JP7755251B2 - Computer program, cognitive function estimation method, learning model generation method, and information processing device - Google Patents

Description

The present invention relates to a computer program for estimating a subject's cognitive function, a cognitive function estimation method, a learning model generation method, and an information processing device.

In recent years, society has been aging, and the number of elderly people living alone is also increasing. There is a risk that cognitive function will decline as people get older, but it is difficult for the average person to determine the level of cognitive function, and in the case of elderly people living alone, it is particularly difficult for a decline in cognitive function to be noticed. Conventionally, cognitive function has been assessed by, for example, doctors or specialists conducting interviews with subjects, but this requires the cooperation of the subjects and only allows assessment of cognitive function at the time of the interview.

Patent Document 1 proposes a dementia information output system that identifies a user's sleeping hours for each day based on the results of measuring the user's body movements, determines whether electrical devices used by the user are turned on, calculates the frequency of days on which the electrical devices are determined to be turned on during the identified sleeping hours as the frequency of forgetting to turn them off, and determines the possibility that the user has developed mild dementia or the like based on the calculated frequency of forgetting to turn them off.

International Publication No. 2017/191697

The dementia information output system described in Patent Document 1 requires that equipment for measuring body movements be installed in the subject's room, etc., making it difficult to implement due to issues such as cost and installation work. Furthermore, while it is possible to measure body movements when the subject's range of movement is limited to a small space such as a single room, such as in a nursing home, in a home with multiple rooms, for example, it is not possible to measure the subject's body movements in rooms other than the one where the measuring device is installed, and there are concerns that installing measuring devices in every room would result in further increases in costs.

The present invention was made in light of these circumstances, and its purpose is to provide a computer program, a cognitive function estimation method, a learning model generation method, and an information processing device that are expected to facilitate the introduction of systems for diagnosing dementia.

In one embodiment, a computer program causes a computer to acquire estimated usage information for each electrical device in a facility, input the acquired usage information into a learning model that has undergone machine learning to output cognitive function information related to the cognitive functions of subjects using the facility when the usage information is input, acquire the cognitive function information output by the learning model, and output the acquired cognitive function information.

In one embodiment, it is expected that systems for diagnosing dementia will be easier to implement.

1 is a schematic diagram for explaining an overview of an information processing system according to a first embodiment; 10 is a graph showing an example of the results of a survey on the use of electrical appliances and the state of cognitive function. 10 is a graph showing an example of the results of a survey on the use of electrical appliances and the state of cognitive function. 10 is a graph showing an example of the results of a survey on the use of electrical appliances and the state of cognitive function. FIG. 2 is a block diagram showing a configuration of a server device according to the present embodiment. FIG. 2 is a schematic diagram for explaining the configuration of a learning model according to the first embodiment. FIG. 2 is a schematic diagram illustrating an example of a subject information DB. FIG. 2 is a schematic diagram illustrating an example of a power information DB. FIG. 2 is a block diagram showing the configuration of a terminal device according to the present embodiment. 10 is a flowchart showing the procedure of a learning model generation process performed by the server device according to the present embodiment. 10 is a flowchart showing the procedure of a cognitive function estimation process performed by the server device according to the present embodiment. FIG. 10 is a schematic diagram illustrating an example of a notification screen displayed by the terminal device. FIG. 10 is a schematic diagram for explaining the configuration of a learning model according to the second embodiment. FIG. 10 is a schematic diagram for explaining the configuration of a learning model according to the third embodiment.

Specific examples of information processing systems according to embodiments of the present invention will be described below with reference to the drawings. Note that the present invention is not limited to these examples, but is defined by the claims, and is intended to include all modifications within the meaning and scope of the claims.

<System Overview>
FIG. 1 is a schematic diagram for explaining an overview of an information processing system according to a first embodiment. The information processing system according to this embodiment is a system that estimates the cognitive function of a person living in a facility 101, for example, a house or a room in a nursing home, by using power information including values of current, voltage, power, or the like measured by a sensor 1 installed in the facility 101. The sensor 1 is installed in a distribution board or the like of the facility 101, and measures the power consumed within the facility 101, for example, at a frequency of once to several times per second. The sensor 1 has a wireless communication function using, for example, a wireless LAN (Local Area Network) or a mobile phone communication network. The sensor 1 transmits measurement results, such as power consumption measured for the facility 101, to a server device 3 as power information.

The server device 3 receives power information transmitted from the sensor 1 of the facility 101 and stores and accumulates the received power information in a database or the like. Based on the power information acquired from the sensor 1, the server device 3 performs processing to estimate the individual power consumption and usage time, etc., of multiple electrical devices installed within the facility 101, such as various electrical devices such as televisions (television devices), air conditioners (air conditioners), microwave ovens, washing machines, and lighting fixtures. The power consumption, etc. measured by the sensor 1 is the overall power consumption, etc. of the facility 101, and is the total value of the power consumption, etc. of multiple electrical devices. Based on the information on the overall power consumption, etc. of the facility 101 acquired from the sensor 1, the server device 3 calculates the individual power consumption, etc. of each electrical device installed within the facility 101, for example, by performing pattern analysis of temporal changes in current consumption or frequency component analysis. The server device 3 also calculates the time when each electrical device started operating, the time when it stopped operating, and the operating time, etc. The process performed by the server device 3 to estimate the power consumption and operating time of individual electrical devices uses a technology called NILM (Non-Instrusive Load Monitoring). Because NILM is an existing technology, a detailed description of it will be omitted in this embodiment.

The server device 3 according to this embodiment performs processing to estimate the cognitive function of a subject residing in the facility 101, using information such as the power consumption and usage time of individual electrical devices estimated using NILM technology from power information acquired from the sensor 1 in the facility 101. The inventors of the present application have discovered that it is possible to estimate to a certain extent the behavior of people residing in the facility 101 based on the power consumption or usage time of electrical devices within the facility 101, and that it is possible to estimate the degree of risk of dementia, etc., based on human behavior. The inventors of the present application collected information on the power consumption or usage time of electrical devices installed in the facility 101 and the level of cognitive function of people residing in the facility 101, and used this collected information to generate a learning model (so-called AI (Artificial Intelligence)) using machine learning. The server device 3 according to this embodiment uses the generated learning model to estimate the cognitive function of people residing in the facility 101 based on information such as the power consumption or usage time of electrical devices.

The server device 3 estimates the cognitive functions of people living in the facility 101 using the machine-learned learning model and transmits the estimation results to the terminal device 5 of the user 102. The user 102 may be, for example, a person living in the facility 101, i.e., the person whose cognitive function is to be estimated, or may be, for example, a spouse, child, parent, grandchild, doctor, or caregiver of the person whose cognitive function is to be estimated, i.e., a person other than the person whose cognitive function is to be estimated. The terminal device 5 may be, for example, a general-purpose information processing device such as a smartphone, tablet terminal device, or personal computer. Upon receiving the estimation results of cognitive function from the server device 3, the terminal device 5 notifies the user 102 of the receipt of the estimation results and displays the estimation results of the subject's cognitive function.

Note that the information processing system according to this embodiment is configured such that a single server device 3 performs the process of generating a learning model and the process of estimating cognitive function using the generated learning model, but this is not limited to this. For example, a server device that performs the process of generating a learning model and a server device that performs the process of estimation using the learning model may be provided separately. Furthermore, the estimation process using the learning model may be performed by the terminal device 5. The process of generating a learning model may also be performed by the terminal device 5.

