JP7606067B2 - Environmental Control System - Google Patents

Description

Related to a learning device, an estimation device, and an environmental adjustment system.

Non-Patent Document 1 (Paper: The effect of body temperature changes before sleep on falling asleep (general)) discloses that deep body temperature affects sleep.

Accurately estimates the sleep state of a sleeper, taking into account that deep body temperature affects sleep.

The learning device of the first aspect learns the quality of sleep of a sleeping person. The learning device includes an acquisition unit, a learning unit, and a generation unit. The acquisition unit acquires state variables. The state variables are features related to the deep body temperature of the sleeping person or features related to the skin temperature of the sleeping person. The features related to the deep body temperature are determined based on at least a first deep body temperature when the sleeping person falls asleep. The features related to the skin temperature are determined based on at least a first skin temperature when the sleeping person falls asleep. The learning unit learns by associating the state variables with the quality of sleep. The generation unit generates a learning model based on the results of learning by the learning unit. The learning model takes as input at least the features related to the deep body temperature or the features related to the skin temperature of the sleeping person, and outputs the quality of sleep.

This allows the learning device to generate a learning model that can accurately estimate the quality of a sleeper's sleep as a sleeping state using deep body temperature.

The learning device of the second aspect is the learning device of the first aspect, and the feature amount related to the deep body temperature is further determined based on a second deep body temperature obtained before the first deep body temperature. The feature amount related to the skin temperature is further determined based on a second skin temperature obtained before the first skin temperature.

This allows for more accurate learning results.

The learning device of the third aspect is the learning device of the first or second aspect, and the acquisition unit further acquires features related to the sleeper's normal deep body temperature or features related to the sleeper's normal skin temperature.

This allows for more accurate learning results.

The learning device of the fourth aspect is a learning device of any one of the first aspect to the third aspect, in which the acquisition unit acquires the ambient temperature when the sleeper falls asleep, the fingertip blood flow of the sleeper, or an RGB image of the sleeper as a state variable.

This allows for more accurate learning results.

The learning device of the fifth aspect is a learning device of any one of the first to fourth aspects, in which the quality of sleep is determined based on at least one of the following: the time from when the sleeper gets into bed until he or she falls asleep, the deep sleep rate of the sleeper, the number of times the sleeper wakes up during the night, the time the sleeper wakes up during the night, a questionnaire on sleep quality administered to the sleeper, a questionnaire on daytime performance administered to the sleeper, the amount of hormone secretion by the sleeper, and the hormone concentration of the sleeper.

The learning device of the sixth aspect is a learning device of any one of the first to fifth aspects, in which the feature quantity related to the deep body temperature is determined based on at least one of the amount of change in the deep body temperature of the sleeper over a specified period, the maximum value, the minimum value, the mode, the average value, the difference between the maximum and minimum values, the maximum or minimum slope, the average slope, and an index related to the change.

The learning device of the seventh aspect is a learning device of any one of the first to sixth aspects, in which the feature quantity related to skin temperature is determined based on at least one of the amount of change in the skin temperature of the sleeper over a specified period of time, the maximum value, the minimum value, the mode, the average value, the difference between the maximum and minimum values, the maximum or minimum slope, the average slope, and an index related to the change.

The learning device of the eighth aspect is a learning device of any one of the first aspect to the seventh aspect, and the acquisition unit further acquires, as a state variable, any one of the humidity, illuminance, chromaticity, scent, and airflow surrounding the sleeper when the sleeper falls asleep.

This allows for more accurate learning results.

The learning device of the ninth aspect is the learning device of any one of the first aspect to the eighth aspect, in which the learning unit learns using a plurality of teacher data. The teacher data includes state variables and sleep quality.

The estimation device of the tenth aspect estimates sleep quality using a learning model generated based on the results of learning by a learning device of any one of the first to ninth aspects.

This allows the estimation device to estimate sleep quality with high accuracy.

The estimation device of the eleventh aspect is the estimation device of the tenth aspect, further comprising a deep body temperature feature estimation unit. The deep body temperature feature estimation unit estimates the feature related to the deep body temperature of the sleeper from the feature related to the skin temperature.

This allows the estimation device to reduce the burden on the sleeper caused by measuring deep body temperature.

The estimation device of the twelfth aspect is the estimation device of the tenth aspect, in which the deep body temperature feature estimation unit estimates the feature related to the deep body temperature of the sleeping person by inputting the feature related to the skin temperature of the sleeping person based on the learning results of a learning model that has learned the feature related to the deep body temperature of the sleeping person in association with the feature related to the skin temperature of the sleeping person. The feature related to the skin temperature of the sleeping person is determined based on at least a first skin temperature when the sleeping person falls asleep. The feature related to the deep body temperature of the sleeping person is determined based on at least a first deep body temperature when the sleeping person falls asleep.

An environmental adjustment system according to a thirteenth aspect includes an estimation device according to any one of the tenth to twelfth aspects, and an environmental adjustment device. The environmental adjustment device includes an actuator and a control unit. The actuator adjusts the environment around the sleeper. The control unit controls the operation of the actuator based on the sleep quality of the sleeper estimated by the estimation device. This enables the environmental adjustment system to adjust the environment around the sleeper and improve the sleep quality of the sleeper.

FIG. 1 illustrates an example of a learning system and an estimation system. FIG. 1 is a diagram illustrating a configuration of a learning system. FIG. 2 is a diagram showing the configuration of an estimation system and an environment adjustment system 3.

First Embodiment
(1) Overview Fig. 1 is a diagram showing an example of a learning system 1, an estimation system 2, and an environmental adjustment system 3 according to the first embodiment. The learning system 1 acquires various data, learns based on the acquired data, and generates a trained learning model based on the learning results. The estimation system 2 acquires various data, and estimates the sleep quality of a sleeper using the trained learning model generated in the learning system 1. The environmental adjustment system 3 adjusts the environment around the sleeper based on the sleep quality of the sleeper estimated by the estimation system 2.
The learning system 1, the estimation system 2, and the environment adjustment system 3 will be described in detail below.

(2) Learning System 1
2 is a schematic diagram showing an example of a learning system 1 according to the first embodiment. The learning system 1 mainly includes a data acquisition device 10, a teacher data generation device 20, and a learning device 30. Each device is capable of communicating with each other via a network such as the Internet.

