JP7730645B2 - Information processing device, information processing method, program, and data structure - Google Patents

Description

The present invention relates to an information processing device, an information processing method, a program, and a data structure.

Technology for determining deterioration of storage cells in systems that use energy storage modules in which multiple storage cells are connected in series is known (Patent Document 1). In the technology proposed in Patent Document 1, a learning model determines the deterioration of the storage cells and outputs the presence or absence of environmental abnormalities based on the storage element information of the storage cells, actual measurement value time series data, and predicted value time series data.

Japanese Patent Application Laid-Open No. 2019-168451

Incidentally, for example, in data centers, storage battery systems are sometimes used as uninterruptible power supplies to ensure that the computers installed in the data center can operate at all times. Storage battery systems as uninterruptible power supplies charge multiple storage batteries, and in the event of a power outage, supply power from these batteries to the equipment in the data center.

Some battery storage systems have a configuration in which, for example, 100 or more batteries are connected in series. In such battery storage systems, there is a risk of losing power supply function if an abnormality occurs in one battery, causing a disconnection. Therefore, it is necessary to detect signs of abnormality in the batteries early and perform appropriate maintenance, such as replacement.

On the other hand, in countries where power outages are infrequent, battery storage systems rarely operate during actual power outages. Despite this, configuring a battery storage system to constantly monitor the status of each battery by installing sensors in each battery makes the battery system complex and expensive. Even if the battery storage system is configured so that it does not use constant monitoring by sensors, it may still be possible to properly maintain the battery storage system by manually inspecting the batteries at specified intervals (e.g., every one to several months).

Given this background, there is a demand for technology in battery storage systems that can accurately predict signs of abnormalities in the batteries while utilizing data measured during regular inspections.

The present invention was made in consideration of the above-mentioned problems, and its purpose is to realize technology that can accurately predict abnormalities in storage batteries using data on storage batteries measured during regular inspections of storage battery systems.

In order to solve this problem, for example, the program of the present invention has the following configuration:
A program that causes a computer to function as each means of an information processing device for detecting an abnormality sign related to a plurality of storage batteries that constitute an uninterruptible power supply, the information processing device comprising:
an acquisition means for acquiring a data set for a specific storage battery, the data set including information on measurement results for the specific storage battery and information on a usage status of the specific storage battery, the data being recorded for each periodic inspection performed at a predetermined time interval;
a determination means for determining whether or not there is a sign of abnormality in the specific storage battery by inputting a data set for the specific storage battery into a first learning model related to xgboost that has been trained in advance,
the information regarding the measurement result for the specific storage battery includes a degree of change in an internal resistance value of the specific storage battery for each of the predetermined time intervals, and the information regarding the usage status of the specific storage battery includes the number of days of use from the date of charging;
The determination means is further characterized in that it is capable of determining whether or not there are signs of abnormality in the plurality of storage batteries by inputting the data set for the plurality of storage batteries into a second learning model related to clustering or SVM that has been pre-trained and that determines that a storage battery that is determined to be normal among the plurality of storage batteries and that does not fall within the class to which the information in the data set belongs is a storage battery that is showing signs of abnormality.

According to the present invention, it is possible to accurately predict abnormalities in storage batteries using data on storage batteries measured during regular inspections of the storage battery system.

FIG. 1 is a diagram illustrating an overview of a storage battery abnormality sign detection system according to an embodiment of the present invention. FIG. 1 is a diagram illustrating a configuration example of a storage battery group in a data center according to an embodiment of the present invention. FIG. 1 is a diagram illustrating an example of the configuration of a storage battery in a specific storage battery group according to the present embodiment. A diagram explaining the relationship between changes in the internal resistance value of a storage battery and abnormalities in the storage battery. FIG. 10 is a diagram showing an example of learning data according to the present embodiment; FIG. 10 is a diagram showing another example of learning data according to the embodiment; FIG. 1 is a diagram showing an example of the functional configuration of an information processing apparatus according to an embodiment of the present invention; FIG. 2 is a diagram illustrating an example of the functional configuration of a communication device according to the present embodiment. FIG. 10 is a diagram showing an example of a data set stored in a measurement data DB according to the present embodiment. FIG. 1 is a diagram showing an example of a data set input to a learning model in an estimation stage according to the present embodiment. FIG. 10 is a diagram showing an example of the relationship between the timing of periodic inspection and discharge time according to the present embodiment; A flowchart showing a series of operations in a learning stage by the information processing device according to the present embodiment. A flowchart showing a series of operations in an estimation stage performed by the information processing device according to the present embodiment. FIG. 10 is a diagram showing an example of an abnormality sign notification screen according to the embodiment; FIG. 10 is a diagram schematically illustrating data distribution of a plurality of storage batteries according to the present embodiment.