The process of generating a learning model or the process of estimating cognitive function using the generated learning model may be performed by, for example, a personal computer, smartphone, tablet device, game console, wearable device, or various other information processing devices. Furthermore, an information processing device that performs the process of generating a learning model or the process of estimating cognitive function using the generated learning model may be configured not to communicate with other devices, i.e., a standalone configuration. When performing these processes on a standalone information processing device, the information processing device may obtain power information for facility 101 via, for example, a recording medium, or the user may input information printed on paper or the like into the information processing device.

<Relationship between electrical device usage and cognitive function>
The inventors of the present application conducted a survey on the use of electrical devices and the state of cognitive function and found that there is a relationship between the two. Figures 2 to 4 are graphs showing examples of the survey results on the use of electrical devices and the state of cognitive function.

The graph shown shows the results of a survey conducted over several years from 2019 on the relationship between the amount of time electrical devices were used and the state of the subjects' cognitive function, targeting approximately 80 healthy individuals living in Nobeoka City, Miyazaki Prefecture. Information obtained from sensors 1 installed in the facility 101 where the survey subjects live was collected and stored, including information on electrical quantities such as the amount of power or current consumed throughout the facility 101, as well as information on changes in these electrical quantities over time or frequency components. This accumulated information was then processed, including missing value completion and format conversion, and analyzed using NILM technology to estimate the usage time for each electrical device within the facility 101.

The cognitive function status of the subjects was assessed by the National Cerebral and Cardiovascular Center based on the MMSE-J (Mini-Mental State Examination-Japanese) and GDS-15 (Geriatric Depression Scale 15). For reference, cognitive function testing was also conducted using the TICS-J (Telephone Interview for Cognitive Status in Japanese), a telephone cognitive function screening, and the test results were obtained. Detailed explanations of the cognitive function assessment methods based on the MMSE-J, GDS-15, and TICS-J are omitted here. In this study, on a maximum score of 30 points using the MMSE-J, a score of 23 or less was considered cognitively impaired, 24 to 27 points was considered mild cognitive impairment, and 28 points or higher was considered normal. Furthermore, when assessing depression using the GDS-15, a score of 5 or higher was considered depressed.

Figure 2 shows a graph of the relationship between the usage time of an induction heating (IH) cooker as an electrical appliance and the dementia assessment results of the subject. The usage time of the electrical appliance was calculated as the average daily usage time for each season (spring, summer, autumn, and winter) for each subject. The graph shows the average usage time of the electrical appliance calculated for each dementia assessment result, along with high and low lines indicating the confidence interval for this average. Based on the graph in Figure 2, it can be seen that the usage time of an induction cooker tends to decrease as cognitive function declines.

Figure 3 shows a graph of the relationship between the amount of time spent using a microwave oven (an electrical appliance) and the assessment results regarding the subject's state of depression. Figure 4 shows a graph of the relationship between the amount of time spent using a television (an electrical appliance) and the assessment results regarding the subject's state of depression. Based on the graphs in Figures 3 and 4, it can be seen that people with depressive symptoms tend to spend less time using microwave ovens and televisions. For example, with regard to microwave use, it is estimated that people with depressive symptoms have a reduced appetite, activity level, etc., and therefore spend less time using the microwave.

In this embodiment, the results of a survey on the relationship between the usage time of three electrical appliances, an induction cooker, a microwave oven, and a television, and the state of cognitive function of the subject were presented. However, a relationship between usage time and the state of cognitive function was also observed for other electrical appliances, such as air conditioners, lighting fixtures, audio equipment, refrigerators, and rice cookers. From these results, an estimated usage time for each electrical appliance in the facility 101 can be obtained, and the cognitive function of the subject can be estimated based on the obtained usage time for each electrical appliance. Furthermore, in this embodiment, the cognitive function of the subject can be estimated based on the estimated usage time for individual or multiple electrical appliances in the facility 101, rather than the overall power usage or the power usage of individual appliances in the facility 101.

Furthermore, the characteristics of cognitive function and the amount of time spent using electrical devices each differ depending on the subject's attributes, such as age, gender, and educational background. Therefore, by acquiring attribute information such as the subject's age, gender, and educational background and using it as input information for estimating cognitive function, it is expected that the accuracy of the estimation will be improved. For example, when setting criteria for determining the state of cognitive function based on the amount of time spent using electrical devices, it is possible to set criteria that correspond to the subject's attribute information.

<Device configuration>
5 is a block diagram showing the configuration of the server device 3 according to this embodiment. The server device 3 according to this embodiment is configured to include a processing unit 31, a memory unit (storage) 32, and a communication unit (transceiver) 33. Note that although this embodiment will be described assuming that processing is performed by one server device, processing may also be performed in a distributed manner by a plurality of server devices.

The processing unit 31 is configured using an arithmetic processing device such as a CPU (Central Processing Unit), MPU (Micro-Processing Unit) or GPU (Graphics Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), etc. By reading and executing the server program 32a stored in the memory unit 32, the processing unit 31 performs various processes, such as collecting power information from sensors 1 installed in the facility 101, estimating the power consumption and usage time of each electrical device installed in the facility 101 based on the collected power information, estimating the cognitive functions of people living in the facility 101 based on the estimated usage time of the electrical devices, and generating a learning model used to estimate cognitive functions.

The memory unit 32 is configured using a large-capacity storage device such as a hard disk. The memory unit 32 stores various programs executed by the processing unit 31 and various data required for the processing of the processing unit 31. In this embodiment, the memory unit 32 stores the server program 32a executed by the processing unit 31, and is also provided with a training data memory unit 32b that stores training data used in the process of generating a learning model, a learning model memory unit 32c that stores information about unlearned or learned learning models, a subject information DB (database) 32d that stores information about subjects whose cognitive function is to be estimated, and a power information DB 32e that stores power information collected from sensors 1 in the facility 101.

In this embodiment, the server program (program product) 32a is provided in a form recorded on a recording medium 99 such as a memory card or optical disc, and the server device 3 reads the server program 32a from the recording medium 99 and stores it in the memory unit 32. However, the server program 32a may also be written to the memory unit 32, for example, during the manufacturing stage of the server device 3. Alternatively, the server program 32a may be distributed by another remote server device or the like and acquired by the server device 3 via communication. For example, the server program 32a may be read from the recording medium 99 by a writing device and written to the memory unit 32 of the server device 3. The server program 32a may be provided in a form distributed via a network, or in a form recorded on the recording medium 99.

The training data storage unit 32b stores multiple pieces of training data used in the generation (learning) process of the learning model. The training data is, for example, data that associates input information and output information for the learning model. In this embodiment, data that associates information such as the usage time of individual electrical devices and the attributes of the subject whose cognitive function is to be estimated with labels that indicate the cognitive function of the subject is used as the training data. The training data is created in advance, for example, by a designer of the information processing system related to this embodiment, and is stored in the training data storage unit 32b of the storage unit 32 of the server device 3.

The learning model storage unit 32c stores information about a learning model that estimates the cognitive function of a subject. The information stored in the learning model storage unit 32c is, for example, information about the structure of the learning model and information such as parameters determined by machine learning. In addition to information about learning models for which machine learning has been completed, the learning model storage unit 32c can also store learning models in their initial state before machine learning is performed, or learning models temporarily saved as an intermediate step in machine learning.