(2-1) Data Acquisition Device 10
The data acquisition device 10 includes, for example, a first acquisition device 11 that acquires data on the temperature around the sleeper, a second acquisition device 12 that acquires data on the skin temperature of the sleeper, a third acquisition device 13 that acquires data on the deep body temperature of the sleeper, and a fourth acquisition device 14 that acquires data on the quality of sleep of the sleeper. In the first embodiment, the first acquisition device 11, the second acquisition device 12, and the third acquisition device 13 each acquire data continuously from 30 minutes before the sleeper goes to bed until the sleeper wakes up, but the data may be acquired at a predetermined time interval or at any timing. In the first embodiment, the fourth acquisition device 14 acquires data after the sleeper wakes up. The data acquired by the first acquisition device 11, the second acquisition device 12, the third acquisition device 13, and the fourth acquisition device 14 is linked to information on the time when the data was acquired.

The data acquired by the first acquisition device 11, the second acquisition device 12, the third acquisition device 13, and the fourth acquisition device 14 is transmitted to the teacher data generating device 20 via a network. Note that the first acquisition device 11, the second acquisition device 12, the third acquisition device 13, and the fourth acquisition device 14 may transmit only a portion of the acquired data in response to a request from the teacher data generating device 20 to the teacher data generating device 20.

(2-1-1) First acquisition device 11
The first acquisition device 11 is a device such as a temperature sensor or a thermograph, or a device including them. For example, the first acquisition device 11 may be an air conditioning device including a temperature sensor. Although not particularly limited, 11 is at least a device capable of acquiring data on the temperature around a sleeping person.

(2-1-2) Second acquisition device 12
The second acquisition device 12 is a device such as a skin temperature sensor or a device including a skin temperature sensor. For example, the second acquisition device 12 may be a wearable sensor including a skin temperature sensor. The acquisition device 12 is not particularly limited, but is at least a device capable of acquiring data on the skin temperature of a sleeper.

(2-1-3) Third acquisition device 13
The third acquisition device 13 is a device such as a deep body temperature sensor or a device including a deep body temperature sensor. For example, the third acquisition device 13 may be a tympanic thermometer. The third acquisition device 13 is Although there is no particular limitation, the device is at least capable of acquiring data on the deep body temperature of a sleeper.

(2-1-4) Fourth acquisition device 14
The fourth acquisition device 14 is a device for acquiring the quality of sleep of the sleeper. In the first embodiment, the quality of sleep is acquired by performing a sleep quality test on the sleeper and a daytime performance test. The sleep quality and daytime performance are determined based on a questionnaire. The questionnaire regarding sleep quality and daytime performance is, for example, the OSA sleep inventory MA version. The fourth acquisition device 14 is, for example, a mobile terminal or the like. The mobile terminal as the fourth acquisition device 14 may acquire data on sleep by, for example, having the sleeper complete the OSA sleep questionnaire MA version via the mobile terminal after waking up. The acquisition device 14 is not limited to a mobile terminal, and may be any device capable of acquiring the results of a sleeper taking the OSA sleep questionnaire MA version.

(2-2) Teacher data generating device 20
The teacher data generating device 20 may be a computer having a known hardware configuration such as a CPU (Central Processing Unit), memory, storage, an input/output interface, a communication interface, an input device, a display device, and a data bus, a known operation system, etc. installed, and further having a server function. A GPU (Graphics Processing Unit) may also be provided as necessary. Alternatively, it may be a virtual server provided using one or more computers. It is preferable that the teacher data generating device 20 in the present disclosure has a large capacity memory.

The teacher data generating device 20 has a data acquisition unit 21, a feature extraction unit 22, a teacher data generating unit 23, an acquired data storage unit 24, and a teacher data storage unit 25.

(2-2-1) Data Acquisition Unit 21
The data acquisition unit 21 performs a process of acquiring various data acquired by the data acquisition device 10 by receiving the data via a network. The various data acquired by the data acquisition unit 21 are stored in the acquired data storage unit 24. The various data stored in the acquired data storage unit 24 are linked with information on the time when each data was acquired by the data acquisition device 10.

In the first embodiment, the data acquisition unit 21 acquires data of the first ambient temperature and the second ambient temperature from the first acquisition device 11, data of the first skin temperature and the second skin temperature from the second acquisition device 12, data of the first deep body temperature and the second deep body temperature from the third acquisition device 13, and data of the quality of sleep from the fourth acquisition device 14. Each data will be described below. In the following description, the time when the sleeper falls asleep is preferably acquired at the timing when the sleeper falls asleep by measuring the sleeper's falling asleep using a sensor or the like, but is not limited to this. The time when the sleeper falls asleep is at least the timing after the sleeper goes to bed, and may be, for example, the timing after a predetermined time has passed since the sleeper goes to bed. In the first embodiment, the time when the sleeper falls asleep is 30 minutes after the sleeper goes to bed, but is not limited to this.

(2-2-1-1) Data on the first ambient temperature and the second ambient temperature The first ambient temperature is the temperature around the sleeper when the sleeper falls asleep. The second ambient temperature is the temperature around the sleeper acquired at a timing earlier than the first ambient temperature. The second ambient temperature data is acquired at least at a timing before the sleeper goes to bed. For example, the second ambient temperature data may be data acquired any time before the sleeper goes to bed. In the first embodiment, the second ambient temperature data is data on the temperature around the sleeper 30 minutes before the sleeper goes to bed, but is not limited to this.

(2-2-1-2) Data on First Skin Temperature and Second Skin Temperature The first skin temperature is the skin temperature when the sleeper falls asleep.

The second skin temperature is the skin temperature of the sleeper acquired at a timing earlier than the first skin temperature. It is preferable that the second skin temperature data is acquired at a timing before the sleeper goes to bed. For example, it is preferable that the second skin temperature data is data acquired any time before the sleeper goes to bed. In the first embodiment, the second skin temperature data is data of the sleeper's skin temperature 30 minutes before the sleeper goes to bed, but is not limited to this.