The following embodiments are described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the claimed invention, and not all combinations of features described in the embodiments are necessarily essential to the invention. Two or more of the multiple features described in the embodiments may be combined in any desired manner. Furthermore, the same reference numbers are used for identical or similar components, and duplicate descriptions will be omitted.

(Overview of the storage battery abnormality detection system)
An overview of the storage battery abnormality sign detection system will be described with reference to FIG. 1 . The storage battery abnormality sign detection system 10 includes an information processing device 100, which is an information processing server, and a communication device 210. The communication device 210 is, for example, a personal computer or a mobile terminal located in a data center 200. The communication device 210 accepts input of measurement results and usage status data related to the storage batteries by a maintenance technician for the storage battery group 230, and transmits the input information related to the measurement results and usage status to the information processing device 100 via a network. The communication device 210 may also accept data input by a maintenance technician for the storage battery group 230 or an administrator of the storage battery abnormality sign detection system 10, and transmit correct answer data (described below) from the learning data to the information processing device 100.

The information processing device 100 receives and records information about measurement results and usage status from each communication device 210 in multiple data centers, and detects signs of abnormalities in the storage battery based on a machine learning model.

The data center 200 operates a server group 220 that executes arbitrary processes (e.g., processes of tenants that use the server resources of the server group). The server group 220 normally operates on power supplied from a power supply (not shown) in the data center 200. The storage battery group 230 includes multiple storage batteries and functions as an uninterruptible power supply, supplying power to the server group 220 in emergencies such as power outages.

The battery group 230 in the data center 200 is composed of multiple battery systems (referred to here as UPS battery systems), as shown in FIG. 2A, for example. Each of the multiple UPS battery systems is distinguished, for example, as "UPS No. 1," "UPS No. 2," "UPS No. 3," and "UPS standby," and can independently provide power to the server group 220. The "UPS standby" system operates, for example, when at least one of "UPS No. 1," "UPS No. 2," and "UPS No. 3" fails or is undergoing maintenance, or when the required power cannot be supplied due to insufficient power supply.

Each UPS battery system includes multiple battery groups 240, such as battery group A and battery group B. These battery groups 240 may be connected in series. For example, as shown in FIG. 2B, the battery group 240 is configured to generate the required voltage by connecting multiple batteries 250 in series. If one battery 250 experiences an abnormality such as a disconnection, the UPS battery system may lose its function as a UPS power supply. Therefore, accurately detecting signs of an abnormality in each battery is an important issue for the system to function as a UPS power supply.

(Detection of signs of abnormality in storage batteries)
Detection of signs of abnormality in a storage battery will be described with reference to FIG. 2C. FIG. 2C shows an example of the transition of the internal resistance values of several individual storage batteries. The horizontal axis of FIG. 2C represents time, and the vertical axis represents the internal resistance value (mΩ). Graph 261 shows the transition of the internal resistance value measured for a specific storage battery at times t-3, ..., t, and t+1. Times t-3, ..., t, and t+1 correspond to the timing of measurements during periodic inspections that were carried out in sequence. In other words, graph 261 is a graph of the measurement results for the storage battery recorded during the periodic inspection. Graphs 262 and 263 also represent the measurement results for individual storage batteries.

A conventional anomaly detection method involves determining that a battery's internal resistance value exceeds a predetermined threshold value 264 is a sign of an abnormality and replacing or inspecting the battery. However, as shown in graph 262, even though the threshold value was not exceeded when inspection timing was t, if the battery's internal resistance value increases rapidly between inspection timing t and t+1, there is a risk that the battery will fail before inspection timing t+1.

In response to this issue, this embodiment detects signs of abnormality in a storage battery by taking into account not only the internal resistance value but also the trend in changes in the internal resistance value, in order to detect storage batteries whose internal resistance will exceed a threshold value in the future. In fact, even if the internal resistance value of a normal storage battery increases over time, the rate of increase decreases. In contrast, the internal resistance value of a storage battery in which an abnormality occurs increases at an accelerated rate (the internal resistance value increases and the rate of increase also increases). This embodiment focuses on the trend in changes in the internal resistance value.

