Abnormal detection device, abnormal detection method, and computer program

Abstract

To provide an anomaly detection device, an anomaly detection method, and a computer program.SOLUTION: An anomaly detection device includes a creation unit that creates learning data by statistically processing a plurality of pieces of measurement data of a power storage element that may include abnormal measurement data, a storage unit that stores a model that is trained to output a score corresponding to whether the measurement data contains abnormal measurement data when the measurement data is input, using the created learning data, and a detection unit that detects an abnormality or a sign of an abnormality in the power storage element on the basis of a score that is output by inputting the plurality of pieces of measurement data into the model.SELECTED DRAWING: Figure 11

Description

Technical Field

[0001] The present invention relates to an abnormality detection device, an abnormality detection method, and a computer program that detect an abnormality based on measurement data of a power storage element.

Background Art

[0002] Power storage elements are widely used in uninterruptible power supply devices, DC or AC power supply devices included in stabilized power supplies, etc. Further, the use of power storage elements in large-scale systems that store electric power generated by renewable energy or existing power generation systems is expanding.

[0003] In a system using a power storage element, it is necessary to detect the state of the power storage element. Patent Document 1 discloses the use of a model for determining the safety or abnormality of a power storage element. In Patent Document 1, data determined to be normal is acquired in advance, and the model is created by machine learning such as deep learning based on the acquired data.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] A model for abnormality detection is machine-learned using learning data in which data of normal products and data of products that are not normal (abnormal products) are separated in advance. However, it is not easy to prepare learning data including classification as to whether data of a power storage element is data of a normal product.

[0006] An object of the present invention is to provide an abnormality detection device, an abnormality detection method, and a computer program that detect an abnormality or a sign thereof based on measurement data of a power storage element.

Means for Solving the Problem

[0007] The abnormality detection device includes a creation unit that creates learning data by statistically processing a plurality of measurement data that may include abnormal measurement data of a power storage element, and a storage unit that stores a model that is learned to output a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input, using the created learning data. The abnormality detection device further includes a detection unit that detects an abnormality or a sign of an abnormality of the power storage element based on the score output by inputting the plurality of measurement data into the model.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Embodiments for Carrying Out the Invention

[0009] The abnormality detection device includes a creation unit that statistically processes a plurality of measurement data that may include abnormal measurement data of a power storage element to create learning data, and a storage unit that stores a model that is learned to output a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input, using the created learning data, and a detection unit that detects an abnormality or a sign of an abnormality of the power storage element based on the score output by inputting the plurality of measurement data into the model. Here, the "plurality of measurement data that may include abnormal measurement data" means a plurality of measurement data in which measurement data that should be determined as abnormal or heterogeneous is not completely excluded artificially or mechanically. The "plurality of measurement data that may include abnormal measurement data" includes, in that sense, a plurality of measurement data in which measurement data that should be determined as abnormal or heterogeneous is not excluded at all artificially or mechanically. The "plurality of measurement data that may include abnormal measurement data" also includes, in that sense, a plurality of measurement data in which a part of the measurement data that should be determined as abnormal or heterogeneous (for example, extreme outliers) is excluded artificially or mechanically. The "plurality of measurement data that may include abnormal measurement data" also includes, in that sense, measurement data in which the power storage element is new or in a good state and does not actually include abnormal measurement data (measurement data that has not been subjected to a process of artificially or mechanically excluding abnormal measurement data). The score may be a numerical value or classification output from a model trained without a teacher. The score may be, for example, a reconstruction error obtained from an autoencoder. Alternatively, the score may be a numerical value or classification output from a model trained with a teacher. It tends to be difficult to prepare measurement data of other systems operated under the same conditions as the power storage system actually operated, or to prepare appropriate learning data by virtual methods such as simulation. Therefore, it is preferable to adopt unsupervised learning that can analyze the characteristics of the measurement data of the power storage system actually operated.

[0010] With the above configuration, in order to prepare learning data from measurement data obtained during operation, there is no need to separate data that should be judged as normal and data that should be judged as abnormal (eliminating the labor for data selection). The preparation work for learning data is simplified, and it becomes possible to automate part or all of the preparation work. Measurement data indicating the state of the energy storage element (or indirectly indicating the state of the system surrounding the energy storage element) may change in characteristics depending on the aging deterioration of the energy storage element and the usage environment. Even when operating with the same charge-discharge pattern, the current measurement data of the energy storage element is different from the measurement data several months or several years later. Depending on the usage period and usage environment, the energy storage element deteriorates, and the measurement data inevitably changes little by little. Among them, it is highly difficult to distinguish whether the obtained measurement data is abnormal data using a mathematical model or a threshold value. A very complicated operation is required to accurately distinguish abnormal / normal and prepare learning data. On the contrary, by "statistically processing a plurality of measurement data that may include abnormal measurement data of the energy storage element to create learning data" as in the above configuration, complicated operations can be eliminated or simplified.