Figure 6 is a schematic diagram illustrating the configuration of a learning model according to embodiment 1. The learning model 7 according to this embodiment is a learning model that has undergone machine learning to receive as input the usage time of individual electrical devices estimated by the server device 3 based on power information measured by the sensor 1 in the facility 101 and attribute information of the subject whose cognitive function is to be estimated, and to output the state of the subject's cognitive function. The learning model 7 may employ models of various configurations, such as a neural network, a deep neural network, an SVM (Support Vector Machine), a decision tree, a random forest, or a logistic regression. Furthermore, the machine learning may include deep learning, ensemble learning, etc.

The usage time of electrical devices input into the learning model 7 may be statistical values such as the daily average and standard deviation of usage time calculated for each electrical device based on, for example, power information collected from the sensor 1 of the facility 101 over several months (e.g., one month or three months). For example, the learning model 7 may be input with numerical values for the average usage time of each electrical device, such as the average usage time of a television, the average usage time of an air conditioner, the average usage time of a microwave oven, etc.

In this embodiment, statistical values such as the average and standard deviation of usage time are used as input data for learning model 7, but this is not limited to this. For example, time-series data on usage time of electrical devices may be input to learning model 7. In this case, learning model 7 may employ a configuration such as an RNN (Recurrent Neural Network), LSTM (Long Short Term Memory), Seq2Seq (sequence to sequence), or Transformer.

The subject's attribute information input into the learning model 7 may include, for example, the subject's age, gender, educational background, etc. Educational background may be, for example, the highest level of education (junior high school graduation, high school graduation, university graduation, etc.).

The cognitive function state output by learning model 7 may include, for example, a normal state in which cognitive function is not impaired, a mild cognitive impairment state in which cognitive function is slightly impaired, and a cognitive impairment state in which cognitive function is further impaired. Learning model 7 is a learning model that classifies the subject into one of these three states, and outputs three values corresponding to the normal state, mild cognitive impairment, and cognitive impairment. The output value of learning model 7 is a numerical value indicating the probability (likelihood, confidence, etc.) of the corresponding state, and the state corresponding to the maximum value of the three output values can be estimated as the cognitive function state of the subject.

In this embodiment, the learning model 7 is configured to output three values: a normal state, a mild cognitive impairment state, and a cognitive impairment state; however, this is not limited to this. The learning model 7 may be configured to output two values, for example, a normal state and a state of cognitive decline including mild cognitive impairment and cognitive impairment. The learning model 7 may also be configured to output a single value, for example, outputting whether or not the state is cognitive decline. The learning model 7 may also be configured to output four or more values. Furthermore, the learning model 7 may be configured to output a value such as a predicted probability value of a state of cognitive decline, or a predicted cognitive function score.

In addition, in this embodiment, the server device 3 stores multiple learning models 7 in the learning model storage unit 32c according to household composition, such as a learning model 7 for single people and a learning model 7 for married couple households. These multiple learning models 7 have the same composition, input information, output information, etc., but different training data used during machine learning and different internal parameters. The server device 3 selects one of the learning models 7 according to the subject's household composition, and inputs the electrical device usage time and the subject's attribute information into the selected learning model 7 to obtain the state of cognitive function output by the learning model 7.

Note that, while the learning model 7 in this embodiment is configured to accept the usage time of the electrical device and the subject's attribute information as input information, this is not limiting. For example, the learning model 7 may be configured to accept the usage time of the electrical device as input (without accepting the subject's attribute information as input) and output the state of the subject's cognitive function based on the input usage time. The learning model 7 may also be configured to accept input of various information other than the usage time of the electrical device and the subject's attribute information, and output the state of the subject's cognitive function. The learning model 7 may also be configured to input, for example, the amount of power consumption or current consumption of the electrical device instead of the usage time of the electrical device, or the number of times the electrical device is used estimated from the amount of power consumption or current consumption. The usage status information of the electrical device may include various information such as measured electrical values such as the amount of power consumption or current consumption of the electrical device, or estimated values such as the usage time or number of times the electrical device is used based on these measured values. This information may be appropriately combined and input to the learning model 7.

Figure 7 is a schematic diagram showing an example of the subject information DB 32d. The subject information DB 32d of the server device 3 according to this embodiment stores information such as "sensor ID," "subject ID," "age," "gender," "educational history," and "household composition" in association with each other. The "sensor ID" is identification information uniquely assigned to each sensor 1 installed in a distribution board or the like of the facility 101. The "subject ID" is identification information uniquely assigned to the subject whose cognitive function state is to be estimated, and may be the subject's name, etc. The "age" is numerical information indicating the subject's age. The "gender" is the gender of the subject, and is set to "male" or "female." The "educational history" is the subject's highest level of education, and information such as "high school," "university," or "graduate school" is set. The "household composition" is information indicating the number of people living in the facility 101 where the sensor 1 is installed, and information such as "single person (living alone)" or "couple (living together)" is set. The information stored in the subject information DB 32d is entered by the subject or the subject's agent, for example, when accepting an application to use a service provided by the information processing system according to this embodiment.

Figure 8 is a schematic diagram showing an example of the power information DB 32e. The power information DB 32e of the server device 3 according to this embodiment stores, for example, a "sensor ID" and "date" in association with the "power consumption" and "usage time" of multiple electrical devices, such as a "television," "air conditioner," and "microwave oven." The "sensor ID" is identification information uniquely assigned to each sensor 1 installed in a distribution board or the like in the facility 101, and is the same as the sensor ID stored in the subject information DB 32d described above. The "date" is the date on which the sensor 1 performed measurement, and is set to information such as "6/4/2021." The "power consumption" of each electrical device is the total daily power consumption of each electrical device estimated by the server device 3 based on the power information from the sensor 1. The "usage time" of each electrical device is the total daily usage time of each electrical device estimated by the server device 3 based on the power information from the sensor 1.

The communication unit 33 of the server device 3 communicates with various devices via a network N, which may include a mobile phone communication network, a wireless local area network (LAN), the Internet, etc. In this embodiment, the communication unit 33 communicates with the sensor 1 and the terminal device 5 via the network N. The communication unit 33 transmits data provided by the processing unit 31 to other devices, and provides data received from other devices to the processing unit 31.

The storage unit 32 may be an external storage device connected to the server device 3. The server device 3 may also be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed using software. The server device 3 is not limited to the above configuration and may also include, for example, a reading unit that reads information stored on a portable storage medium, an input unit that accepts operational input, or a display unit that displays images.

In addition, in the server device 3 according to this embodiment, the processing unit 31 reads and executes the server program 32a stored in the memory unit 32, thereby realizing a learning model generation unit 31a, a usage time estimation unit 31b, a cognitive function estimation unit 31c, and the like as software functional units in the processing unit 31. Note that in this diagram, functional units of the processing unit 31 related to the generation process of the learning model 7 and the estimation process of the subject's cognitive function are shown, and functional units related to other processes are not shown.

The learning model generation unit 31a performs processing to generate a learning model 7 that estimates cognitive function. As shown in FIG. 6, the learning model 7 according to this embodiment is a learning model that receives as input the individual usage times of electrical devices estimated by the server device 3 and attribute information of the person being estimated, and outputs information indicating the state of the cognitive function of the person. The learning model generation unit 31a acquires pre-created training data from the training data storage unit 32b, and performs processing to machine-learn a learning model 7 of the desired configuration using the acquired training data, thereby generating a learning model 7 that estimates the cognitive function of the person being estimated. Since the supervised learning processing of the learning model is an existing technology, detailed explanation will be omitted; however, the learning model generation unit 31d can train the learning model 7 using techniques such as gradient descent, stochastic gradient descent, or backpropagation.