Furthermore, in detecting signs of abnormality according to this embodiment, a machine learning model is trained using data that can take into account the progression of changes in internal resistance value as training data, and the presence or absence of signs of abnormality is determined using the machine learning model. For example, in the example of Figure 2C, data from times t, t-1, and t-2 is input into the (trained) machine learning model, and a determination is made as to whether signs of abnormality are present (exceeding threshold value 264) by time t+1. It is determined that there are no signs of abnormality for the storage batteries corresponding to graphs 261 and 263, and that there are signs of abnormality for the storage battery corresponding to graph 262.

Figure 3 shows an example of learning data (also called teacher data) used in this embodiment. This learning data is stored in a learning data DB (described later) of the information processing device 100. The learning data includes, for example, a storage battery identification ID, information related to the measurement results corresponding to time t-2 to time t, information related to the usage status, and a judgment.

"Judgment" represents the correct answer data (also known as a label). Based on past regular inspections and the occurrence of abnormalities, "NG" (indicating an abnormality) is assigned to storage batteries that exceeded threshold 264 between time t and t+1. On the other hand, "OK" (indicating no abnormality) is assigned to storage batteries that did not exceed threshold 264 between time t and t+1.

The battery identification ID is an ID for identifying each battery. It may be possible to not only simply identify the battery, but also to specify, for example, which data center the battery is located in, which UPS battery system it is located in, and which battery in that UPS battery system it is a part of. In this way, when using learning data, it is possible to train a model using learning data from a specific data center, or to train a model using learning data related to batteries in a specific UPS battery system. If batteries with different properties (i.e., batteries manufactured by different manufacturers or types of batteries with different characteristics) are installed in each data center or each UPS battery, it is possible to train a model for each battery with different properties, and to realize a learning model appropriate for each battery with different properties.

The measurement data corresponding to time t-2 includes the internal resistance value (mΩ) of the target storage battery at time t-2 and the slope of the internal resistance value between time t-1 and time t-2. The slope of the internal resistance value may be, for example, the difference in the internal resistance value between time t-1 and time t-2, or the value obtained by dividing this difference by the number of days between time t-1 and time t-2. Furthermore, these values (such as the value obtained by dividing this difference by the number of days between time t-1 and time t-2) may be multiplied by a predetermined scaling factor.

The measurement data corresponding to time t-1 includes the internal resistance value (mΩ) of the target storage battery at time t-1 and the slope of the internal resistance value between time t and time t-1. The slope of the internal resistance value may be, for example, the difference in the internal resistance value between time t and time t-1, or the value obtained by dividing this difference by the number of days between time t and time t-1. Furthermore, these values may be multiplied by a predetermined scaling factor.

By adding the gradient of the internal resistance value to each of the measurement data from time t-2 and time t-1, the change in the rate of change in the internal resistance value can be explicitly provided to the machine learning model. In other words, unlike when the learning data simply includes the internal resistance value, in this embodiment, the occurrence of an abnormality in the storage battery can be predicted by explicitly taking into account the change in the rate of change in the internal resistance value (i.e., the accelerated increase in the internal resistance value).

Measurement data corresponding to time t includes the internal resistance value (mΩ) of the target storage battery at time t. Data indicating usage status includes information on the usage period, which indicates the number of days elapsed since the storage battery was first charged (number of days of use from the charging date). It has been found that the number of days elapsed since the start of storage battery use correlates with the occurrence of an abnormality.

By using this type of learning data and utilizing the results of past regular inspections, it is possible to predict trends in battery deterioration, enabling necessary battery countermeasures to be implemented before an abnormality occurs. Furthermore, it is possible to predict the occurrence of an abnormality in a battery where the internal resistance value changes suddenly between time t (the current inspection time) and time t+1 (the next inspection time) by explicitly taking into account the changes in the degree of change in the internal resistance value.