(2-2-1-3) Data on the first deep body temperature and the second deep body temperature The first deep body temperature is the deep body temperature when the sleeper falls asleep. The second deep body temperature is the deep body temperature of the sleeper acquired at a timing earlier than the first deep body temperature. The second deep body temperature data is preferably acquired at a timing before the sleeper goes to bed. For example, the second deep body temperature data is preferably data acquired any time before the sleeper goes to bed. In the first embodiment, the second deep body temperature data is data on the deep body temperature of the sleeper 30 minutes before the sleeper goes to bed, but is not limited to this.

(2-2-1-4) Sleep quality data Sleep quality is determined based on the OSA Sleep Questionnaire MA version as described above. Sleep quality data is obtained after the sleeper completes the OSA Sleep Questionnaire MA version after waking up.

(2-2-2) Feature Extraction Unit 22
The feature extraction unit 22 performs processing to extract skin temperature feature quantities and deep body temperature feature quantities from the first skin temperature and second skin temperature data and the first deep body temperature and second deep body temperature data stored in the acquired data storage unit 24. The skin temperature feature quantities and deep body temperature feature quantities extracted by the feature extraction unit 22 are stored in the acquired data storage unit 24. The skin temperature feature quantities and deep body temperature feature quantities will be described below.

(2-2-2-1) Skin Temperature Feature The skin temperature feature is a feature related to the skin temperature of a sleeper, and is extracted from the data of the first skin temperature and the second skin temperature. Specifically, the skin temperature feature is a feature determined based on any of the change value, maximum value, minimum value, mode, average value, difference between the maximum value and the minimum value, maximum or minimum slope, average slope, and index related to the change of the first skin temperature and the second skin temperature. The index related to the change is, for example, an index indicating a change tendency, such as the slope of the deep body temperature correlated with the skin temperature. Note that the method of extracting the skin temperature feature from the data of the first skin temperature and the second skin temperature is not particularly limited, but it is preferable to select useful features by using a filter method, a wrapper method, an embedding method, or the like to select features. In addition, an interaction feature may be used.

(2-2-2-2) Deep Body Temperature Feature The deep body temperature feature is a feature related to the deep body temperature of the sleeper, and is extracted from the data of the first deep body temperature and the second deep body temperature. Specifically, the deep body temperature feature is a feature determined based on any of the change value, maximum value, minimum value, mode, average value, difference between the maximum value and the minimum value, maximum or minimum slope, average slope, and index related to the change of the first deep body temperature and the second deep body temperature. The index related to the change is, for example, an index indicating a change tendency, such as the slope of the skin temperature correlated with the deep body temperature. The method of extracting the deep body temperature feature from the data of the first deep body temperature and the second deep body temperature is not particularly limited, but it is preferable to select a useful feature by performing feature selection using a filter method, a wrapper method, an embedding method, or the like. In addition, an interaction feature may be used.

(2-2-3) Teacher data generation unit 23
The teacher data generating unit 23 extracts state variables and correct answer data from the acquired data storage unit 24 to generate teacher data (teacher data set). The teacher data is a data set for the learning device 30 to learn about sleep quality. In the first embodiment, data on the first ambient temperature, the second ambient temperature, the skin temperature feature, the deep body temperature feature, and the sleep quality are used as the teacher data.

Specifically, the teacher data generation unit 23 first extracts data on the first ambient temperature, the second ambient temperature, the skin temperature feature, the deep body temperature feature, and the quality of sleep from the acquired data storage unit 24. At this time, the teacher data generation unit 23 extracts the data based on the time information linked to each piece of data. Next, the teacher data generation unit 23 generates teacher data using the first ambient temperature, the second ambient temperature, the skin temperature feature, and the deep body temperature feature as state variables and the data on the quality of sleep as correct answer data. The teacher data generated by the teacher data generation unit 23 is stored in the teacher data storage unit 35.

(2-3) Learning device 30
The learning device 30 is a high-performance computer that includes a known hardware configuration, such as a CPU (Central Processing Unit), memory, storage, an input/output interface, a communication interface, an input device, a display device, and a data bus (not shown), has a known operating system installed, and further has a server function. A GPU (Graphics Processing Unit) may also be provided as necessary. Alternatively, the learning device 30 may be a virtual server provided by using one or more computers.

The learning device 30 includes a teacher data acquisition unit 31, a learning unit 32, and a learning model generation unit 33.

(2-3-1) Teacher data acquisition unit 31
The teacher data acquiring unit 31 acquires a plurality of teacher data from the teacher data generating device 20 via a network. The teacher data acquired by the teacher data acquiring unit 31 is stored in a memory or the like.

(2-3-2) Learning unit 32
A learning model is provided in advance in the learning unit 32. The learning unit 32 causes the learning model to learn (for example, supervised learning) based on the multiple teacher data acquired by the teacher data acquisition unit 31. The learning method performed by the learning model is not particularly limited, but it is preferable to use multiple regression analysis, random forest, or the like.

The learning model in the first embodiment learns with the aim of outputting an estimated value of sleep quality from input data input to the estimation system 2 (Figure 3) described below.

(2-3-3) Learning model generation unit 33
The learning model generation unit 33 outputs the learning model as a trained learning model at a stage where the learning in the learning unit 32 progresses and the estimation accuracy of the learning model is improved, thereby generating a learning model to be used in the estimation device 50. The learning model output from the learning model generation unit 33 is capable of outputting an estimated value of sleep quality from input data input to the estimation system 2 with a predetermined accuracy or higher.

The learning model generated by the learning model generation unit 33 is distributed to the estimation device 50 (described later) via a network. Alternatively, the learning model generated by the learning model generation unit 33 is distributed to the estimation device 50 via a storage medium.

(3) Features of Learning System 1 (3-1)
The learning device 30 included in the learning system 1 of the first embodiment is a learning device 30 that learns the quality of sleep of a sleeper. The learning device 30 includes a teacher data acquisition unit 31 as an acquisition unit, a learning unit 32, and a learning model generation unit 33 as a generation unit. The teacher data acquisition unit 31 acquires a first ambient temperature, a second ambient temperature, a feature amount related to a deep body temperature, and a feature amount related to a skin temperature as state variables. The first ambient temperature is the temperature around the sleeper when the sleeper falls asleep. The second ambient temperature is the temperature around the sleeper acquired before the first ambient temperature. The feature amount related to the deep body temperature is determined based on at least the first deep body temperature when the sleeper falls asleep. The feature amount related to the skin temperature is determined based on at least the first skin temperature when the sleeper falls asleep. The learning unit 32 learns by associating the state variables with the quality of sleep. The learning model generation unit 33 generates a learning model based on the result of learning by the learning unit 32. The learning model estimates sleep quality using input features related to the first ambient temperature, the second ambient temperature, the deep body temperature, and the skin temperature.