[0011] In the detection of abnormalities in measurement data obtained after the start of operation using a model learned with measurement data obtained before the start of operation or in the initial stage of operation of the energy storage element, there is a possibility of erroneously detecting measurement data that is not abnormal as an abnormality or its sign. For example, if a model is learned using measurement data obtained in the initial stage of operation as data of normal products, the model will detect changes in the characteristics of the energy storage element due to mere aging changes in the characteristics of the energy storage element or changes in the operating environment (seasonal changes or changes in the degree of charge and discharge) as an abnormality or its sign. This is called deterioration diagnosis, not abnormality detection.

[0012] In the abnormality detection device with the above configuration, the measurement data used for learning the model is the measurement data that is the target of abnormality detection. According to the above configuration, it is not affected (or the influence is small) by the difference in the period or operating environment between the time of learning the model and the time of abnormality detection using the model. When the model is trained using data of normal products including simply abnormal measurement data, the trained model cannot detect abnormal measurement data as an abnormality or its sign during detection. The inventors have found that, as in the above configuration, by statistically processing a plurality of measurement data that may include abnormal measurement data, appropriate training data can be easily prepared and model training can be executed. In the abnormality detection device with the above configuration, additional training of the model and reconstruction of the model can be realized relatively easily.

[0013] The training data used for training the model in the abnormality detection device may be created using the average of a plurality of measurement data that may include abnormal measurement data of the energy storage element.

[0014] The inventors have found that by using the average of a plurality of measurement data that may include abnormal measurement data of the energy storage element, pseudo-normal data (training data) can be obtained. In an actual energy storage system, the occurrence of abnormalities in the energy storage element and system failures is extremely rare. The inventors have found that a small number of abnormal data included in a large number of measurement data are appropriately rounded by the average and do not have a negative impact on the training of the model for detecting abnormalities in the energy storage element. Rather, the inventors have found that appropriate training data can be prepared from data in which normal and abnormal (or heterogeneous) data are mixed. The training data thus obtained is suitably applied, for example, to the training of an autoencoder.

[0015] The energy storage element may be configured by connecting a plurality of modules each including a plurality of energy storage cells in series. The creation unit may create the training data by averaging the measurement data of the energy storage cells of the same rank in the plurality of modules. The energy storage element may have a configuration (also referred to as a domain) in which a plurality of configurations (also referred to as banks) each connecting a plurality of modules including a plurality of energy storage cells in series are connected in parallel. The creation unit may create the training data by averaging the measurement data of the energy storage cells of the same rank in the plurality of modules included in the domain.

[0016] By using the method for calculating the average considering the configuration of the energy storage element as described above, appropriate training data can be created.

[0017] In the anomaly detection device, the creation unit may create the training data based on the measurement data read out for the read target period among the measurement data measured in time series from the energy storage element. The detection unit may input the measurement data for the detection target period, which is the same period as the read target period, into the model trained with the training data, and detect an anomaly or a sign of an anomaly in the energy storage element for the detection target period based on the score output from the model.

[0018] With the above configuration, by sequentially reconstructing the model, the influence due to differences in the period or environment between the model training time and the anomaly detection time using the model can be eliminated.

[0019] In the anomaly detection device, the creation unit may create the training data based on the measurement data read out for the read target period among the measurement data measured in time series from the energy storage element. The detection unit may input the measurement data for the detection target period, which partially overlaps with the read target period, into the model trained with the training data, and detect an anomaly or a sign of an anomaly in the energy storage element for the detection target period based on the score output from the model.

[0020] When the variation in the measurement data is small, it is not always necessary to make the training period and the detection period the same, and anomaly detection may be performed using a model trained with measurement data from a slightly earlier time. When it is not possible to sufficiently acquire measurement data, such as when the energy storage system is stopped, anomaly detection is still possible using a model trained with measurement data from a slightly earlier time.

[0021] The abnormality detection method statistically processes a plurality of measurement data that may include abnormal measurement data of a power storage element to create learning data, uses the created learning data, and outputs a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input. The model is learned, the learned model is stored, and based on the score output by inputting the plurality of measurement data into the model, an abnormality or a sign of abnormality of the power storage element is detected.

[0022] The abnormality detection method may be implemented using a computer installed in the vicinity of the power storage element, or may be implemented using a computer installed remotely.

[0023] The computer program statistically processes a plurality of measurement data that may include abnormal measurement data of a power storage element to create learning data, uses the created learning data, and outputs a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input. The model is learned, the learned model is stored, and based on the score output by inputting the plurality of measurement data into the model, the computer program executes a process of detecting an abnormality or a sign of abnormality of the power storage element.

[0024] The computer program may be executed by a computer installed in the vicinity of the power storage element, or may be executed by a computer installed remotely.

[0025] The present invention will be specifically described with reference to the drawings showing its embodiments.

[0026] FIG. 1 is a diagram showing an overview of a remote monitoring system 100. The remote monitoring system 100 enables remote access to information regarding power storage elements and power supply related devices included in a megasolar power generation system S, a thermal power generation system F, and a wind power generation system W. A rectifier (DC power supply device or AC power supply device) D disposed in an uninterruptible power supply (UPS) U, a stabilized power supply system for railways, etc. may be remotely monitored.