In this embodiment, we will explain the case where xgboost, a gradient boosting method based on a decision tree algorithm, is used as the machine learning model. However, the machine learning model may be configured using other machine learning models, for example, a deep neural network. In addition to using a deep neural network with a simple hidden layer configuration, it may also be possible to use a neural network that uses an RNN (recurrent neural network) such as LSTM, which can more appropriately take into account the characteristics of temporal changes in time-series data. It may also be possible to use a deep neural network with an attention mechanism (also called a caution mechanism) that can provide output that focuses on characteristic data in the time-series data.

The learning data may include additional information, as shown in Figure 4. It may also include other measurement values, such as the battery voltage (V) at time t-2, time t-1, and time t. Furthermore, the information about usage status may include connection location information in addition to information about the usage period. The connection location information is the connection location (order from the beginning) of the target battery among a group of series-connected batteries. As described above, a group of batteries includes multiple batteries connected in series. The batteries are arranged, for example, in a two- or three-dimensional configuration within a rack or box. Therefore, when a battery discharges during a power outage or a periodic inspection, temperature changes within the box affect battery degradation. However, the timing of battery measurements during a periodic health check may not be during battery discharge, as shown in Figure 8, for example. For example, battery discharge is expected to occur during very limited inspection periods to check operation or when power is needed due to a power outage. Therefore, the timing of measurement and recording during a periodic inspection may not necessarily be such that temperatures that could affect the battery can be measured. For this reason, in this embodiment, the connection positions of the storage batteries (or the arrangement of storage batteries arranged in two or three dimensions) are used as input parameters for the learning model. Alternatively, instead of or in addition to the connection positions of the storage batteries, the discharge time may be used. The discharge time is the cumulative time during which the storage battery has discharged. This allows the impact of the operation of the storage battery to be reflected in the detection of signs of deterioration.

In this embodiment, an example will be described in which the communication device 210 transmits information related to the measurement results, information related to the usage status, and correct answer data to the information processing device 100, and the information processing device 100 generates the learning data. However, the communication device 210 may also generate the learning data related to this embodiment and transmit the generated learning data to the information processing device 100. In this case, the information processing device 100 simply stores the acquired learning data in the learning data DB 522.

(Configuration of information processing device)
Next, an example of the functional configuration of the information processing device 100 will be described with reference to Fig. 5. Note that each of the functional blocks described with reference to the following figures may be integrated or separated, and the described functions may be realized by different blocks. Also, what is described as hardware may be realized by software, and vice versa.

The communication unit 501 includes a communication circuit or communication module that communicates with the communication device 210 used by the storage battery maintenance personnel via a network. The communication unit 501 communicates with multiple communication devices 210.

The control unit 502 includes a central processing unit (CPU) 510 and a RAM 511. The CPU 510 may be composed of one or more CPUs. The control unit 502 executes a computer program stored in the recording unit 504 to perform the storage battery abnormality sign detection process described below and control the operation of each unit of the information processing device 100. The control unit 502 may further include a computing unit (e.g., a GPU) or dedicated hardware for faster execution of statistical processing such as machine learning.

RAM 511 is a volatile storage medium such as DRAM, and temporarily stores parameters and processing results used by the control unit 502 to execute computer programs.

The power supply unit 503 is a circuit or module that provides power for the operation of each unit of the information processing device 100. The power supply unit 503 may also be configured to include a battery.

The recording unit 504 includes a non-volatile recording medium such as a hard disk or semiconductor memory, and records setting values and calculation results necessary for the operation of the information processing device 100. The recording unit 504 also stores information related to measurement results and usage conditions received from the communication device 210 in the measurement data DB 521. The recording unit 504 also stores the above-mentioned learning data in the learning data DB 522.

The storage battery data receiving unit 512 receives information about measurement results and usage conditions entered by the storage battery maintenance staff from the communication device 210 and records it in the measurement data DB 521 of the recording unit 504. The information about measurement results and usage conditions stored in the measurement data DB 521 is configured, for example, as shown in Figure 7A.

The data shown in FIG. 7A includes a battery identification ID 701, information about measurement results for each hour, information about usage status, and information about usage status that is invariant over time 704. The information about measurement results for each hour and information about usage status includes, for example, information about measurement results measured at time t-n and information about usage status 702, information about measurement results measured at time t-n-1 and information about usage status, and information about measurement results measured at time t and information about usage status 703.

