Machine learning device and battery state determination device

Abstract

To provide a mechanical learning device and a storage battery state determination device that allow an inspection person to make an accurate determination regardless of the experience or the ability of the inspection person.SOLUTION: A mechanical learning device 10 includes: a first input data acquisition unit 111; a first label acquisition unit 112; and a first learning model construction unit 113. The first input data acquisition unit 111 acquires actual object picture information of at least one of the appearance and the inside of a storage battery 40 as input data. The first label acquisition unit 112 acquires actual object data showing the state of the storage battery 40 corresponding to the actual object picture information, as a label. The first learning model construction unit 113 performs a learning with a teacher by using a combination of the actual object picture information and actual object data as teacher data, thereby constructing a first learning model as a learning model to determine the state of the storage battery 40.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a machine learning device and a storage battery state determination device. [Background technology]

[0002] Conventionally, there are known techniques for monitoring the battery status of electronic devices. Patent Document 1 describes this type of technique. Patent Document 1 describes a battery management system that includes an electronic device equipped with a battery and a management server connected to the electronic device via a network, in which the electronic device stores battery information including a status indicating the battery status in a storage medium, and the management server predicts the current status of the battery of the electronic device based on the battery information of the electronic device, and transmits control information for battery control of the electronic device to the electronic device based on the predicted result. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Application No. 2020-523219 Summary of the Invention [Problem to be solved by the invention]

[0004] By the way, maintenance and inspection of storage batteries during float charging may involve visually inspecting the appearance of the battery case and lid, as well as the condition of the positive and negative plates inside the battery. Visual inspection requires experience and skill, and there is room for improvement in terms of the accuracy of the inspection when the inspector lacks experience or skill.

[0005] The present invention aims to provide a machine learning device and a battery state determination device that enable more accurate determinations regardless of the experience or ability of an inspector. [Means for solving the problem]

[0006] (1) A machine learning device according to the present invention includes an input data acquisition unit that acquires image data of at least one of the exterior and interior of a storage battery as input data, a label acquisition unit that acquires status information indicating the status of the storage battery corresponding to the image data as a label, and a model construction unit that constructs a learning model for determining the status of the storage battery by performing supervised learning using a pair of the image data and the status information as training data.

[0007] (2) In the machine learning device according to the present invention, the label acquisition unit acquires degree information indicating the degree of abnormality as the state information, and the model construction unit constructs a learning model that outputs the degree information in accordance with the input image data.

[0008] (3) In the machine learning device according to the present invention, the label acquisition unit acquires, as the status information, abnormality location information indicating the location where an abnormality has occurred in the storage battery corresponding to the image data, and the model construction unit constructs a learning model for determining the location where the abnormality has occurred by performing supervised learning using a pair of the image data and the abnormality location information as training data.

[0009] (4) In the machine learning device according to the present invention, the input data acquisition unit acquires, as input data, specific gravity information of the electrolyte at a predetermined timing before the end of the life of the storage battery, the label acquisition unit acquires, as the status information, remaining life period information indicating the remaining life of the storage battery from the predetermined timing to the end of the life of the storage battery, and the model construction unit constructs a learning model for determining the remaining life by performing supervised learning using a pair of the specific gravity information and the remaining life period information as training data.

[0010] (5) The storage battery state determination device according to the present invention includes a learning model acquisition unit that acquires the learning model constructed by the machine learning device described in any one of (1) to (4), a new input data acquisition unit that acquires image data of at least one of the exterior and interior of the storage battery or specific gravity information of the electrolyte of the storage battery as new input data, and a storage battery state determination unit that determines the state of the storage battery based on the new input data and the learning model and outputs the determination result. [Effects of the Invention]