Research has been conducted on the effect of body temperature fluctuations before sleep on falling asleep, as in Non-Patent Document 1. For example, Non-Patent Document 1 reports that extreme drowsiness occurs during periods when deep body temperature drops, and in particular, the time at which a rapid drop in deep body temperature occurs has a significant effect on sleep quality.

Specifically, when humans become sleepy, they release heat by transferring their core body temperature to the extremities and expanding the temperature-regulating blood vessels called AVAs (arteriovenous anastomoses) that exist in the palms of the hands, soles of the feet, and face. This lowers the core body temperature and increases the efficiency of rest for the brain and body during sleep. By raising body temperature during the day by exercising or bathing, the core body temperature tends to drop when going to bed, but if the core body temperature does not rise during the day or the bedroom is hot, the core body temperature does not drop and deep sleep cannot be achieved, resulting in poor quality sleep. Deep sleep is characterized by being most common in the first three hours after falling asleep and becoming less frequent thereafter. It has been found that the transition of core body temperature affects the comfort of sleep, which can be rephrased as deep and high-quality sleep. To summarize the above, it has been found that the relationship between core body temperature, skin temperature, and sleep depth in ideal sleep is as shown in Figure 5. Additionally, it was found that with age, the time it takes to fall asleep increases, the percentage of deep sleep decreases, and the total sleep time decreases.

The learning device 30 of the first embodiment acquires features related to the deep body temperature of a sleeper as a state variable, and through learning, is capable of generating a learning model that can more accurately estimate the quality of sleep as a sleep state of the sleeper.

(3-2)
The second feature amount acquired by the learning device 30 is determined based on at least the first skin temperature and the second skin temperature. The first skin temperature is the skin temperature of the sleeper when the sleeper falls asleep. The second skin temperature is the skin temperature of the sleeper acquired before the first skin temperature.

In this embodiment, the learning device 30 acquires data on the skin temperature of the sleeper 30 minutes after the sleeper goes to bed as the first skin temperature, and data on the skin temperature of the sleeper 30 minutes before the sleeper goes to bed as the second skin temperature. In addition, the learning device 30 of this embodiment acquires the first deep body temperature at the same time as the first skin temperature, and acquires the second deep body temperature at the same time as the second skin temperature.

This enables the learning device 30 to generate a learning model that outputs more accurate deep body temperature features.

(3-3)
The learning device 30 learns using a plurality of teacher data sets. The teacher data sets include data on sleep quality, data on a first ambient temperature, data on a second ambient temperature, a deep body temperature feature as a feature related to the deep body temperature, and a skin temperature feature as a feature related to the skin temperature.

In this embodiment, sleep quality is determined based on at least one of the following: the time it takes for the sleeper to fall asleep after getting into bed, the deep sleep rate, the number of awakenings during sleep, the duration of awakenings during sleep, and the OSA sleep inventory MA version.

The deep body temperature feature is determined based on at least one of the following: the amount of change in the deep body temperature of the sleeper during a specified period, the maximum value, the minimum value, the mode, the average value, the difference between the maximum and minimum values, the maximum or minimum slope, and the average slope.

Furthermore, the skin temperature feature is determined based on at least one of the following: the amount of change in the sleeper's skin temperature over a specified period of time, the maximum value, the minimum value, the mode, the average value, the difference between the maximum and minimum values, the maximum or minimum slope, and the average slope.

As a result, the learning device 30 of this embodiment is able to generate a learning model that outputs sleep quality with high accuracy.

(4) Modifications (4-1) Modification 1
In the first embodiment, the sleep quality is determined based on the OSA sleep questionnaire MA version, but the sleep quality may be determined based on other methods. Also, the fourth acquisition device 14 may be a device other than a mobile terminal.

For example, sleep quality may be determined from the sleeper's heart rate while asleep, body movements, the percentage of deep sleep time, the time it takes for the sleeper to fall asleep after getting into bed, the sleeper's hormone secretion amount, the sleeper's hormone concentration, or any questionnaire regarding sleep quality and daytime performance administered to the sleeper. The sleep quality data is not particularly limited, and the fourth acquisition device 14 is not particularly limited as long as it is capable of acquiring at least sleep quality data.

(4-2) Modification 2
The data acquisition device 10 may acquire data other than the data shown above. For example, the data acquisition device 10 may acquire environmental data, target data, or sleep data. Specifically, the environmental data is data such as humidity, illuminance, chromaticity, scent, or airflow around the sleeper. The target data is data such as the fingertip blood flow of the sleeper, an RGB image for estimating the fingertip blood flow of the sleeper, heart rate, heart rate variability, respiratory rate, galvanic skin reaction, amount of clothing/type of bedding, daytime activity status, and pillow temperature. The sleep data is data such as the body movement, sleep state, or time until falling asleep of the sleeper. Note that these data may be acquired by the first acquisition device 11, the second acquisition device 12, the third acquisition device 13, and the fourth acquisition device 14 included in the data acquisition device 10, or by other devices. The learning device 30 can use the data acquired by those devices as teacher data.

By using the above-mentioned data as state variables, the learning device 30 can generate a learning model that estimates core body temperature features with greater accuracy.

(4-3) Modification 3
In the first embodiment, the first ambient temperature and the second ambient temperature are acquired by the same device (first acquisition device 11), but the first ambient temperature and the second ambient temperature may be acquired by different devices.

The first acquisition device 11, etc. may also acquire data on the temperature around the sleeper at a specified time as a third ambient temperature in addition to the first ambient temperature and the second ambient temperature.

(4-4) Modification 4
In the first embodiment, the teacher data generation device 20 extracts the skin temperature feature and the deep body temperature feature, but the skin temperature feature and the deep body temperature feature may be extracted in the learning device 30. Alternatively, the skin temperature feature may be extracted in the second acquisition device 12, and the deep body temperature feature may be extracted in the third acquisition device 13. The process of extracting the feature from each data may be performed in any device.