The battery identification ID 701 is the same as that described above in Figures 3 and 4. The internal resistance and voltage values in the information regarding hourly measurement results are the same as those described above in Figure 4. The date is the date on which data such as internal resistance was measured. The usage period and discharge time in the information regarding usage status represent the cumulative number of days up to the corresponding time (e.g., t-n). The information regarding usage status 704 includes connection location information, which is the same as the connection location information shown in Figure 4.

Referring again to FIG. 5 , the learning data generation unit 513 generates the above-mentioned learning data using data on a specific storage battery (e.g., designated by the storage battery maintenance staff or the administrator of the anomaly sign detection system 10) from the data stored in the measurement data DB 521. For example, the learning data generation unit 513 calculates the slope of the internal resistance value at time t-1 from the date and internal resistance value at time t-1 and the date and internal resistance value at time t-1 in the measurement data DB 521. Furthermore, the learning data generation unit 513 generates the usage information shown in FIG. 3 or 4 based on usage information at time t and time-invariant usage information 704. The specific storage battery is a storage battery that has actually experienced an abnormality or has been confirmed to be operating normally. According to the inventor's experiments, anomaly signs can be effectively detected even when the learning data is composed of, for example, data from three regular inspections (time t-2, time t-1, time t). However, this example is not limiting; data from more regular inspections, such as five, may also be used.

The learning data generation unit 513 assigns learning data judgment information (i.e., correct answer data indicating "OK" or "NG") to the data of the storage battery identification ID of the specific storage battery. The learning data generation unit 513 receives correct answer data via the communication device 210 from, for example, a storage battery maintenance technician or an administrator of the storage battery abnormality sign detection system 10, and adds it to the learning data. The learning data generation unit 513 stores the generated learning data in the learning data DB 522.

The measurement data acquisition unit 514 generates information related to measurement data for a specific storage battery and information related to usage status from the measurement data DB 521 in order to execute the abnormality sign detection process in the estimation stage described below. The storage battery processed by the measurement data acquisition unit 514 is the storage battery that is to be judged for the presence or absence of abnormality signs. The measurement data acquisition unit 514 performs processing similar to the processing for generating data other than "judgment" described for the learning data generation unit 513 to generate a dataset to be input into the learning model for abnormality sign detection. As shown in Figure 7B, the dataset generated by the measurement data acquisition unit 514 includes the internal resistance value and the slope of the internal resistance value at time t-2 and time t-1, the internal resistance value at time t, and information on the usage period. Note that the example dataset shown in Figure 7B is in a format corresponding to the learning data shown in Figure 3. Furthermore, when generating a dataset in a format corresponding to the training data shown in Figure 4, the generated dataset includes the internal resistance value, internal resistance value slope, and voltage at time t-2 and time t-1, the internal resistance value and voltage at time t, and information on the usage period, connection position, and discharge time.

The preprocessing unit 515 performs common preprocessing on the training data and the data shown in Figure 7B before inputting the data into the learning model. This preprocessing reduces the impact of variations in characteristics on the learning model and estimation results when storage batteries are manufactured by multiple manufacturers and models and the characteristics of storage batteries vary depending on the manufacturer and model. Preprocessing may include, for example, normalizing each parameter related to the measurement results and the usage information. Normalization may involve scaling each parameter so that it falls within the range of 0 to 1 (e.g., using the maximum value for each manufacturer or model). Alternatively, the preprocessing unit 515 may use values obtained by dividing measurement results, such as voltage values and internal resistance values, by device specification information. For example, the measurement voltage value may be divided by a standard voltage, which varies for each storage battery product, to calculate the ratio to the standard voltage, and the calculated ratio to the standard voltage is input into the learning model. The standard voltage varies depending on the number of series cells in the product. The preprocessing unit 515 may also divide the internal resistance value by a standard internal resistance value, which varies for each product, to calculate a ratio to the standard internal resistance, and use this ratio. The standard internal resistance value varies depending on the number of series cells in the product and the terminal structure. The preprocessing unit 515 may also divide the storage battery's usage period by an average lifespan, which varies for each product, to calculate a ratio to the average lifespan, and use this ratio. Information such as the maximum value used for normalization, as well as the standard voltage, standard internal resistance, and average lifespan for each manufacturer and model, may be pre-recorded in the recording unit 504.