[0011] According to the present invention, it is possible to provide a machine learning device and a battery state determination device that enable more accurate determinations regardless of the experience or ability of an inspector. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a diagram showing the overall configuration of a storage battery management system according to an embodiment of the present invention; [Figure 2] FIG. 2 is a functional block diagram of a first learning unit included in the machine learning device according to the embodiment of the present invention. [Figure 3] FIG. 2 is a functional block diagram of a second learning unit included in the machine learning device according to the embodiment of the present invention. [Figure 4] FIG. 2 is a functional block diagram of a third learning unit included in the machine learning device according to the embodiment of the present invention. [Figure 5] 1 is a functional block diagram of a battery state determination device according to an embodiment of the present invention; [Figure 6] 4 is a flowchart showing the operation of a first learning unit in the operation of the machine learning device according to the embodiment of the present invention. [Figure 7] 4 is a flowchart showing the operation of a second learning unit in the operation of the machine learning device according to the embodiment of the present invention. [Figure 8] 4 is a flowchart showing the operation of a third learning unit in the operation of the machine learning device according to the embodiment of the present invention. [Figure 9] 4 is a flowchart illustrating an operation of the battery state determination device according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present invention will be described with reference to FIGS.

[0014] [Configuration of one embodiment] [1.1 Overall Structure] First, the configuration of a battery management system S according to this embodiment will be described. Fig. 1 is a diagram showing the overall configuration of the battery management system. The battery management system S is a system for managing the status of multiple batteries 40 in floating charge, installed in a power plant or the like.

[0015] The storage battery 40 according to this embodiment is composed of multiple batteries called cells, and for example has six cells, 40a, 40b, 40c, 40d, 40e, and 40f, as shown in FIG. 1. The number of cells is not limited to this and can be freely set. As shown in FIG. 1, the storage battery management system S includes a machine learning device 10, a storage battery state determination device 20, a terminal device 30, and a network N. The machine learning device 10, the storage battery state determination device 20, and the terminal device 30 may be one or more.

[0016] Furthermore, the machine learning device 10, the battery state determination device 20, and the terminal device 30 are each connected to a network N and can communicate with each other via the network N. The network N is, for example, a local area network (LAN), the Internet, a public telephone network, or a combination of these. The specific communication method of the network N, whether it is a wired connection or a wireless connection, etc., are not particularly limited. The machine learning device 10 and the battery state determination device 20 may be directly connected via a connection unit, rather than communicating via the network N.

[0017] 1, the machine learning device 10 includes a first learning unit 11, a second learning unit 12, and a third learning unit 13. The first learning unit 11, the second learning unit 12, and the third learning unit 13 are learning models that perform supervised learning based on input training data and are used by the battery state determination device 20 to determine the state of the battery.

[0018] In this embodiment, as training data for constructing a learning model capable of determining an abnormality in storage battery 40 from an actual photograph, for example, actual photograph information obtained by photographing the exterior or interior of the actual storage battery 40 is used as input data. Furthermore, as the training data, abnormality determination result information is used as a label, which is status information of the storage battery 40 determined by a skilled inspector. That is, as the training data, a set of data consisting of actual photograph information and abnormality determination result information is used.

[0019] In this embodiment, the above-described actual photograph information is used as input data for training data for constructing a learning model capable of determining the location of an abnormality in storage battery 40 from an actual photograph. The training data also uses abnormality location information as status information of storage battery 40 determined by a skilled inspector as a label. That is, the training data uses a set of data consisting of actual photograph information and abnormality location information.

[0020] Furthermore, in this embodiment, historical information on actual data such as the voltage of storage battery 40 that has reached the end of its life is used as input data for training data for constructing a learning model capable of determining the remaining life of storage battery 40 from the actual data of storage battery 40. The actual data according to this embodiment includes measurement data measured during routine inspections such as the liquid level and temperature of the electrolyte in storage battery 40, specific gravity information, voltage, and outside air temperature, as well as equipment data such as aging information from the start of use and the manufacturing date. The historical information includes actual data from the start of use of storage battery 40 until the end of its life.

[0021] The information included in the actual data is not limited to this, and may also include, for example, the rate at which the electrolyte decreases, information on the state of the positive and negative plates, information on variations in electrolyte temperature and voltage between cells, battery capacity, degree of battery deterioration, and battery charge / discharge efficiency.