The usage time estimation unit 31b performs a process to estimate the usage time of each electrical device installed in the facility 101 based on the power information obtained from the sensor 1 of the facility 101. The power information repeatedly transmitted by the sensor 1 includes measurement results that combine the power consumption or current consumption, etc., of multiple electrical devices installed in the facility 101. The usage time estimation unit 31b repeatedly receives the power information from the sensor 1 and stores and accumulates the received power information in the power information DB 32e. The usage time estimation unit 31b reads the power information accumulated in the power information DB 32e and estimates the individual power consumption or current consumption, etc., of each electrical device installed in the facility 101 based on the temporal increase/decrease pattern or frequency domain distribution of the combined power consumption or power consumption, etc., for the facility 101. The usage time estimation unit 31b also estimates the time each electrical device has been operating (used) based on the temporal changes in the individual power consumption or current consumption, etc., of each electrical device. The estimation process performed by the usage time estimation unit 31b is performed using NILM technology. Since NILM is an existing technology, detailed explanation will be omitted in this embodiment.

The cognitive function estimation unit 31c performs processing to estimate the cognitive function of the subject using the trained learning model 7 stored in the learning model storage unit 32c. The cognitive function estimation unit 31c acquires the usage time of each electrical device in the facility 101 estimated by the usage time estimation unit 31b, and acquires attribute information of the subject living in this facility 101 from the subject information DB 32d. The cognitive function estimation unit 31c inputs the acquired usage time of each electrical device and the attribute information of the subject living in the facility 101 into the learning model 7, and acquires information on the state of cognitive function output by the learning model 7. Based on the information acquired from the learning model 7, the cognitive function estimation unit 31c estimates whether the subject is in, for example, a normal state, a state of mild cognitive impairment, or a state of cognitive impairment. If the learning model 7 is configured to output likelihoods, etc. for these three states, the cognitive function estimation unit 31c can estimate that the state with the highest likelihood is the state of the subject's cognitive function. The cognitive function estimation unit 31c transmits the estimation result of the cognitive function to a terminal device 5 of the subject or the subject's family member, etc., and notifies them of the estimation result.

Figure 9 is a block diagram showing the configuration of a terminal device 5 according to this embodiment. The terminal device 5 according to this embodiment is configured to include a processing unit 51, a memory unit (storage) 52, a communication unit (transceiver) 53, a display unit (display) 54, and an operation unit 55. The terminal device 5 is a device used by the subject or the subject's family, etc., whose cognitive function is to be estimated by the information processing system according to this embodiment, and may be configured using an information processing device such as a smartphone, tablet terminal device, or personal computer.

The processing unit 51 is configured using a processing unit such as a CPU or MPU, ROM, RAM, etc. The processing unit 51 reads and executes the program 52a stored in the storage unit 52 to perform various processes, such as receiving information transmitted from the server device 3 and displaying the received information to notify the user 102.

The storage unit 52 is configured using a non-volatile memory element such as a flash memory. The storage unit 52 stores various programs executed by the processing unit 51 and various data required for processing by the processing unit 51. In this embodiment, the storage unit 52 stores the program 52a executed by the processing unit 51. In this embodiment, the program (program product) 52a is distributed by a remote server device or the like, and the terminal device 5 acquires it via communication and stores it in the storage unit 52. However, the program 52a may also be written to the storage unit 52, for example, during the manufacturing stage of the terminal device 5. For example, the program 52a may be read by the terminal device 5 from a recording medium 98 such as a memory card or optical disc and stored in the storage unit 52. For example, the program 52a may be read by a writing device from the recording medium 98 and written to the storage unit 52 of the terminal device 5. The program 52a may be provided by being distributed via a network or by being recorded on the recording medium 98.

The communication unit 53 communicates with various devices via a network N, which may include a mobile phone communication network, a wireless LAN, the Internet, etc. In this embodiment, the communication unit 53 communicates with the server device 3, etc., via the network N. The communication unit 53 transmits data provided by the processing unit 51 to other devices, and provides data received from other devices to the processing unit 51.

The display unit 54 is configured using a liquid crystal display or the like, and displays various images, characters, etc. based on processing by the processing unit 51. The operation unit 55 accepts operations from the user 102 and notifies the processing unit 51 of the accepted operations. For example, the operation unit 55 accepts operations from the user 102 using an input device such as a mechanical button or a touch panel provided on the surface of the display unit 54. For example, the operation unit 55 may also be an input device such as a mouse and keyboard, and these input devices may be configured to be detachable from the terminal device 5.

In addition, in the terminal device 5 according to this embodiment, the processing unit 51 reads and executes the program 52a stored in the storage unit 52, thereby realizing the display processing unit 51a and the like as software functional units in the processing unit 51. Note that the program 52a may be a program dedicated to the information processing system according to this embodiment, or may be a general-purpose program such as an internet browser or web browser.

The display processing unit 51a receives information regarding the estimated results of the subject's cognitive function transmitted from the server device 3, and processes the received information to display the received information on the display unit 54. In the information processing system according to this embodiment, the server device 3 also transmits information regarding the usage time and power consumption of each electrical device installed in the facility 101 to the terminal device 5. The display processing unit 51a receives information regarding the usage time and power consumption of the electrical devices transmitted from the server device 3, and displays the received information on the display unit 54. The display processing unit 51a of the terminal device 5 according to this embodiment displays information regarding the usage time or power consumption of each electrical device estimated by the server device 3, and information regarding the subject's cognitive function estimated by the server device 3, on the display unit 54.

<Learning model generation process>
In the information processing system according to this embodiment, a process for generating a learning model 7 for estimating the cognitive function of a subject is performed in advance by the server device 3. The process performed by the server device 3 according to this embodiment to generate the learning model 7 is a so-called supervised learning machine learning process. The server device 3 performs the supervised learning process for the learning model 7 using a plurality of training data in which input information and output information of the learning model 7 are associated with each other.

The information processing system according to this embodiment collects and stores power information obtained from multiple sensors 1 installed in multiple facilities 101, as well as information related to the state of cognitive function of people living in the facilities 101. The power information obtainable from the sensors 1 is the measured values of power consumption or current consumption of all electrical devices installed in the facility 101. In the information processing system according to this embodiment, the power consumption or current consumption of each electrical device installed in the facility 101 is estimated using NILM technology from the power information obtained from the sensors 1, and the usage time of each electrical device per day, for example, is estimated from this estimation result.

In the information processing system according to this embodiment, the daily usage time of each electrical device is estimated using NILM technology based on power information collected over several months to several years, and statistical values such as the average and standard deviation of the usage time of each electrical device are calculated for each season, for example, spring (March to May), summer (June to August), fall (September to November), and winter (December to February). In this embodiment, the average and standard deviation of the usage time of each electrical device for each season for, for example, one year are used as input information for the training data of the learning model 7.

In this embodiment, statistical values such as the average and standard deviation of the usage time of electrical appliances by season are used as input information for the training data of the learning model 7, but this is not limited to this. For example, statistical values of the usage time of electrical appliances per month or per day may be used as input information, or time series data such as the usage time or power consumption of electrical appliances may be used, or other information may be used.