The model processing unit 516 inputs the learning data shown in FIG. 3 or FIG. 4 to perform processing to train an internal machine learning model, and also inputs the input data shown in FIG. 7B using the trained machine learning model to perform storage battery abnormality sign detection processing. The machine learning model used by the model processing unit 516 in this embodiment functions as a classifier that outputs a prediction result as a binary value (OK or NG). As described above, in this embodiment, the learning model uses, for example, xgboost, a gradient boosting method based on a decision tree algorithm. However, the machine learning model may also be configured using, for example, a deep neural network. The abnormality information providing unit 517 transmits the results of the abnormality sign detection processing performed by the model processing unit 516 in the estimation stage to the communication device 210. The abnormality information providing unit 517 transmits the results of the abnormality sign detection processing to the communication device 210, for example, as information on an abnormality sign notification screen, which will be described later with reference to FIG. 10.

(Configuration of communication device)
Next, an example of the functional configuration of the communication device 210 will be described with reference to FIG. 6 . In this embodiment, a personal computer will be used as an example of the communication device, but the communication device may be other electronic devices such as a smartphone or a tablet terminal. Note that each of the functional blocks described with reference to the following figures may be integrated or separated, and the described functions may be realized by different blocks. Furthermore, what is described as hardware may be realized by software, and vice versa.

The communication unit 601 includes communication circuits and the like, and communicates with the information processing device 100, for example, by connecting to the Internet via wireless communication such as a wireless LAN, or by connecting to a network via wireless communication for mobile phones.

The control unit 602 includes a CPU 610 and RAM 611, and controls the operation of each unit within the communication device 210, for example, by the CPU 610 executing a computer program recorded in the recording unit 607.

The operation unit 603 includes at least one of a keyboard, mouse, or touch panel, and can perform operations on various operation GUIs displayed on the display unit 606 (for example, inputting information related to measurement results and usage conditions, and providing correct answer data). The power supply unit 604 provides power to each unit of the communication device 210. The imaging device 305 is, for example, a camera mechanism including an imaging element, and takes images in response to instructions from the control unit 602.

The display unit 606 includes a display device such as an LCD or OLED. In response to instructions from the control unit 602, the display unit 606 displays a GUI for inputting information about measurement results and usage conditions, for providing correct data, a GUI for an abnormality warning screen, etc.

The recording unit 607 includes non-volatile memory such as semiconductor memory, and stores programs and setting values executed by the control unit 602. Computer programs stored in the recording unit 607 include an operating system and various applications for implementing the various functions of the communication device 210.

(A series of operations for abnormality sign detection processing in the learning stage)
Next, a series of operations of the abnormality sign detection process in the learning stage executed in the information processing device 100 will be described with reference to FIG. 9A . This process is realized by the CPU 510 of the control unit 502 executing a computer program recorded in the recording unit 504. In the following description, for ease of explanation, the processing entity of each step will be collectively described as the control unit 502, but each unit functioning within the control unit 502 executes the corresponding process depending on the processing content. This process is a process for training a machine learning model in the model processing unit 516, and the learning of the machine learning model is not yet complete.

In S401, the control unit 502 receives information about the measurement results and information about the usage status from the communication device 210. As described above, the control unit 502 stores the information about the measurement results and information about the usage status in the measurement data DB 521 in the format shown in FIG. 7A.

In S402, the control unit 502 acquires information necessary for generating training data from the communication device 210. The information necessary for generating training data is the correct answer data shown in FIG. 3 or FIG. 4.

In S404, the control unit 502 generates the learning data shown in FIG. 3 or 4 based on the data shown in FIG. 7A (information related to the measurement results and information related to the usage status) stored in the measurement data DB 521 and the correct answer data acquired in S402. The learning data is generated as described above for the learning data generation unit 513.

In S405, the control unit 502 performs learning on the learning model. The control unit 502 trains the machine learning model using the generated training data. Since well-known learning methods can be used for learning the machine learning model, such as decision trees and deep neural networks, detailed explanations will be omitted. When the control unit 502 finishes learning the machine learning model using the training data, it ends the series of operations for the anomaly sign detection process in the learning stage.