(4-5) Modification 5
In the first embodiment, the learning system 1 extracts the skin temperature feature and the deep body temperature feature to learn, but the learning system 1 may use deep learning to learn. In other words, the learning unit 32 may automatically extract appropriate feature values from the skin temperature, the deep body temperature, the first ambient temperature, and the second ambient temperature to learn. In this case, the skin temperature and the deep body temperature may be acquired only when the sleeper falls asleep.

(4-6) Modification 6
In the first embodiment, the teacher data is a teacher data set related to a sleeping person. Here, the multiple teacher data used for learning by the learning unit 32 may be teacher data sets related to different targets (sleeping people).

On the other hand, the multiple teacher data used for learning by the learning unit 32 in the first embodiment may be a teacher data set related to a specific sleeper. In this case, the data acquisition device 10 may have a personal information acquisition device that acquires personal information of the specific sleeper. The personal information acquisition device is, for example, a mobile terminal, into which the personal information of the specific sleeper is input. The personal information is, for example, information such as the specific sleeper's normal skin temperature, normal core body temperature, age, sex, height, weight, and the average normal skin temperature of the same age.

This enables the learning device 30 to contribute to generating a learning model that estimates core body temperature features with high accuracy for a specific sleeper.

(5) Estimation System 2
3 is a schematic diagram showing an example of the estimation system 2 and the environment adjustment system 3 according to the first embodiment. The estimation system 2 has devices such as an input data acquisition device 40 and an estimation device 50. Each device is capable of communicating via a predetermined network such as the Internet.

(5-1) Input data acquisition device 40
The input data acquisition device 40 includes a fifth acquisition device 41 that acquires data on the temperature around the sleeping person, a sixth acquisition device 42 that acquires data on the skin temperature of the sleeping person, and a seventh acquisition device 43 that acquires data on the deep body temperature of the sleeping person. The fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 each acquire predetermined data.

In the first embodiment, the fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 each continuously acquire data starting 30 minutes before the sleeper goes to bed, but the data may be acquired at a predetermined time interval or at any timing. The data acquired by the fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 is linked to information on the time when the data was acquired.

The data acquired by the fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 is transmitted to the estimation device 50 via the network. Note that the fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 may transmit to the estimation device 50 only a portion of the acquired data in response to a request from the estimation device 50.

The fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43, and the data acquired by the fifth acquisition device 41, the sixth acquisition device 42, and the seventh acquisition device 43 will be described below. In the following description, when the sleeper falls asleep, it is preferable to measure the sleeper's fall asleep using a sensor or the like and acquire the data at the timing when the sleeper falls asleep, but this is not limited to this. The time when the sleeper falls asleep is at least the timing after the sleeper goes to bed, and may be, for example, the timing after any time has passed since the sleeper goes to bed. In the first embodiment, the data of the second ambient temperature is data of the temperature around the sleeper 30 minutes before the sleeper goes to bed, but this is not limited to this.

(5-1-1) Fifth acquisition device 41
The fifth acquisition device 41 is a device such as a temperature sensor or a thermograph, or a device including them. For example, the fifth acquisition device 41 may be an air conditioning device including a temperature sensor. The device 41 is not particularly limited, but is at least a device capable of acquiring data on the temperature around the sleeper.

In the first embodiment, the fifth acquisition device 41 acquires at least the third ambient temperature and the fourth ambient temperature.

The third ambient temperature is data on the temperature around the sleeper when the sleeper falls asleep. The fourth ambient temperature is data on the temperature around the sleeper that was acquired before the third ambient temperature. The fourth ambient temperature is acquired at least before the sleeper goes to bed. For example, the fourth ambient temperature may be data acquired any time before the sleeper goes to bed. In the first embodiment, the fourth ambient temperature data is data on the temperature around the sleeper 30 minutes before the sleeper goes to bed, but is not limited to this.

(5-1-2) Sixth acquisition device 42
The sixth acquisition device 42 is a device such as a skin temperature sensor or a device including the same. For example, the sixth acquisition device 42 may be a wearable sensor including a skin temperature sensor or the like. The device 42 is not particularly limited, but is at least a device capable of acquiring data on the third skin temperature and the fourth skin temperature of the sleeper.

The third skin temperature is the skin temperature of the sleeper when he/she falls asleep. The fourth skin temperature is the skin temperature of the sleeper acquired at a time before the third skin temperature. The fourth skin temperature data is preferably acquired before the sleeper goes to bed. For example, the fourth skin temperature data is preferably data acquired any time before the sleeper goes to bed. In the first embodiment, the fourth skin temperature data is data of the sleeper's skin temperature 30 minutes before the sleeper goes to bed, but is not limited to this.

(5-1-3) Seventh acquisition device 43
The seventh acquisition device 43 is a device such as a deep body temperature sensor or a device including the same. For example, the seventh acquisition device 43 may be a tympanic thermometer or the like. The seventh acquisition device 43 is not particularly limited. Although it is not intended to be used as a temperature measuring device, it is at least an apparatus capable of obtaining data on the third and fourth deep body temperatures of a sleeper.

The third deep body temperature is the deep body temperature when the sleeper falls asleep. The fourth deep body temperature is the deep body temperature of the sleeper acquired at a timing earlier than the third deep body temperature. The fourth deep body temperature data is preferably acquired at a timing before the sleeper goes to bed. For example, the fourth deep body temperature data is preferably data acquired any time before the sleeper goes to bed. In the first embodiment, the fourth deep body temperature data is data of the sleeper's deep body temperature 30 minutes before the sleeper goes to bed, but is not limited to this.

(5-2) Estimation device 50
The estimation device 50 includes a known hardware configuration including a CPU (Central Processing Unit), a memory, a storage device, an input/output interface, a communication interface, an input device, a display device, and a data bus, and has a known operation system installed. The computer is a high-performance computer having a server function. If necessary, a GPU (Graphics Processing Unit) may be provided. Alternatively, the computer is provided using one or more computers. The virtual server may be a virtualized virtual server.

The memory or other storage unit of the estimation device 50 stores a trained learning model previously generated by the learning system 1 and various programs for executing this learning model.

The estimation device 50 includes an input data acquisition unit 51, a feature extraction unit 52, an estimation unit 53, and an output unit 54.