[0027] In the megasolar power generation system S, the thermal power generation system F, and the wind power generation system W, a power conditioner (PCS: Power Conditioning System) P and an energy storage system (ESS: Energy Storage System) 101 are arranged in parallel. The energy storage system 101 may be configured by arranging a plurality of containers C accommodating energy storage module groups L in parallel. Alternatively, the energy storage module group L and the power conditioner P may be arranged in a building (energy storage room). The energy storage module group L includes a plurality of energy storage elements. The energy storage elements are preferably rechargeable ones such as secondary batteries like lead-acid batteries and lithium-ion batteries, and capacitors. A part of the energy storage elements may be non-rechargeable primary batteries.

[0028] In the remote monitoring system 100, a communication device 1 is mounted / connected to each of the energy storage systems 101 in the systems S, F, W to be monitored, or the devices (P, U, D and the management device M described later). The remote monitoring system 100 includes the communication device 1, a server device 2 (abnormality detection device) that collects information from the communication device 1, a client device 3 for viewing the collected information, and a network N that is a communication medium between the devices.

[0029] The communication device 1 may be a terminal device (measurement monitor) that communicates with a battery management device (BMU) provided in the energy storage element to receive information on the energy storage element, or may be an ECHONET / ECHONETLite (registered trademark)-compatible controller. The communication device 1 may be an independent device, or may be a network card-type device that can be mounted on the power conditioner P or the energy storage module group L. The communication device 1 is provided one by one for each group composed of a plurality of energy storage modules in order to acquire information on the energy storage module group L in the energy storage system 101. A plurality of power conditioners P are connected in serial communication, and the communication device 1 is connected to the control unit of one of the representative power conditioners P.

[0030] The server device 2 includes a Web server function and presents information obtained from the communication devices 1 mounted / connected to each device to be monitored in response to access from the client device 3.

[0031] The network N includes a public communication network N1 which is the so-called Internet and a carrier network N2 that realizes wireless communication according to a predetermined mobile communication standard. The public communication network N1 includes a general optical line, and the network N includes a dedicated line to which the server device 2 is connected. The network N may include a network compatible with ECHONET / ECHONETLite. The carrier network N2 includes a base station BS, and the client device 3 can communicate with the server device 2 via the network N from the base station BS. An access point AP is connected to the public communication network N1, and the client device 3 can transmit and receive information with the server device 2 via the network N from the access point AP.

[0032] The battery module group L of the power storage system 101 has a hierarchical structure. The communication device 1 that transmits information on the battery elements to the server device 2 acquires information on the battery module group from the management device M provided in the battery module group L. FIG. 2 is a diagram showing an example of the hierarchical structure of the battery module group L and the connection form of the communication device 1. The battery module group L is configured in a hierarchical structure including, for example, battery modules (also referred to as modules) in which a plurality of battery cells (also referred to as cells) are connected in series, banks in which a plurality of battery modules are connected in series, and domains in which a plurality of banks are connected in parallel. In the example of FIG. 2, a management device M is provided for each of the banks numbered (#) 1 to N and for the domain to which the banks are connected in parallel. The management device M provided for each bank communicates with a control board with a communication function (CMU: Cell Management Unit) built in each battery module by serial communication, and acquires measurement data (current, voltage, temperature) for the battery cells inside the battery module. The management device M of the bank executes management processes such as detection of abnormalities in the communication state. The management device M of each bank transmits the measurement data obtained from the battery modules of each bank to the management device M provided in the domain. The management device M of the domain aggregates the measurement data obtained from the management device M of the banks belonging to that domain, information such as detected abnormalities, etc. In the example of FIG. 2, the communication device 1 is connected to the management device M of the domain. Alternatively, the communication device 1 may be connected to each of the management device M of the domain and the management device M of the bank. The management device M can acquire the identification data (identification number) of the domain or bank of the device to which it is connected.

[0033] In one example, the hierarchical structure of the power storage system 101 is composed of 12 power storage modules each formed by connecting 12 power storage cells in series, and includes 12 banks each formed by connecting 12 of these modules in series (domain). In one example, the power storage system 101 may include two domains, and in this case, the power storage system 101 includes 3,456 power storage cells. As another example, the power storage system 101 has a hierarchical structure including a plurality of banks each formed by connecting 18 power storage modules each formed by connecting 16 power storage cells in series. The hierarchical structure of the power storage system 101 is not limited to these examples. Instead of the configuration in which a plurality of banks are connected in parallel as shown in FIG. 2, the power storage system 101 may be composed of a single bank.

[0034] In the remote monitoring system 100, in a large-scale ESS as described above, the server device (abnormality detection device) 2 uses the communication devices 1 installed in each device to collect data such as the SOC (State Of Charge) and SOH (State Of Health) in the power storage system 101. The server device 2 processes the collected data to detect the state of the power storage system 101, and presents it to the user via the client device 3.