(Series of operations in the abnormality sign detection process at the estimation stage)
Next, a series of operations of the abnormality sign detection process in the estimation stage executed by the information processing device 100 will be described with reference to FIG. 9B . This process is realized by the CPU 510 of the control unit 502 executing a computer program stored in the recording unit 504. In the following description, for ease of explanation, the processing entity of each step will be collectively described as the control unit 502, but each unit functioning within the control unit 502 executes the corresponding process depending on the processing content. This process is a process for determining the presence or absence of an abnormality sign from newly input storage battery data using a trained machine learning model in the model processing unit 516. While the process shown in FIG. 9B describes an example in which processing is performed on one specific storage battery, abnormality sign detection may be performed on multiple storage batteries by repeating the processes of S921 and S922.

In S921, the control unit 502 generates the data shown in FIG. 7B from the data shown in FIG. 7A stored in the measurement data DB 521 for the specific storage battery that is the target of the abnormality sign detection process. After generating the data shown in FIG. 7B, the control unit 502 may apply the preprocessing by the preprocessing unit 515 described above to the generated data.

In S922, the control unit 502 executes anomaly sign detection processing. The control unit 502 uses the trained machine learning model to determine whether or not there are signs of anomaly for the specific storage battery being processed. The control unit 502 inputs the data generated in S921 into the trained machine learning model and obtains output from the machine learning model (i.e., a binary value of OK or NG).

In S923, the control unit 502 transmits information about the storage battery in which a sign of abnormality was detected (e.g., the storage battery identification ID and the progress of the degree of change in the internal resistance value) to the communication device based on the determination result obtained from the machine learning model. After transmitting the storage battery information to the communication device 210, the control unit 502 terminates this series of processes.

The processes of S921 and S922 may be repeatedly performed for multiple storage batteries, and abnormality sign information for multiple storage batteries (also referred to as an abnormality sign notification screen) as shown in FIG. 10 may be transmitted to the communication device 210. The abnormality sign notification screen 1000 shown in FIG. 10 displays the storage battery IDs of one or more storage batteries for which an abnormality sign has been detected, and storage battery information 1002 corresponding to the storage battery IDs. The storage battery information 1002 may display at least one of the characteristics of the storage battery indicating the abnormality sign, such as the change in the slope of the internal resistance value 1003, the usage period 1004, and the connection position 1005. In this way, it is possible to simultaneously identify storage batteries that are estimated to have an abnormality sign at the current time (time t), obtain an overview of the characteristics of each storage battery, and easily confirm the validity of the estimation results.

As described above, in this embodiment, a data set for a specific storage battery, including information about the measurement results for the specific storage battery and information about its usage status, is acquired and input into a trained machine learning model to determine whether or not there are signs of an abnormality in the specific storage battery. The information about the measurement results for the specific storage battery includes the degree of change in the internal resistance value of the specific storage battery at each predetermined time interval, and the machine learning model is explicitly provided with the trend in the degree of change in the internally provided value. This makes it possible to accurately predict signs of an abnormality in the storage battery using data on the storage battery measured during regular inspections of the storage battery system. According to this embodiment, even with a small amount of data from, say, three regular inspections, providing the trend in the degree of change in the internally provided value allows for accurate prediction of signs of an abnormality in the storage battery, thereby enabling the storage battery system to have a simple configuration (one that does not require sensors that constantly monitor the characteristics of the storage battery).

In the above-described embodiment, the internal resistance value, the slope of the internal resistance value, and the voltage were given as examples of information related to the measurement results of the storage battery, but the information related to the measurement results may also include other information. For example, it may include information such as at least one of the temperature of the storage battery obtained during periodic inspection, the number of discharges, the discharge time, and the discharge capacity.

Furthermore, in the above-described embodiment, the abnormality sign detection process determines whether or not there is an abnormality sign for each individual storage battery. However, in addition to the abnormality sign determination for each individual storage battery described above, the model processing unit 516 may use machine learning algorithms such as clustering or SVM (support vector machine) to determine that a storage battery that does not fall into the class of batteries determined to be normal among multiple storage batteries is a storage battery showing an abnormality sign.

For example, as shown in Figure 11, if we consider a space in which information related to the measurement results of each storage battery and information related to usage status are vector components, data for multiple storage batteries will be distributed at various positions within the space (e.g., 1103 and 1104). By applying machine learning algorithms such as clustering and SVM to the data for these storage batteries, storage batteries 1103 that belong to class 1102 of normal storage batteries can be distinguished from other storage batteries 1104. The distinguished storage batteries 1104 are then determined to be storage batteries with signs of abnormality.