(5-2-1) Input data acquisition unit 51
The input data acquisition unit 51 acquires the third ambient temperature and fourth ambient temperature data acquired by the fifth acquisition device 41, the third skin temperature and fourth skin temperature data acquired by the sixth acquisition device 42, and the third deep body temperature and fourth deep body temperature data acquired by the seventh acquisition device 43 by receiving them via a network. The data acquired by the input data acquisition unit 51 is stored in a memory or the like.

(5-2-2) Feature Extraction Unit 52
The feature extraction unit 52 extracts skin temperature features from the third skin temperature and fourth skin temperature data acquired by the input data acquisition unit 51. The feature extraction unit 52 also extracts deep body temperature features from the third deep body temperature and fourth deep body temperature data acquired by the input data acquisition unit 51. The skin temperature features and deep body temperature features are substantially the same features as the skin temperature features and deep body temperature features extracted by the feature extraction unit 22 of the learning system, and therefore a description thereof will be omitted.

(5-2-3) Estimation unit 53
The estimation unit 53 estimates the third and fourth ambient temperature data acquired by the input data acquisition unit 51 and the skin temperature feature amount and the deep body temperature feature amount extracted by the feature amount extraction unit 22 as input data. The sleep quality is estimated by classifying the quality of the sleep of the sleeper into three stages (good, normal, poor, etc.), for example. This is supervised learning. The estimation method is not particularly limited, but it is preferable to use multiple regression analysis, random forest, etc. For example, the results of experiments using these methods showed that the accuracy rate according to the input data was The results were 0.762, 0.930, and 0.803, respectively, and it was possible to estimate the quality of sleep with high accuracy.

(5-2-4) Output unit 54
The output unit 54 outputs the estimated value of the sleep quality estimated by the estimation unit 53. Specifically, the output unit 54 transmits the estimated value of the sleep quality as an output to another device, or displays it on a display or the like.

(6) Features of the Estimation System 2 The estimation device 50 included in the estimation system 2 in the first embodiment estimates the sleep quality of a sleeper using a trained learning model generated based on the learning results of the learning device 30.

As a result, the estimation system 2 of the first embodiment can accurately estimate the sleep quality of a sleeper using the learning model generated in the learning system 1.

(7) Modifications (7-1) Modification 1
The input data acquisition unit 51 may acquire input data other than the data described above. For example, the input data acquisition unit 51 may acquire environmental data, target data, sleep data, etc. The input data acquisition unit 51 acquires input data according to the specifications of the learning model.

(7-2) Modification 2
In the first embodiment, the estimation device 50 extracts the skin temperature feature amount, but the skin temperature feature amount may be extracted by the sixth acquisition device 42. Alternatively, the skin temperature feature amount may be extracted by a device other than the sixth acquisition device 42.

(7-3) Modification 3
In the first embodiment, the estimation system 2 extracts the skin temperature feature amount and performs learning, but the estimation system 2 may perform learning using deep learning. In other words, the estimation unit 53 may automatically extract appropriate feature amounts from the third ambient temperature, the fourth ambient temperature, the third skin temperature, and the fourth skin temperature and perform learning.

(8) Environmental Adjustment System 3
The environment adjustment system 3 is a system that adjusts the environment around the sleeper based on the sleep quality of the sleeper estimated by the estimation system 2. The environment adjustment system 3 includes the estimation device 50 of the first embodiment and an environment adjustment device 60. An example of the environment adjustment device 60 will be described below.

(8-1) Environment adjustment device 60
The environmental adjustment device 60 is a device that adjusts the environment around a sleeper, and is, for example, an air conditioner. The environmental adjustment device 60 mainly includes an acquisition unit 61, a control unit 62, an actuator 63, and Equipped with.

(8-1-1) Acquisition unit 61
The acquisition unit 61 acquires an estimated value of sleep quality from the estimation device 50 via a network such as the Internet. The acquisition unit 61 also acquires various data from the input data acquisition device 40 via the network. The unit 61 stores the information acquired from the estimation device 50 and the input data acquisition device 40 in a storage device of the control unit 62, which will be described later.

(8-1-2) Control unit 62
The control unit 62 is realized by a computer and includes a control and arithmetic device and a storage device (not shown). A processor such as a CPU or a GPU can be used as the control and arithmetic device. The control and arithmetic device reads a program stored in the storage device and performs predetermined arithmetic processing according to the program. Furthermore, the control and arithmetic device can write the results of calculations to the storage device and read information stored in the storage device according to the program.

Specifically, the storage device of the control unit 62 in the first embodiment stores information acquired by the acquisition unit 61. The storage device also stores control instructions for a predetermined period before the acquisition unit 61 acquires an estimated value of sleep quality. The control instructions are instruction signals for controlling the operation of the actuator 63, which will be described later.

The control unit 62's control and calculation device performs calculations based on the information stored in the storage device and the control instructions, and sends control instructions to the actuator 63 to improve sleep quality.

(8-1-3) Actuator 63
The actuator 63 operates in response to control instructions received from the control and arithmetic device of the control unit 62. The actuator 63 is, for example, a compressor, an expansion valve, a fan, a flap, or the like of an air conditioner.

(9) Features The environment adjustment system 3 in the first embodiment includes an estimation device 50 and an environment adjustment device 60. The environment adjustment device 60 has an actuator 63 and a control unit 62. The actuator 63 adjusts the environment around the sleeper. The control unit 62 controls the operation of the actuator based on the quality of sleep of the sleeper estimated by the estimation device 50.

As described in Non-Patent Document 1, the environmental adjustment system 3 uses the human characteristic of feeling very sleepy when the core body temperature drops, and for example, when it predicts that the quality of sleep will decrease, it lowers the temperature around the sleeper (e.g., by 1 degree) so that the quality of sleep improves (the core body temperature decreases). Alternatively, the environmental adjustment system 3 performs environmental adjustment to strengthen the airflow around the sleeper. In this way, the environmental adjustment system 3 contributes to improving the quality of sleep of the sleeper.

(10) Modifications In the first embodiment, the acquisition unit 61 acquires various data from the input data acquisition device 40 , but the acquisition unit 61 may acquire various data from sensors appropriately attached to the environment adjustment system 3 .