[0035] FIGS. 3 and 4 are block diagrams showing the internal configuration of the devices included in the remote monitoring system 100. As shown in FIG. 3, the communication device 1 includes a control unit 10, a storage unit 11, a first communication unit 12, and a second communication unit 13. The control unit 10 is a processor using a CPU (Central Processing Unit), and uses memories such as a built-in ROM (Read Only Memory) and RAM (Random Access Memory) to control each component and execute processing.

[0036] The storage unit 11 uses a non-volatile memory such as a flash memory. The storage unit 11 stores a device program that is read and executed by the control unit 10. The device program 1P includes a communication program conforming to SSH (Secure Shell), SNMP (Simple Network Management Protocol), etc. The storage unit 11 stores information collected by the processing of the control unit 10, information such as event logs. The information stored in the storage unit 11 can also be read out via a communication interface such as USB whose terminals are exposed on the housing of the communication device 1.

[0037] The first communication unit 12 is a communication interface that realizes communication with the monitoring target device to which the communication device 1 is connected. The first communication unit 12 uses, for example, a serial communication interface such as RS-232C or RS-485. For example, the power conditioner P is equipped with a control unit having a serial communication function conforming to RS-485, and the first communication unit 12 communicates with the control unit. When the control boards provided in the power storage module group L are connected by a CAN (Controller Area Network) bus and communication between the control boards is realized by CAN communication, the first communication unit 12 is a communication interface based on the CAN protocol. The first communication unit 12 may be a communication interface corresponding to the ECHONET / ECHONETLite standard.

[0038] The second communication unit 13 is an interface that realizes communication via the network N, and uses, for example, a communication interface such as Ethernet (registered trademark) or a wireless communication antenna. The control unit 10 can communicate and connect with the server device 2 via the second communication unit 13. The second communication unit 13 may be a communication interface corresponding to the ECHONET / ECHONETLite standard.

[0039] In the communication device 1 configured as described above, the control unit 10 acquires measurement data for the power storage element obtained by the device to which the communication device 1 is connected via the first communication unit 12. By reading and executing the SNMP program, the control unit 10 can function as an SNMP agent and respond to information requests from the server device 2.

[0040] The client device 3 is a computer used by an operator such as an administrator or a maintenance person of the power storage system 101 of the power generation systems S, F, and W. The client device 3 may be a desktop or laptop personal computer, or a so-called smartphone or tablet communication terminal. The client device 3 includes a control unit 30, a storage unit 31, a communication unit 32, a display unit 33, and an operation unit 34.

[0041] The control unit 30 is a processor using a CPU. Based on the client program 3P including a web browser stored in the storage unit 31, the control unit 30 causes the display unit 33 to display a web page provided by the server device 2 or the communication device 1.

[0042] The storage unit 31 uses a non-volatile memory such as a hard disk or a flash memory. Various programs including the client program 3P are stored in the storage unit 31. The client program 3P may be a copy obtained by reading the client program 6P stored in the recording medium 6 and copying it to the storage unit 31.

[0043] The communication unit 32 uses a communication device such as a network card for wired communication, a wireless communication device for mobile communication connected to the base station BS (see FIG. 1), or a wireless communication device corresponding to connection to the access point AP. The control unit 30 can communicate and connect or transmit and receive information with the server device 2 or the communication device 1 via the network N by the communication unit 32.

[0044] The display unit 33 uses a display such as a liquid crystal display or an organic EL (Electro Luminescence) display. The display unit 33 displays an image of a web page provided by the server device 2 or the communication device 1 through processing based on the client program 3P of the control unit 30. The display unit 33 is preferably a display with a built-in touch panel, but may also be a display without a built-in touch panel.

[0045] The operation unit 34 is a user interface such as a keyboard and a pointing device or a voice input unit capable of input and output with the control unit 30. The operation unit 34 may use the touch panel of the display unit 33 or physical buttons provided on the housing. The operation unit 34 notifies the control unit 30 of operation information by the user.

[0046] As shown in FIG. 4, the server device (abnormality detection device) 2 uses a server computer and includes a processing unit 20, a storage unit 21, and a communication unit 22. In the present embodiment, the server device 2 is described as a single server computer, but processing may be distributed among a plurality of server computers.

[0047] The processing unit 20 is a processor using a CPU or a GPU (Graphics Processing Unit), and uses a memory such as a built-in ROM and RAM to control each component and execute processing. The processing unit 20 executes communication and information processing based on the server program 21P stored in the storage unit 21. The server program 21P includes a web server program, and the processing unit 20 functions as a web server that provides a web page to the client device 3. The processing unit 20 collects information from the communication device 1 as an SNMP server based on the server program 21P. The processing unit 20 executes an abnormality detection process based on the measurement data collected based on the abnormality detection program 22P stored in the storage unit 21.

[0048] The storage unit 21 uses a non-volatile memory such as a hard disk or a flash memory. The storage unit 21 stores the above-described server program 21P and the abnormality detection program 22P. The storage unit 21 stores a model 2M used in the process based on the abnormality detection program 22P. The storage unit 21 stores the measurement data of the power conditioner P and the power storage module group L of the power storage system 101 to be monitored, which is collected by the processing of the processing unit 20.