Furthermore, in the information processing system according to this embodiment, the training data used for machine learning in the learning model 7 includes, as input information, the average and standard deviation values of the usage time of each electrical device for each season as well as attribute information of the people living in the facility 101. The attribute information may be, for example, age, gender, and educational background.

In addition, in this embodiment, diagnostic results obtained by doctors or other medical professionals assessing the cognitive function status of people living in facility 101, for example, once to several times a year, are collected. In this embodiment, the diagnostic results are information indicating whether the subject is in a normal state, a state of mild cognitive impairment, or a state of cognitive impairment. These diagnostic results become the output information of the training data for learning model 7, i.e., information on the correct label for the input information.

As such, the training data used for machine learning of the learning model 7 in the information processing system according to this embodiment is data that associates the average and standard deviation values of the annual usage time of each electrical device installed in the facility 101 by season with attribute information of the people living in the facility 101 and is labeled with a label indicating the state of cognitive function of the people living in the facility 101. It is preferable that training data related to as many different facilities 101 and people as possible be collected. The collection of information necessary for this training data and the creation of training data based on the collected information may be performed, for example, by the server device 3, by a device other than the server device 3, or manually by a system designer, etc. The created training data is stored in advance in the training data storage unit 32b of the server device 3. Arithmetic processing, such as estimating the usage time of each electrical device from the power information obtained from the sensor 1 and calculating the average and standard deviation values of the usage time of each electrical device by season, may be performed in advance, for example, before the machine learning of the learning model 7 is performed. Alternatively, for example, unprocessed power information may be stored and performed as preprocessing when the machine learning processing is performed.

For example, suppose facility 101 is equipped with three types of electrical appliances: a television, an air conditioner, and a microwave. In this case, the training data would be the average and standard deviation of television usage time in the spring, the average and standard deviation of television usage time in the summer, the average and standard deviation of television usage time in the fall, the average and standard deviation of television usage time in the winter, the average and standard deviation of air conditioner usage time in the spring, ..., the average and standard deviation of air conditioner usage time in the winter, the average and standard deviation of microwave usage time in the spring, ..., the average and standard deviation of microwave usage time in the winter, and attribute information such as the age, gender, and education history of the people living in facility 101, with corresponding correct labels indicating whether the people are in a normal state, a state with mild cognitive impairment, or a state with cognitive impairment. The amount of information (features, dimensions) of the training data may be reduced using techniques such as RFE (Recursive Feature Elimination).

Furthermore, in the information processing system of this embodiment, single people living alone in facility 101 and people in married couple households living with two people are treated as subjects for estimating the state of cognitive function, and households with more than one person are excluded from the estimation of the state of cognitive function. For this reason, the information processing system of this embodiment generates two learning models 7: a learning model 7 that estimates the state of cognitive function of single subjects, and a learning model 7 that estimates the state of cognitive function of subjects in married couple households. The learning model 7 that estimates the state of cognitive function of single subjects is trained by machine learning using training data created based on information collected at the facility 101 for single people. The learning model 7 that estimates the state of cognitive function of subjects in married couple households is trained by machine learning using training data created based on information collected at the facility 101 for married couple households. In this embodiment, only single-person or married couple households are targeted, but this is not limited to this and other households (for example, two generations of a married couple and their children, two generations of a married couple and their parents, or three generations of a married couple, their children, and their parents) may also be targeted. In this case, training data is created based on information collected from the target households, and machine learning is performed on the learning model 7.

The server device 3 generates a learning model 7 by performing machine learning using the above-mentioned training data created in advance. Figure 10 is a flowchart showing the steps of the learning model generation process performed by the server device 3 according to this embodiment. The learning model generation unit 31a of the processing unit 31 of the server device 3 according to this embodiment reads out training data stored in the training data storage unit 32b of the storage unit 32 (step S1). The learning model generation unit 31a performs preprocessing such as data shaping on the read out data as necessary (step S2). In step S2, for example, calculations may be performed to aggregate information on the daily usage time of electrical appliances by season.

The learning model generation unit 31a acquires one piece of data from among the multiple pieces of training data (step S3). The learning model generation unit 31a performs calculations using the acquired data and updates the parameters of the learning model (step S4). Note that the learning model generation unit 31a may perform steps S3 and S4 in parallel using multiple pieces of data. The learning model generation unit 31a determines whether the processing of step S4 has been completed for all of the training data (step S5). If the processing has not been completed for all of the training data (S5: NO), the learning model generation unit 31a returns to step S3, acquires other data, and repeats the processing.

If processing has been completed for all training data (S5: YES), the learning model generation unit 31a determines whether the parameter update process for the learning model using all training data has been performed a predetermined number of times and completed (step S6). The predetermined number of times used for the determination in step S6 is determined in advance by the designer of the system, etc. If the predetermined number of iterations has not been completed (S6: NO), the learning model generation unit 31a returns to step S3 and repeatedly performs the parameter update process for the learning model 7 using the same training data. If the predetermined number of iterations has been completed (S6: YES), the learning model generation unit 31a stores information such as the parameters of the finally determined learning model 7 in the learning model memory unit 32c of the memory unit 32 (step S7) and terminates processing.

<Cognitive function estimation processing>
The server device 3 uses the learning model 7 generated by the above-described generation process to perform a process of estimating the state of cognitive function of a subject living in the facility 101. The server device 3 receives power information periodically transmitted from the sensor 1 installed in the facility 101, and stores and accumulates the received power information in the power information DB 32e. The server device 3 performs a process of estimating the state of cognitive function of the subject, for example, every time a predetermined period elapses (every month or every season, etc.), or when a request is received from the terminal device 5, for example.

In the cognitive function estimation process, the server device 3 first, as preprocessing, reads out power information for a predetermined period (e.g., the past year) stored in the power information DB 32e and calculates the average value and standard deviation of usage time by season. The server device 3 also reads out the subject's attribute information (age, gender, educational background, household composition, etc.) stored in the subject information DB 32d and checks whether the subject's household composition is single or married. The server device 3 reads out a learning model 7 corresponding to the household composition from the learning model storage unit 32c of the storage unit 32. The server device 3 inputs into the learning model 7 the average value and standard deviation of usage time by season for each electrical device installed in the facility 101 where the subject lives, as well as the subject's attribute information such as age, gender, and educational background.

The learning model 7 accepts input of the average and standard deviation values of the usage time for each electrical device by season, as well as attribute information such as the subject's age, gender, and educational background, and outputs an estimation result of the subject's cognitive function state. In this embodiment, the learning model 7 outputs likelihood (probability, confidence) values for each of the normal state, mild cognitive impairment state, and cognitive impairment state as an estimation result of the cognitive function state. The server device 3 obtains the three likelihood values output by the learning model 7, and the cognitive function state corresponding to the largest value is set as the estimation result of the subject's cognitive function state.

In this embodiment, the learning model 7 is configured to accept as input statistical values such as the average value and standard deviation of the usage time of the electrical appliances for each season, but this is not limited to this. The learning model 7 may also be configured to accept as input statistical values of the usage time of the electrical appliances for each month or each day, or may be configured to accept as input time-series data such as the usage time or power consumption of the electrical appliances, or may be configured to accept information other than these as input.