The environmental adjustment system 3 and the estimation system 2 may be stored in the same housing. Therefore, the acquisition unit 61 may acquire information about each of them without going through a network.

Second Embodiment
Hereinafter, the learning system 1 and the estimation system 2 according to the second embodiment of the present invention will be described in detail, focusing on the differences from the systems according to the first embodiment.

(11) Learning System 1 of the Second Embodiment
(11-1) Data Acquisition Device 10
The data acquisition device 10 in the learning system 1 of the second embodiment is generally the same as the data acquisition device 10 of the first embodiment.

(11-2) Teacher data generating device 20
In the learning system 1 of the second embodiment, the teacher data generating device 20 generates teacher data for the learning device 30 to learn the deep body temperature feature of a sleeping person. For convenience, the teacher data generated to learn the sleep quality of the sleeping person described in the first embodiment is referred to as first teacher data, and the teacher data to learn the deep body temperature feature is referred to as second teacher data.

The teacher data generation unit 23 of the teacher data generation device 20 first extracts data on the first ambient temperature, the second ambient temperature, the skin temperature feature, and the deep body temperature feature from the acquired data storage unit 24. Next, the teacher data generation unit 23 generates second teacher data using the first ambient temperature, the second ambient temperature, and the skin temperature feature as state variables and the data on the deep body temperature feature as correct answer data. The second teacher data generated by the teacher data generation unit 23 is stored in the teacher data storage unit 35.

Note that the teacher data generation unit 23 in the second embodiment also generates the first teacher data, as in the first embodiment.

(11-3) Learning device 30
The learning device 30 causes the learning model to learn based on a plurality of teacher data acquired by the teacher data acquisition unit 31. For convenience, the learning model that learns the sleep quality of a sleeper using the first teacher data is referred to as the first learning model, and the learning model that learns the deep body temperature feature amount using the second teacher data is referred to as the second learning model. The first learning model is the learning model described in the first embodiment.

The learning device 30 performs learning using a plurality of second teacher data acquired by the teacher data generating device 20, with the aim of outputting an estimated value of the deep body temperature feature from the input data input to the estimation system 2.

The learning model generation unit 33 generates a second learning model to be used in the estimation device 50 by outputting the second learning model as a trained second learning model at a stage where the learning in the learning unit 32 progresses and the estimation accuracy of the learning model is improved. The second learning model output from the learning model generation unit 33 is capable of outputting an estimate of the deep body temperature feature from the input data input to the estimation system 2 with a predetermined accuracy or higher.

In addition, the learning device 30 of the second embodiment also generates a first learning model that learns the sleep quality of a sleeper using the first teacher data, as in the first embodiment.

(12) Features of the Learning System 1 of the Second Embodiment The learning device 30 included in the learning system 1 of the second embodiment is a learning device 30 that learns the quality of sleep of a sleeper. The learning device 30 includes a teacher data acquisition unit 31 as an acquisition unit, a learning unit 32, and a learning model generation unit 33 as a generation unit. The teacher data acquisition unit 31 acquires a first ambient temperature, a second ambient temperature, a feature amount related to a deep body temperature, and a feature amount related to a skin temperature as state variables. The first ambient temperature is the temperature around the sleeper when the sleeper falls asleep. The second ambient temperature is the temperature around the sleeper acquired before the first ambient temperature. The feature amount related to the deep body temperature is determined based on at least the first deep body temperature when the sleeper falls asleep. The feature amount related to the skin temperature is determined based on at least the first skin temperature when the sleeper falls asleep. The learning unit 32 learns by associating the state variables with the quality of sleep. The learning model generation unit 33 generates a learning model based on the result of learning by the learning unit 32. The learning model estimates the quality of sleep using the feature amounts related to the first ambient temperature, the second ambient temperature, and the skin temperature as input.

A sleeper's deep body temperature has a significant impact on the quality of sleep. However, to measure deep body temperature, a device such as a rectal thermometer or tympanic thermometer must be worn, and these devices place a greater burden on the sleeper than when measuring skin temperature.

The learning device 30 shown in the second embodiment acquires information for calculating features related to the deep body temperature of the sleeping person in advance, and through learning, generates a learning model that outputs a first feature, which is information related to the deep body temperature. This makes it possible to reduce the burden on the sleeping person associated with acquiring information related to the deep body temperature.

(13) Modification In the second embodiment, the first teacher data and the second teacher data are generated by the same teacher data generation device 20. However, this is not limited to this. The first teacher data and the second teacher data may be generated by different hardware.

In addition, the first learning model and the second learning model are trained on the same learning device 30, but this is not limited to the above. The first learning model and the second learning model may be trained on different hardware.

(14) Estimation System 2
(14-1) Input data acquisition device 40
The input data acquisition device 40 of the second embodiment includes the fifth acquisition device 41 and the sixth acquisition device 42 of the first embodiment. In other words, the input data acquisition device 40 of the second embodiment does not include the seventh acquisition device 43. The fifth acquisition device 41 and the sixth acquisition device 42 acquire substantially the same data as in the first embodiment.

(14-2) Estimation device 50
The estimation device 50 of the second embodiment stores a first learning model and a second learning model that have been trained in advance and generated by the learning system 1, as well as various programs for executing each learning model.

The estimation unit 53 estimates the deep body temperature feature using the second learning model with the data of the third and fourth ambient temperatures acquired by the input data acquisition unit 51 and the skin temperature feature extracted by the feature extraction unit 22 as input data. The estimation unit 53 can estimate whether the deep body temperature feature, for example, will be an increase of 0.2°C or more, a plateau (between a decrease of 0.2°C and an increase of 0.2°C), or a decrease of 0.2°C or more when comparing the average tympanic membrane temperature for each minute from 30 minutes to 16 minutes before falling asleep with the average tympanic membrane temperature for each minute from 15 minutes before falling asleep to when falling asleep. The estimation is, for example, supervised learning. Here, if there is variation in the number of samples, it is possible to use a method such as undersampling or oversampling. Table (1) shown below shows the accuracy rate in an experiment in which the deep body temperature feature was estimated using the estimation unit 53. In this way, a combination of experimental conditions for estimating the deep body temperature feature was compared.