[0049] The server program 21P, the abnormality detection program 22P, and the model 2M stored in the storage unit 21 may be obtained by reading the server program 51P, the abnormality detection program 52P, and the model 5M stored in the recording medium 5 and then replicating them to the storage unit 21.

[0050] The communication unit 22 is a communication device that realizes communication connection via the network N and transmission and reception of information. Specifically, the communication unit 22 is a network card corresponding to the network N.

[0051] In the remote monitoring system 100 configured as described above, the communication device 1 transmits the measurement data of each power storage cell acquired from the management device M after the previous timing to the server device 2 each time at a predetermined timing. The predetermined timing may be, for example, a fixed cycle or when the data volume satisfies a predetermined condition. The communication device 1 may transmit all the measurement data obtained via the management device M, may transmit the measurement data thinned out at a predetermined ratio, or may transmit the average value of the measurement data. The server device 2 acquires information including the measurement data from the communication device 1 and stores the acquired measurement data in the storage unit 21 in association with the acquisition time information and the information for identifying the device (M, P) where the information is acquired.

[0052] The server device 2 can present the latest data of the energy storage system 101 stored therein in response to an access from the client device 3. The server device 2 can present the states of each energy storage cell, each energy storage module, bank, or domain. The server device 2 can use the measurement data to perform abnormality diagnosis, degradation diagnosis, estimation of SOC, SOH, etc., or life prediction of the energy storage system 101, and present the implementation results.

[0053] Based on the abnormality detection program 22P and the model 2M shown in FIG. 4, the server device 2 determines, for each energy storage cell, whether the energy storage cell is abnormal or has a sign of abnormality from the measurement data of the energy storage cell. The server device 2 performs state detection for each energy storage module, bank, or domain based on the determination result.

[0054] FIG. 5 is a flowchart showing an example of the processing procedure for model creation and storage by the server device 2. The processing unit 20 of the server device 2 periodically executes the following processing procedure for each target energy storage element. The execution period is longer than the period in which the measurement data is transmitted from the communication device 1. The processing procedure shown in FIG. 5 corresponds to the "creation unit" and the "storage unit".

[0055] The processing unit 20 of the server device 2 reads out the measurement data stored in the storage unit 21 in association with the time information for each energy storage cell for the read target period (step S101).

[0056] The measurement data is, for example, voltage values measured in time series. Alternatively, the measurement data may be the voltage values at each time point obtained by smoothing by taking the moving average of the time series voltage values. The measurement data may be a graph of the time transition of the voltage value. The measurement data may be a set of voltage value and temperature, a set of voltage value, current value, and temperature. The measurement data may be each of the voltage value, current value, and temperature, and the model 2M may be created for each of these data types. The measurement data may be a value calculated using two or three of the voltage value, current value, and temperature. The measurement data may be, for example, the SOC value acquired from the management device M (see FIG. 2).

[0057] The read target period in step S101 is, for example, the period from the arrival timing of the previous execution cycle to the arrival timing of the current execution cycle. The read target period is determined for each power storage system 101 in an arbitrary unit such as one day, one week, two weeks, one month, etc.

[0058] The processing unit 20 groups the read measurement data (step S102), and creates learning data by calculating the average for each group of the measurement data (step S103).

[0059] In step S103, the processing unit 20 groups the measurement data based on the configuration (hierarchical structure) of the power storage system 101. For example, among the power storage cells connected in series included in different banks of power storage modules, the processing unit 20 groups the power storage cells with the same connection order into the same group. The processing unit 20 may group the measurement data within a bank existing in the same environment (location, building, room, shelf, etc.).

[0060] In step S103, the processing unit 20 may create learning data by other statistical processes instead of the average. The statistical process may be calculation of the mode value or calculation of the median value.

[0061] The processing unit 20 creates a model 2M for the measurement data in the detection target period using the created learning data (step S104). The model 2M is learned to output a score corresponding to the possibility (also referred to as abnormality degree or heterogeneity degree) that the measurement data of a power storage cell not homogeneous with the learning data is included in the input measurement data (see FIG. 6).

[0062] In step S104, the processing unit 20 learns the learning data (average of the measurement data) created in step S103 as the measurement data of a normal power storage element (pseudo-normal data).

[0063] In the first example, the detection target period in step S104 is the period during which measurement data was obtained, that is, the period that coincides with the reading target period (see Fig. 6A). In the first example, it is determined whether the learning data, which is the average of the measurement data, and the individual measurement data are homogeneous. In the second example, the detection target period is the reading target period of the measurement data and the period after that period (see Fig. 6B). For example, the processing unit 20 may determine whether the measurement data measured in a two-week period that is one week after and overlaps with the two-week period from which the learning data created from the measurement data for a certain two weeks was learned by the model 2M learned by the learning data is homogeneous with the learning data.

[0064] The processing unit 20 stores the model 2M created in step S104 in the storage unit 21 in association with the identification data (step S105), and ends the creation process and storage process of the model 2M. The identification data in step S105 may be a numerical value indicating the reading target period or a serial number.