The control unit 502 may execute multiple machine learning algorithms using the model processing unit 516, and if a storage battery in which an abnormality sign was detected by the processing of S922 matches a storage battery 1104 that does not belong to the class of normal storage batteries, ultimately detect the storage battery as a storage battery with an abnormality sign. In this way, maintenance can be performed on storage batteries that are determined to have a high probability of having an abnormality sign by multiple machine learning algorithms. On the other hand, the control unit 502 may execute multiple machine learning algorithms using the model processing unit 516, and if any one of the machine learning algorithms determines that there is an abnormality sign, ultimately detect the storage battery as a storage battery with an abnormality sign. In this case, maintenance can be performed widely on storage batteries that may have an abnormality sign.

The invention is not limited to the above-described embodiments, and various modifications and variations are possible within the scope of the invention.

100... Information processing device, 210... Communication device, 513... Learning data generation unit, 514... Measurement data acquisition unit, 515... Preprocessing unit, 516... Model processing unit

Claims (8)

A program that causes a computer to function as each means of an information processing device for detecting an abnormality sign related to a plurality of storage batteries that constitute an uninterruptible power supply, the information processing device comprising:
an acquisition means for acquiring a data set for a specific storage battery, the data set including information on measurement results for the specific storage battery and information on a usage status of the specific storage battery, the data being recorded for each periodic inspection performed at a predetermined time interval;
a determination means for determining whether or not there is a sign of abnormality in the specific storage battery by inputting a data set for the specific storage battery into a first learning model related to xgboost that has been trained in advance,
the information regarding the measurement result for the specific storage battery includes a degree of change in an internal resistance value of the specific storage battery for each of the predetermined time intervals, and the information regarding the usage status of the specific storage battery includes the number of days of use from the date of charging;
The program is characterized in that the determination means is further capable of determining whether or not the plurality of storage batteries have signs of abnormality by inputting the data sets for the plurality of storage batteries into a second learning model related to clustering or SVM that has been pre-trained and that determines that a storage battery that is determined to be normal among the plurality of storage batteries and that does not fall within the class to which the information in the data set belongs is a storage battery that has signs of abnormality. The program described in claim 1, characterized in that the information regarding the usage status of the specific storage battery further includes the connection position of the specific storage battery among the multiple storage batteries connected in series. The program described in claim 1 or 2, characterized in that the information regarding the usage status of the specific storage battery further includes the arrangement of the specific storage battery among the multiple storage batteries arranged in a two-dimensional or three-dimensional configuration. The program described in any one of claims 1 to 3, characterized in that the information regarding the measurement results for the specific storage battery further includes at least one of the number of times discharged, the discharge time, or the discharge capacity. The program described in claim 1 further comprises a detection means for detecting at least one of the specific storage battery determined to have an abnormality sign by the first learning model and the storage battery determined to have an abnormality sign by the second learning model as a storage battery with an abnormality sign. The program described in any one of claims 1 to 5, characterized in that the information regarding the measurement results is based on measurement results measured when the multiple storage batteries are not discharging. the plurality of storage batteries include storage batteries from different manufacturers,
7. The program according to claim 1, wherein the information about the measurement results uses values that are normalized in advance for different storage batteries manufactured by different manufacturers. An information processing method executed by an information processing device for detecting an abnormality sign related to a plurality of storage batteries constituting an uninterruptible power supply, comprising:
an acquisition step of acquiring a data set for a specific storage battery, the data set including information on measurement results for the specific storage battery and information on a usage status of the specific storage battery, the data being recorded for each periodic inspection performed at predetermined time intervals;
a determination step of determining whether or not there is a sign of abnormality in the specific storage battery by inputting a dataset for the specific storage battery into a first learning model related to xgboost that has been trained in advance,
the information regarding the measurement result for the specific storage battery includes a degree of change in an internal resistance value of the specific storage battery for each of the predetermined time intervals, and the information regarding the usage status of the specific storage battery includes the number of days of use from the date of charging;
The information processing method is characterized in that the judgment step further comprises inputting the data set for the plurality of storage batteries into a second learning model related to clustering or SVM that has been pre-trained and that judges a storage battery that is determined to be normal among the plurality of storage batteries and that falls outside the class to which the information in the data set belongs as a storage battery that is showing signs of abnormality, thereby making it possible to determine whether or not the plurality of storage batteries have signs of abnormality.