The estimation device 50 of the second embodiment estimates sleep quality using the deep body temperature feature estimated by the estimation unit 53. Specifically, the estimation unit 53 estimates sleep quality using a first learning model, using as input data the third and fourth ambient temperature data acquired by the input data acquisition unit 51, the skin temperature feature extracted by the feature extraction unit 22, and the deep body temperature feature estimated by the estimation unit 53. The output unit 54 outputs an estimate of sleep quality estimated using the first learning model.

(15) Features The estimation device 50 included in the estimation system 2 in the second embodiment estimates the sleep quality of a sleeper by using a trained learning model generated based on the learning result of the learning device 30.

The estimation system 2 further includes an estimation unit 53 that functions as a deep body temperature feature estimation unit. The estimation unit 53 estimates the feature related to the deep body temperature of the sleeper from the first ambient temperature, the second ambient temperature, and the feature related to the skin temperature.

This enables the estimation device 50 to accurately estimate the quality of sleep of the sleeper, and also reduces the burden on the sleeper that would be caused by having to wear a device such as a rectal thermometer or tympanic thermometer to obtain information about deep body temperature.

(16) Modifications In the second embodiment, the estimation of the deep body temperature feature amount and the estimation of the sleep quality are performed by the same estimation device 50. However, this is not limited to this. The estimation of the deep body temperature feature amount and the estimation of the sleep quality may be performed by different hardware.

Although the embodiments of the present disclosure have been described above, it will be understood that various changes in form and details are possible without departing from the spirit and scope of the present disclosure as set forth in the claims.

3 Environmental Adjustment System 30 Learning Device 31 Acquisition Unit 32 Learning Unit 50 Estimation Device 53 Core Body Temperature Feature Estimation Unit 62 Control Unit 63 Actuator

<Paper> The effect of body temperature fluctuation before sleep on falling asleep (general) Shiro Oda http://id.nii.ac.jp/1136/00002386/

Claims (11)

An estimation device (50) that estimates the sleep quality of a sleeper using a learning model;
An environmental conditioning device (60);
Equipped with
The learning model is generated based on a result of learning by a learning device (30) that learns the sleep quality of the sleeper,
The learning device includes:
an acquisition unit (31) that acquires a feature amount related to the deep body temperature of the sleeping person or a feature amount related to the skin temperature of the sleeping person as a state variable;
a learning unit (32) that learns the state variables and the quality of sleep in association with each other;
A generation unit that generates the learning model based on the learning result of the learning unit (32) by inputting the feature amount related to the deep body temperature or the feature amount related to the skin temperature and estimating the quality of sleep of the sleeper;
having
The feature amount related to the deep body temperature is determined based on at least a first deep body temperature when the sleeper falls asleep,
the feature amount related to the skin temperature is determined based on at least a first skin temperature when the sleeper falls asleep,
The environmental adjustment device includes an actuator (63) that adjusts the environment around the sleeper, and a control unit (61) that controls the operation of the actuator (63),
The control unit (61) controls the operation of the actuator (63) based on the quality of sleep of the sleeper estimated by the estimation device (50);
When it is predicted that the quality of sleep will decrease, an environmental adjustment is made to improve the quality of sleep by lowering the temperature around the sleeper or by increasing the airflow around the sleeper.
Environmental control system (3). The feature amount related to the deep body temperature is further determined based on a second deep body temperature acquired before the first deep body temperature,
The feature amount related to the skin temperature is further determined based on a second skin temperature acquired before the first skin temperature.
An environmental conditioning system (3) according to claim 1. The acquisition unit (31) further acquires a feature amount related to the deep body temperature of the sleeping person at normal time or a feature amount related to the skin temperature of the sleeping person at normal time.
An environmental conditioning system (3) according to claim 1 or 2. The acquisition unit (31) acquires an ambient temperature when the sleeper falls asleep, a fingertip blood flow of the sleeper, or an RGB image of the sleeper as a state variable.
An environmental conditioning system (3) according to any one of claims 1 to 3. The quality of sleep learned by the learning unit is determined based on at least one of the following: a time from when the sleeper gets into bed until when he/she falls asleep, a deep sleep rate of the sleeper, a number of times the sleeper wakes up during the night, a time of the sleeper waking up during the night, a questionnaire on the quality of sleep administered to the sleeper, a questionnaire on daytime performance administered to the sleeper, a hormone secretion amount of the sleeper, and a hormone concentration of the sleeper.
An environmental conditioning system (3) according to any one of claims 1 to 4. The feature amount related to the deep body temperature is determined based on at least one of the amount of change in the deep body temperature of the sleeper during a predetermined period, the maximum value, the minimum value, the mode, the average value, the difference between the maximum value and the minimum value, the maximum or minimum slope, the average slope, and an index related to the change.
An environmental conditioning system (3) according to any one of claims 1 to 5. The feature amount related to the skin temperature is determined based on at least one of the amount of change in the skin temperature of the sleeper in a predetermined period, the maximum value, the minimum value, the mode, the average value, the difference between the maximum value and the minimum value, the maximum or minimum slope, the average slope, and an index related to the change.
An environmental conditioning system (3) according to any one of the preceding claims. The acquisition unit (31) further acquires, as the state variable, any one of humidity, illuminance, chromaticity, scent, and airflow around the sleeper when the sleeper falls asleep.
An environmental conditioning system (3) according to any one of the preceding claims. The learning unit (32) learns using a plurality of teacher data,
The teacher data includes the state variables and the sleep quality.
An environmental conditioning system (3) according to any one of the preceding claims. Further comprising a deep body temperature feature estimation unit (53),
The deep body temperature feature amount estimation unit estimates a feature amount related to the deep body temperature of the sleeping person from a feature amount related to the skin temperature of the sleeping person.
An environmental conditioning system (3) according to claim 1. The deep body temperature feature estimation unit (53)
estimating the feature amount related to the deep body temperature of the sleeping person using the feature amount related to the skin temperature of the sleeping person as an input based on a learning result of a learning model that has learned the feature amount related to the deep body temperature of the sleeping person in association with the feature amount related to the skin temperature of the sleeping person;
The feature amount related to the skin temperature of the sleeping person is determined based on at least a first skin temperature when the sleeping person falls asleep,
The feature amount related to the deep body temperature of the sleeping person is determined based on at least a first deep body temperature when the sleeping person falls asleep.
An environmental conditioning system (3) according to claim 10.

Images