Figure 11 is a flowchart showing the steps of the cognitive function estimation process performed by the server device 3 according to this embodiment. The cognitive function estimation unit 31c of the server device 3 according to this embodiment determines whether the timing has arrived to perform the cognitive function estimation process of the subject, for example, when a predetermined period of time has elapsed or when a request to perform the estimation process is received from a terminal device (step S11). If the timing has not arrived to perform the estimation process (S11: NO), the cognitive function estimation unit 31c waits until the timing to perform the estimation process arrives.

When the time has come to perform the estimation process (S11: YES), the cognitive function estimation unit 31c reads out the power information stored in the power information DB 32e of the memory unit 32 (step S12). The cognitive function estimation unit 31c also reads out attribute information related to the subject of cognitive function estimation stored in the subject information DB 32d of the memory unit 32 (step S13). The cognitive function estimation unit 31c performs appropriate preprocessing on the information read out in steps S12 and S13 as necessary (step S14) to generate information to be input to the learning model 7.

The cognitive function estimation unit 31c inputs the input information, which has been appropriately preprocessed in step S14, into the trained learning model 7 stored in the learning model memory unit 32c of the memory unit 32 (step S15). At this time, the cognitive function estimation unit 31c determines whether the subject is a single person or a married couple based on the attribute information read out in step S13, and selects a learning model 7 according to the subject's attributes to input the information. The cognitive function estimation unit 31c acquires an output value output by the learning model 7 in response to the input information. The cognitive function estimation unit 31c estimates whether the subject's cognitive function state is normal, mild cognitive impairment, or cognitive impairment by determining which of the multiple output values acquired from the learning model 7 has the maximum value (step S17). The cognitive function estimation unit 31c transmits information including the estimation result of step S17 to the terminal device 5 (step S18), displays the estimation result on the terminal device 5, and ends the processing.

The server device 3 transmits the estimated result of the state of the subject's cognitive function to the terminal device 5 of a user 102 (the subject himself/herself or a family member of the subject, etc.) predetermined for the subject. Note that in the information processing system according to this embodiment, information regarding the usage time of electrical devices provided in the facility 101 and information regarding the estimated result of the cognitive function of the subject living in the facility 101 are provided to the user 102. Therefore, in this embodiment, the server device 3 transmits to the terminal device 5 the usage time of each electrical device estimated based on power information acquired from the sensor 1 in the facility 101 and the estimated result of the cognitive function of the subject living in the facility 101. However, the information processing system may be configured to provide the user 102 with information regarding the estimated result of the subject's cognitive function, but not with information regarding the usage time of the electrical devices.

The terminal device 5 receives the information transmitted from the server device 3 and displays the received information on the display unit 54, thereby notifying the user 102 of the amount of time the electrical appliances were used and the estimated results of the subject's cognitive function. Figure 12 is a schematic diagram showing an example of a notification screen displayed by the terminal device 5. The notification screen displayed by the terminal device 5 displays the user's ID or name, for example, by displaying a character string such as "User AAAA" at the top of the screen. Below the user name, the notification screen in this example has a usage time display area 111 that displays information about the amount of time the electrical appliances were used, and a cognitive function display area 112 that displays the estimated results of the subject's cognitive function.

The terminal device 5 displays a title string, such as "Usage time by electrical appliance from March to May 1, 2021," and a graph of the usage time of each electrical device installed in the facility 101 in the usage time display area 111 of the notification screen. In the illustrated notification screen, the terminal device 5 associates the names of electrical devices, such as televisions, air conditioners, microwaves, lighting fixtures, and washing machines, with horizontally extending bar graphs indicating the usage time. Based on the information received from the server device 3, the terminal device 5 determines the year, month, and date to be displayed in the title string, generates images of bar graphs corresponding to each electrical device, and displays the illustrated notification screen on the display unit 54.

The terminal device 5 displays a title string such as "State of cognitive function estimated from usage of electrical appliances" and a string indicating the estimated result of cognitive function such as "Possible mild cognitive impairment" in the cognitive function display area 112 of the notification screen. The terminal device 5 displays a predetermined string as the title string. The terminal device 5 also determines the string to display as the estimated result of cognitive function based on the information received from the server device 3. For example, if the estimated result of cognitive function is normal, the terminal device can display a string such as "Cognitive function is normal." Also, for example, if the estimated result of cognitive function is a state of cognitive impairment, the terminal device 5 can display a string such as "Possible cognitive impairment."

<Summary>
In the information processing system according to the present embodiment, the server device 3 acquires usage status information, such as estimated usage time, for each electrical device in the facility 101. The server device 3 inputs the acquired usage status information into a learning model 7, which has undergone machine learning to output cognitive function information regarding the cognitive function of a subject using the facility 101 when the usage status information is input, and acquires the cognitive function information output by the learning model 7. The server device 3 transmits the cognitive function information acquired from the learning model 7 to the terminal device 5, thereby causing the terminal device 5 to display the cognitive function information of the subject. This allows the server device 3 to estimate the cognitive function of the subject based on the usage status of the electrical devices in the facility 101. The usage time of each electrical device can be estimated using NILM technology based on power information, such as power consumption or current consumption, which can be acquired by installing a sensor 1 on a distribution board or the like of the facility 101. Because the sensor 1 only needs to be installed on a distribution board or the like of the facility 101, a cognitive function estimation service using the information processing system according to the present embodiment can be easily implemented. Furthermore, in facilities 101 that are already using a service that notifies users of the usage status of each electrical device, it is highly likely that sensor 1 has already been installed, and this sensor 1 can be used to estimate cognitive function, making it even easier to introduce a cognitive function estimation service.

Note that the information processing system according to this embodiment is configured to collect power information by installing sensors 1 on distribution boards or the like in facility 101, but this is not limited to this. For example, if devices such as smart meters that can acquire similar information are installed in facility 101, server device 3 may acquire power information from these devices. Furthermore, in the information processing system according to this embodiment, the average value and standard deviation of the usage time for each electrical device are used as usage information and input to learning model 7, but this is not limited to this. For example, the average value or standard deviation of the amount of power consumed or current consumed by each electrical device may be input to learning model 7 as usage information, or the average value and standard deviation of the number of times each electrical device is used may be input as usage information.

Furthermore, the learning model 7 according to this embodiment accepts input of attribute information such as the age, gender, and educational background of the subject using the facility 101, along with usage information about the electrical devices in the facility 101, and outputs cognitive function information about the subject. Information such as the subject's age, gender, and educational background has a significant impact on cognitive function, so by using this information, the learning model 7 can more accurately estimate the subject's cognitive function. Note that while the learning model 7 according to this embodiment accepts three pieces of information, age, gender, and educational background, as input of the subject's attribute information, this is not limited to this. For example, the learning model 7 may accept one or two pieces of information, age, gender, or educational background, as input of attribute information. Furthermore, for example, the learning model 7 may accept information other than age, gender, and educational background as input of attribute information.