[0065] Fig. 6 is an explanatory diagram of the reading target period and the detection target period, and shows that the measurement data for the reading target period is periodically read out during the process of storing the measurement data in time series. Fig. 6A shows a case where the reading target period of the measurement data for creating the learning data and the period of the measurement data for the detection target using the learning data (detection target period) coincide. The learning data is created from the read measurement data, and the model 2M is learned from the created learning data. In Fig. 6A, the model 2M is applied to the anomaly detection of the measurement data measured in the same period as the measurement data that is the source of the learning data.

[0066] As shown in Fig. 6A, when the period of the measurement data of the learning data and the detection target period using the model 2M coincide, the influence due to the difference in the period or environment between the time of learning the model 2M and the time of anomaly detection using the model 2M can be eliminated.

[0067] FIG. 6B shows a case where the read period of measurement data for creating learning data and the detection period of measurement data using the learning data are shifted slightly and used. In FIG. 6B, the model 2M is applied to the anomaly detection of measurement data read for a period different from the measurement data that is the source of the learning data.

[0068] In a situation where the environment does not change significantly, for example, within 1 to 2 weeks, or when the power storage system 101 is stopped, as shown in FIG. 6B, the read period of the learning data and the detection period do not necessarily have to match. The model 2M learned from the measurement data in the 2-week read period from 3 weeks ago to 1 week ago may perform anomaly detection on the measurement data in the most recent 2-week detection period.

[0069] FIG. 7 is a schematic diagram of an example of the created model 2M. In one example, the model 2M uses a convolutional neural network, inputs measurement data measured by a plurality of power storage cells, and outputs the possibility that the input measurement data includes measurement data of a power storage cell different from the input measurement data. The model 2M may be an autoencoder.

[0070] In the example shown in FIG. 7, the model 2M includes an input layer 201 that inputs the measurement data of each of the plurality of power storage cells included in the same module. The model 2M includes an output layer 202 that outputs a score based on the input measurement data, and an intermediate layer 203 that includes a convolutional layer or a pooling layer. The model 2M is learned by giving a label of "not heterogeneous" to the learning data created by averaging and providing it to the neural network. The model 2M outputs from the output layer 202 a score corresponding to the possibility that the measurement data of a non-homogeneous power storage cell is included.

[0071] In another example, the model 2M may be a model that inputs time-series data of measurement data (for example, voltage values) of the same power storage cell and outputs a score corresponding to the possibility that the measurement data of a different power storage cell is included. The model 2M may be a classifier that classifies whether the input measurement data is measurement data of an abnormal power storage cell.

[0072] According to the design of Model 2M, the number of groups of measurement data during the reading target period in step S102 shown in FIG. 5 is determined. The Model 2M shown in FIG. 7 inputs the voltage values of, for example, 12 power storage cells included in the module. In step S103 shown in FIG. 5, the processing unit 20 creates a plurality of sets of learning data corresponding to the number of times measured over the reading target period, with 12 average values of voltage values as one set. The number of groups in step S102 may be 12 or a multiple of 12. The grouping may be performed such that the measurement data overlaps between groups.

[0073] FIG. 8 is a schematic diagram of learning data creation. FIG. 8 shows a table representing the identification information (identification number) of the modules in rows and columns. Each module is given identification information representing the [Y]th module in the [X]th bank as B[X]M[Y]. The table in FIG. 7 shows the identification information of 144 modules. The power storage cells are given identification information of C[Z] according to the connection order [Z] in each module. The learning data is created by averaging the measurement data of the power storage cells with the same number (connection order) in each module. The measurement data of the [Z]th power storage cell in the [Y]th module of the [X]th bank is represented as B[X]M[Y]C[Z]. The averaging is performed, for example, as follows. (B1M1C1 +B1M2C1 +…+B1M12C1 +B2M1C1 +…+B12M12C1 ) / 144 (B1M1C2 +B1M2C2 +…+B1M12C2 +B2M1C2 +…+B12M12C2 ) / 144 … (B1M1C12 +B1M2C12 +…+B1M12C12 +B2M1C12 +…+B12M12C12 ) / 144

[0074] As described above, the measurement data is averaged using the measurement data of the power storage cells in the same connection order among the series-connected power storage cells. When there is a non-operating bank (a bank in a standby state), the measurement data of the non-operating bank is excluded from the averaging target.

[0075] Anomaly detection processing based on the model 2M learned by the created learning data will be described. FIG. 9 is a flowchart showing an example of the anomaly detection processing procedure by the server device 2. The processing unit 20 of the server device 2 executes the following processing at the same cycle as the execution cycle of the processing procedure in FIG. 5. The processing procedure shown in FIG. 9 corresponds to the “detection unit”.

[0076] The processing unit 20 reads out the measurement data of the detection target for the detection target period from the measurement data of each power storage cell associated with the time information in the storage unit 21 (step S201). In step S201, the processing unit 20 selects and reads out the measurement data of the power storage cells included in the same module.