Furthermore, the learning model 7 according to this embodiment outputs, as cognitive function information, information classifying the subject's cognitive function as either a normal state, a mild cognitive impairment state, or a cognitive impairment state. This allows the server device 3 to obtain the classification results output by the learning model 7 and easily determine the state of the subject's cognitive function. While the learning model 7 according to this embodiment classifies the subject into three states, namely, a normal state, a mild cognitive impairment state, and a cognitive impairment state, this is not limited to this. The learning model 7 may classify the subject into two states, for example, a normal state or a mild cognitive impairment state, or may classify the subject into four or more states, for example, including a state in addition to the above three states. The server device 3 may, for example, use the learning model 7 that classifies the subject into two states, a normal state or a cognitive impairment, to estimate whether or not the subject has dementia, and may use the learning model 7 that classifies the subject who is estimated to have a cognitive impairment into two states, namely, mild cognitive impairment or a more severe cognitive impairment, to estimate the degree of cognitive impairment.

In addition, in the information processing system according to this embodiment, the server device 3 transmits the estimated results of the usage time of the electrical devices and the estimated results of the cognitive function state to the terminal device 5, and the terminal device 5 receives the results and displays the usage time of the electrical devices and the cognitive function state of the subject on the display unit 54. This allows the user 102 using the terminal device 5 to easily check the usage time of each electrical device in the facility 101 and the cognitive function state of the subject using the facility 101. As the terminal device 5 displays both the usage time of the electrical devices and the cognitive function state of the subject, the user 102 can infer the correspondence between the estimated results of the cognitive function state and the usage time of the electrical devices.

In addition, in the information processing system according to this embodiment, a learning model 7 is individually generated according to attribute information indicating whether the subject is a single person or a married couple, and the server device 3 selects one of multiple learning models 7 according to whether the subject is a single person or a married couple, and performs processing to estimate the state of the subject's cognitive function using the selected learning model 7. In a facility 101 such as a residence, it is predicted that the usage time of electrical devices will vary greatly depending on whether the subject lives alone, with two or more people, or with three or more people. Therefore, by generating a learning model 7 individually according to whether the subject lives alone or with a married couple, it is expected that the accuracy of estimation using the learning model 7 will be improved. However, it is also possible to configure the learning model 7 so that whether the subject is a single person or a married couple is input as attribute information of the subject. Furthermore, individual learning models 7 can be generated for household configurations other than single people and married couples, such as a three-person family or a four-person family, to estimate the state of cognitive function.

Note that the information processing system according to this embodiment is configured to estimate the state of the subject's cognitive function using a learning model 7 that has undergone pre-machine learning, but this is not limited to this. Instead of using a learning model 7, the information processing system may be configured to estimate the state of the subject's cognitive function based on whether or not the state matches pre-defined rules (for example, television usage time per day is X hours or more, air conditioner usage time is Y hours or more, microwave usage time is Z hours or less, etc.).

<Second Embodiment>
An information processing system according to the second embodiment performs processing to estimate a subject's state of depression. FIG. 13 is a schematic diagram illustrating the configuration of a learning model 207 according to the second embodiment. In the information processing system according to the second embodiment, a learning model 207 that receives, as input, the usage time of an electrical device and attribute information of the subject and outputs the subject's state of depression is generated by the server device 3, and is used in the processing to estimate the subject's state of depression by the server device 3. The information input to the learning model 207 according to the second embodiment may be the same as the information input to the learning model 7 according to the first embodiment. The information output by the learning model 207 according to the second embodiment is, for example, information classifying whether the subject is in a depressed state or not, and can be the likelihood of being in a depressed state and the likelihood of being in a normal state (not in a depressed state).

In the information processing system according to the second embodiment, diagnostic results obtained by doctors or other medical professionals assessing the depression of people living in the facility 101, for example, once to several times a year, are collected. In this embodiment, the diagnostic results are information indicating whether or not the subject is depressed. These diagnostic results are equivalent to the output information of the learning model 207 and are assigned as correct answer labels corresponding to the input information, thereby creating training data for generating the learning model 207 by machine learning. The server device 3 generates the learning model 207 by performing machine learning using previously created training data, and stores the generated learning model 207 in the learning model storage unit 32c.

The server device 3 estimates the usage time of each electrical device based on the power information obtained from the sensor 1 of the facility 101, and inputs this estimation result and attribute information of the subject living in the facility 101 into the learning model 207. The server device 3 obtains the information output by the learning model 207, estimates whether the subject is in a depressed state, and transmits the estimation result to the terminal device 5. The terminal device 5 receives the information from the server device 3 and displays the estimation result of the subject's depressed state on the display unit 54. At this time, the server device 3 may transmit information related to the usage time of the electrical device along with the estimation result of the depressed state to the terminal device 5, and the terminal device 5 may display the information related to the usage time of the electrical device along with the estimation result of the depressed state.

In the information processing system according to the second embodiment configured as described above, the learning model 207 receives input of usage information such as estimated usage time for each electrical device in the facility 101 and attribute information for the subject who uses the facility 101, and outputs an estimation result of the subject's state of depression. This allows the server device 3 to use the learning model 207 to estimate the subject's state of depression based on the usage status of the electrical devices in the facility 101.

The server device 3 may perform processing to estimate the state of cognitive function and the state of depression of the subject using both the learning model 7 according to embodiment 1 that estimates the state of cognitive function and the learning model 207 according to embodiment 2 that estimates the state of depression. Furthermore, a learning model may be generated that outputs both the state of cognitive function and the state of depression of the subject as cognitive function information in response to input of usage information and attribute information. The cognitive function information output by the learning model is not limited to classification results into the three states of normal state, mild cognitive impairment, and cognitive impairment, or classification results as to whether or not the subject is in a state of depression, but may also be various other information, such as numerical information indicating the degree of cognitive function.

Furthermore, the other configurations of the information processing system according to embodiment 2 are the same as those of the information processing system according to embodiment 1, so similar parts are given the same reference numerals and detailed descriptions are omitted.

<Third Embodiment>
An information processing system according to a third embodiment estimates the future state of cognitive function of a subject based on time series changes in the usage time of an electrical device and the subject's attribute information. Fig. 14 is a schematic diagram illustrating the configuration of a learning model 307 according to the third embodiment. The learning model 307 according to the third embodiment sequentially accepts input of n pieces of time series information (n = 1, 2, 3, ...) and estimates the future state of cognitive function of the subject. For example, information at time T1, information at time T2, ..., information at time Tn is input in chronological order to the learning model 307.

The information at each point in time input into the learning model 307 includes the usage time of each electrical device in the facility 101, and attribute information of the subjects living in the facility 101. However, the information at each point in time input into the learning model 307 does not necessarily have to include attribute information of the subjects.

For example, when information from times T1 to Tn is input, the future cognitive function state output by the learning model 307 is the cognitive function state of the subject at time Tn+1 (or the cognitive function state of the subject at times T2 to Tn+1). In addition to the future cognitive function state, the learning model 307 may also output the future usage time of the electrical device and the subject's attribute information.

For example, if information from spring 2020, summer 2020, autumn 2020, and winter 2020 is input to the learning model 307 in that order, the learning model 307 will estimate and output the state of the subject's cognitive function in spring 2021. Alternatively, if information from spring 2020, summer 2020, autumn 2020, and winter 2020 is input to the learning model 307 in that order, the learning model 307 may output an estimation result for summer 2020, an estimation result for autumn 2020, an estimation result for winter 2020, and an estimation result for spring 2021. Note that the time series information input to the learning model 307 does not have to be information with a seasonal cycle, and may be various time series information, such as information with a one-year cycle, a one-month cycle, a one-week cycle, or a one-day cycle. The learning model 307 is a learning model that has undergone machine learning to estimate and output information for the next cycle based on the input time series information.