[0077] The processing unit 20 reads out the model 2M corresponding to the detection target period from the storage unit 21 (step S202). As described above, the model 2M corresponding to the detection target period is the model 2M learned by the measurement data of the read target period that matches the detection target period, or the model 2M learned by the measurement data of the read target period that partially overlaps with the detection target period.

[0078] The processing unit 20 provides the measurement data of the detection target read in step S201 to the model 2M read in step S202 (step S203). The processing unit 20 acquires the score output from the model 2M (step S204).

[0079] In step S203, the processing unit 20 provides the measurement data (voltage values) of each of the plurality of power storage cells included in the same module, and in step S204, acquires a score indicating whether the measurement data of the heterogeneous power storage cells is included in the measurement data.

[0080] The processing unit 20 stores the score acquired in step S203 in the storage unit 21 in association with the identification data for identifying the battery cell group of the measurement data of the detection target and the time information of the acquired measurement data (step S205).

[0081] The processing unit 20 reads out the scores for the past predetermined time stored in the storage unit 21 for the measurement data of the detection target (step S206). The processing unit 20 creates a time distribution of the scores for the past predetermined time (step S207).

[0082] Based on the time distribution created in step S207, the processing unit 20 determines whether or not the measurement data of the detection target includes abnormal measurement data (step S208). In step S208, the processing unit 20 may make the determination with reference to the score acquired in step S204. The processing unit 20 may also make the determination with reference to the measurement data itself read out in step S201.

[0083] If it is determined in step S208 that the measurement data includes abnormal measurement data (S208: YES), the processing unit 20 specifies that the measurement data of the detection target is abnormal (step S209), and advances the processing to step S211.

[0084] If it is determined that the measurement data does not include abnormal measurement data (S208: NO), the processing unit 20 specifies that the measurement data of the detection target is not abnormal (step S210), and advances the processing to step S211.

[0085] The processing unit 20 determines whether or not all the measurement data has been selected in step S201 (step S211). If it is determined that not all the data has been selected (S211: NO), the processing unit 20 returns the processing to step S201.

[0086] If it is determined that all the data has been selected (S211: YES), the processing unit 20 ends the abnormality detection process.

[0087] The processing unit 20 determined whether each module in which the power storage cells are connected in series includes abnormal measurement data. Alternatively, depending on the design of the model 2M, the unit of the power storage cells to be detected may be determined. For example, the determination may be made in units of banks, or the determination may be made for each power storage cell.

[0088] FIG. 10 is a graph that simulates the time distribution of measurement data of a plurality of power storage cells. The horizontal axis of FIG. 10 indicates the passage of time. The vertical axis of FIG. 10 indicates the magnitude of the value of the measurement data. In the graph of FIG. 10, the curve indicated by the solid line is the measurement data of normal power storage cells. In the graph of FIG. 10, the curves indicated by the dashed line and the two-dot chain line are the measurement data of abnormal (or heterogeneous) power storage cells.

[0089] As shown in FIG. 10, the measurement data of abnormal power storage cells is either too large or too small compared to the normal measurement data. The amount of measurement data of abnormal power storage cells is very small compared to the amount of measurement data of normal power storage cells. When the measurement data is averaged including these excessive and excessive measurement data, it is presumed that the average value does not vary significantly from the normal measurement data indicated by the solid line. The learning data of the model 2M used for the abnormality detection method is not labeled as normal data that does not include the measurement data of abnormal power storage cells, nor is it labeled as the measurement data of abnormal power storage cells.

[0090] FIG. 11 is a diagram showing the application range of the abnormality detection method. FIG. 11 shows the attributes of a set of measurement data. The measurement data includes the measurement data of normal power storage cells and the measurement data of abnormal power storage cells for the population. The normal power storage cells include standard power storage cells and power storage cells in a state that is normal but different (heterogeneous) from other power storage cells. The abnormal power storage cells include power storage cells showing known abnormalities or their signs and power storage cells showing unknown abnormalities or their signs.

[0091] In FIG. 11, among the attributes of each measurement data, the data attributes to be learned and the data attributes to be detected by the learned model are indicated by hatching. FIG. 11A shows the learning target and the detection target of the learning model used in the conventional anomaly detection. As shown in FIG. 11A, in the conventional anomaly detection, a learned model using teacher data with a label of being abnormal was used for the measurement data of known abnormal energy storage elements. It is necessary to prepare a sufficient number of abnormal data as learning data. In the conventional anomaly detection, the measurement data of known abnormal energy storage elements are detected. In the conventional learned model, the measurement data of an energy storage element with an unknown anomaly may be outside the detection target of the anomaly. The energy storage element may exhibit an unknown pattern of anomaly depending on the usage environment and the usage period. That is, when used in an environment different from the test process of the energy storage element, an anomaly that cannot be detected by the learning model based on the pre-created learning data may occur. It is difficult to distinguish an energy storage cell that may exhibit an unknown pattern of anomaly before the start of operation.