The learning model 307 according to the third embodiment may have the configuration of a learning model that handles time-series information, such as a recurrent neural network (RNN), long short-term memory (LSTM), sequence-to-sequence (Seq2Seq), or transformer. The learning model 307 is generated by using pre-collected time-series information to create training data using information at a certain point in time as input and information at the next point in time as the correct answer label, and then performing supervised machine learning using the created training data.

The server device 3 according to the third embodiment estimates the usage time of electrical devices based on power information acquired from the sensors 1 installed in the facility 101, stores the information on the estimated usage time of the electrical devices, and also aggregates and stores the usage time for a predetermined period, such as for each season or month. The server device 3 acquires input information including the latest usage time of the electrical devices and attribute information of the subject, as well as a predetermined number of previous input information, at a predetermined cycle, for example, for each season or month, and inputs this information in chronological order to the learning model 307 according to the third embodiment. The server device 3 acquires the estimated result of the state of cognitive function finally output by the learning model 307 as the estimated result of the future state of cognitive function of the subject.

The server device 3 transmits information regarding the future state of cognitive function estimated using the learning model 307 to the terminal device 5, along with information regarding the past state of cognitive function. Upon receiving this information, the terminal device 5 can display, for example, an image showing the time-series changes in the subject's cognitive function state in the form of a graph or the like on the display unit 54. Furthermore, for example, if the server device 3 estimates that the future state of cognitive function will worsen compared to the current state, it transmits information to that effect to the terminal device 5, thereby notifying the user.

In the information processing system according to the third embodiment configured as described above, the server device 3 inputs time-series input information, including the usage time of each electrical device in the facility 101 and the subject's attribute information, to the learning model 307, and obtains an estimation result of the subject's future cognitive function state output by the learning model 307. As a result, the information processing system according to the third embodiment can be expected to estimate the subject's future cognitive function state.

In the third embodiment, the learning model 307 is configured to estimate the future state of cognitive function, but this is not limited to this and it may also be configured to estimate the future state of depression. Furthermore, the learning model 307 may be configured to output the current state of cognitive function rather than the future state of cognitive function. In other words, for input information from time T1 to time Tn, the learning model 307 may output the state of cognitive function at time Tn.

Furthermore, in the information processing system according to embodiment 3, the usage time of the electrical device and the subject's attribute information are input to the learning model 307 as a set of input information, but this is not limited to this. The time-series information input to the learning model 307 may be, for example, only the usage time of the electrical device, or may include information other than the usage time and attribute information.

Furthermore, the other configurations of the information processing system according to embodiment 3 are the same as those of the information processing systems according to embodiments 1 and 2, so similar parts are given the same reference numerals and detailed descriptions are omitted.

The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the present invention is defined by the claims, not by the meaning described above, and is intended to include all modifications within the meaning and scope of the claims.

REFERENCE SIGNS LIST 1 Sensor 3 Server device 5 Terminal device 7 Learning model 31 Processing unit 31a Learning model generation unit 31b Usage time estimation unit 31c Cognitive function estimation unit 32 Storage unit 32a Server program 32b Training data storage unit 32c Learning model storage unit 32d Subject information DB
32e Electric power information DB
33 Communication unit 51 Processing unit 51a Display processing unit 52 Memory unit 52a Program 53 Communication unit 54 Display unit 55 Operation unit 98, 99 Recording medium 101 Facility 102 User 111 Usage time display area 112 Cognitive function display area 207 Learning model 307 Learning model N Network

Claims (14)

Obtain estimated usage information for each electrical device in the facility,
When usage status information estimated for each electrical device in a facility is input, machine learning is performed to output cognitive function information regarding the cognitive function of a subject using the facility , and one learning model is selected according to the household composition of the subject related to the acquired usage status information from among a plurality of learning models that have been separately machine-learned according to the household composition of the subject;
inputting the acquired usage information into the selected learning model, and acquiring cognitive function information output by the learning model;
and outputting the acquired cognitive function information. The usage status information is a usage time of each electrical device estimated from the power consumption or current consumption of each electrical device.
2. The computer program of claim 1. The learning model receives the usage information and the subject's age, gender, or educational background as input information, and outputs cognitive function information of the subject according to the input information.
3. A computer program according to claim 1 or claim 2. The cognitive function information output by the learning model is information that classifies the subject's cognitive function as being normal, mild cognitive impairment, or cognitive impairment.
A computer program according to any one of claims 1 to 3. The cognitive function information output by the learning model is information classifying whether the subject is in a depressed state or not.
A computer program according to any one of claims 1 to 3. The learning model receives input of the time-series usage information and outputs the time-series cognitive function information.
A computer program according to any one of claims 1 to 5. The computer program according to claim 1 , further comprising: a process for displaying the usage status information and the cognitive function information on a display unit of a terminal device. outputting display information that associates information related to a period, information related to the estimated usage time of each electrical device for the period, and information related to the cognitive function of the subject for the period;
A computer program according to any one of claims 1 to 7. The plurality of learning models include a learning model for single-person households and a learning model for married couple households,
A computer program according to any one of claims 1 to 8. Correlate usage status information estimated for each electrical device in the facility with cognitive function information regarding the cognitive functions of subjects who use the facility , and acquire multiple types of training data created according to the household composition of the subjects;
By performing machine learning separately according to the household composition using the acquired multiple types of training data, multiple learning models are generated according to the household composition , which output cognitive function information of the subject when usage status information estimated for each electrical device in the facility is input.
A computer program that causes a computer to perform a process. Obtain estimated usage information for each electrical device in the facility,
When usage status information estimated for each electrical device in a facility is input, machine learning is performed to output cognitive function information regarding the cognitive function of a subject using the facility , and one learning model is selected according to the household composition of the subject related to the acquired usage status information from among a plurality of learning models that have been separately machine-learned according to the household composition of the subject;
inputting the acquired usage information into the selected learning model, and acquiring cognitive function information output by the learning model;
A cognitive function estimation method for estimating the cognitive function of a subject based on the acquired cognitive function information. Correlate usage status information estimated for each electrical device in the facility with cognitive function information regarding the cognitive functions of subjects who use the facility , and acquire multiple types of training data created according to the household composition of the subjects;
By performing machine learning separately according to the household composition using the acquired multiple types of training data, multiple learning models are generated according to the household composition , which output cognitive function information of the subject when usage status information estimated for each electrical device in the facility is input.
How to generate a learning model. a usage status information acquisition unit that acquires estimated usage status information for each electrical device in the facility;
a cognitive function information acquisition unit that performs machine learning to output cognitive function information related to the cognitive functions of a subject who uses the facility when usage information estimated for each electrical device in the facility is input , and selects one learning model according to the household composition of the subject related to the acquired usage information from a plurality of learning models that have been separately machine-learned according to the household composition of the subject, inputs the usage information acquired by the usage information acquisition unit into the selected learning model, and acquires the cognitive function information output by the learning model;
and an estimation unit that estimates the cognitive function of the subject based on the acquired cognitive function information. an acquisition unit that associates usage status information estimated for each electrical device in the facility with cognitive function information regarding the cognitive functions of subjects who use the facility , and acquires multiple types of training data created according to the household composition of the subjects;
and a generation unit that generates multiple learning models according to household composition by performing machine learning separately according to household composition using the acquired multiple types of training data, and outputs cognitive function information of a subject when usage information estimated for each electrical device in a facility is input.