[0092] FIG. 11B shows the learning target and the detection target of the learning model in other anomaly detections. The learning model in FIG. 11B learns only the data of energy storage cells with standard characteristics as designed, and is learned to detect data with attributes different from those of the standard energy storage cells. In the case of FIG. 11B, for measurement data in which the measurement data of an energy storage element with an attribute different from that of the learning target energy storage element is mixed, it is determined to be abnormal. In this case, an unknown anomaly or its omen can be detected. However, an energy storage cell that is normal but has a different (heterogeneous) state from other energy storage cells is also determined to be abnormal. For example, when new energy storage elements are mixed with energy storage elements that have been in operation for several years, the new energy storage elements are determined to be abnormal.

[0093] FIG. 11C shows the learning target and detection target of the model 2M of the present embodiment. As shown in FIG. 11C, since the model 2M learns by averaging all data including anomalies and normals, it is possible to detect measurement data that deviates from the average pattern, and it is also possible to detect heterogeneous measurement data such as measurement data of a new power storage element. By using the average value as the learning data, it becomes possible to distinguish heterogeneity in the midst of a certain change (trend) in the entire power storage system 101. For example, in the case where the temperature changes due to seasonal changes, most of the characteristics of the power storage cells included in the power storage system 101 change with a certain characteristic due to the change in temperature. Among them, it becomes possible to extract only heterogeneous power storage cells or modules that do not follow the trend.

[0094] FIG. 12 shows an example of the status screen 331 displayed on the client device 3. The status screen 331 includes an image K1 that visually shows the configuration of the power storage system 101. In the image K1, the arrangement of two domains is shown. Each rectangle in the image K1 represents a bank. The image K1 indicates in a thick frame that the first bank of domain 2 is selected. The rectangles representing the banks in the image K1 indicate the presence or absence of anomalies by colors and patterns shown by hatching. The image K2 shows the arrangement and status of the modules included in the bank selected in the image K1. Each rectangle in the image K2 represents a module. The rectangles of the modules with measurement data for which anomalies have been detected are emphasized by an object 332 with different colors or patterns. The status screen 331 includes an object 333 that visually shows the SOC of the entire selected bank. In this way, the anomalies detected for each power storage cell and module are visually output by the status screen 331.

[0095] The embodiments disclosed as above are illustrative in all respects and not restrictive. The scope of the present invention is indicated by the scope of the claims, and includes all modifications within the meaning and scope equivalent to the scope of the claims.

Explanation of Signs

[0096] 101 Power storage system 2 Server device 20 Processing unit 21 Memory unit 22P, 52P Abnormality detection program 2M, 5M Model 5 Recording medium

Claims

1. A creation unit that creates learning data by statistically processing a plurality of measurement data of a power storage element without excluding abnormal measurement data; A storage unit that stores a model that is learned to output a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input, using the created learning data; A detection unit that detects an abnormality or a sign of an abnormality of the power storage element based on a score output by inputting the plurality of measurement data into the model An abnormality detection device comprising:

2. The creation unit creates the learning data using an average of a plurality of measurement data that may include abnormal measurement data of the power storage element The abnormality detection device according to claim 1.

3. The power storage element is configured by connecting a plurality of modules each including a plurality of power storage cells in series, The creation unit creates the learning data by averaging measurement data of power storage cells of the same rank in the plurality of modules The abnormality detection device according to claim 2.

4. The power storage element is configured such that a domain is formed by connecting a plurality of banks, each bank being configured by connecting a plurality of modules each including a plurality of power storage cells in series, in parallel, The creation unit creates the learning data by averaging measurement data of power storage cells of the same rank in the plurality of modules included in the domain The abnormality detection device according to claim 2.

5. The creation unit creates the learning data using measurement data read out for a read target period among measurement data measured in time series from the power storage element, The detection unit inputs measurement data of a detection target period that is the same period as the read target period into the model learned by the learning data, and detects an abnormality or a sign of an abnormality of the power storage element in the detection target period based on a score output from the model The abnormality detection device according to any one of claims 1 to 4.

6. The creation unit creates the learning data using measurement data read out for a read target period among measurement data measured in time series from the power storage element, The detection unit inputs measurement data of a detection target period that partially overlaps with the read target period into the model learned by the learning data, and detects an abnormality or a sign of an abnormality of the power storage element in the detection target period based on a score output from the model The abnormality detection device according to any one of claims 1 to 4.

7. Statistically process a plurality of measurement data of the energy storage element that have not excluded abnormal measurement data to create learning data, using the created learning data, learn a model to output a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input, store the learned model, detect an abnormality or a sign of an abnormality of the energy storage element based on the score output by inputting the plurality of measurement data into the model Anomaly detection method.

8. On a computer, statistically process a plurality of measurement data of the energy storage element that have not excluded abnormal measurement data to create learning data, using the created learning data, learn a model to output a score corresponding to whether or not the measurement data includes abnormal measurement data when the measurement data is input, store the learned model, detect an abnormality or a sign of an abnormality of the energy storage element based on the score output by inputting the plurality of measurement data into the model A computer program for causing the above processing to be executed.

Images