[0022] The training data uses status information of the storage battery 40 whose life as a label has expired and life information as the remaining life period. That is, the training data uses a set of data consisting of actual data and life information of the storage battery 40. The history information included in the training data includes inspection date and time information, and the remaining life at the time of measurement of the measurement data can be calculated based on the life information and is used for machine learning.

[0023] The storage battery state determination device 20 is a device that determines the state of various storage batteries 40 by acquiring new input data necessary for determining the state of the storage battery 40 and applying the acquired data to the learning model constructed by the machine learning device 10. In this embodiment, the storage battery state determination device 20 determines the state of the storage battery 40 by acquiring new input data for each cell.

[0024] The terminal device 30 is a device that displays the battery state determination result information output from the battery state determination device 20 and outputs input from a user of the terminal device 30 to the battery state determination device 20. The terminal device 30 is realized by, for example, a mobile phone such as a smartphone, or a mobile terminal such as a tablet terminal, but is not limited to this. Furthermore, although multiple terminal devices 30, terminal devices 30a to 30q, are shown in FIG. 1, the present invention is not limited to this and there may be only one terminal device, for example.

[0025] In the battery management system S according to this embodiment, the machine learning device 10 and the battery state determination device 20 are provided separately, but this is not limiting and they may be provided in the same device.

[0026] [1.2 Machine learning device configuration] As described above, the machine learning device 10 includes the first learning unit 11, the second learning unit 12, and the third learning unit 13. The first learning unit 11 according to this embodiment constructs a first learning model for determining an abnormality in the storage battery 40 contained in new actual photograph information that includes the storage battery 40 as a subject. An abnormality in the storage battery 40 refers to a state in which the battery case, lid, positive and negative electrode plates, or electrolyte of the storage battery 40 has cracks, deformation, damage, or the like.

[0027] Fig. 2 is a functional block diagram of the first learning unit 11. As shown in Fig. 2, the first learning unit 11 includes a first input data acquisition unit 111, a first label acquisition unit 112, a first learning model construction unit 113 as a model construction unit, and a first learning model storage unit 114.

[0028] The first input data acquisition unit 111 acquires, as input data, actual photographic information that includes the storage battery 40 as a subject. For example, the first input data acquisition unit 111 executes a process of acquiring information about the storage battery to be inspected that is input to the terminal device 30 and transmitted to the machine learning device 10 via the network N.

[0029] The first label acquisition unit 112 acquires, as a label, the level of abnormality (degree of abnormality) of the storage battery 40 contained in actual photograph information that includes the storage battery 40 as a subject. Note that the setting of the level of abnormality of the storage battery 40 is based, to some extent, on abnormality determination result information, which is the result of a skilled inspector's judgment based on a photograph of the actual battery or on directly viewing the actual battery. However, the present invention is not limited to this.

[0030] In this embodiment, the abnormality of the storage battery 40 is determined according to one of four levels, Level A to Level D. The level of abnormality of the storage battery 40 may be, for example, a score indicated by a numerical value, or may be a level indicating the degree of abnormality using letters or symbols such as excellent, good, fair, and poor.

[0031] Level A indicates an abnormal condition in which an abnormality has occurred in the battery case or lid of the storage battery 40, or in the positive and negative plates or electrolyte inside the storage battery 40, and immediate action is required. For example, Level A is a level at which it is possible to confirm from an actual photograph that the battery case of the storage battery 40 has a crack with an opening through which the electrolyte may leak, or that there is significant deformation that is thought to be caused by abnormal heat generation. Alternatively, Level A indicates a condition in which the storage battery 40 is unable to operate normally, such as when the positive and negative plates inside the storage battery 40 are missing. Alternatively, Level A indicates a condition in which the electrolyte has significantly deteriorated or the liquid level has abnormally decreased, such as when the battery is unable to operate normally.

[0032] Level B indicates a state in which the storage battery 40 needs to be inspected at the next inspection. For example, Level B indicates a state in which, although no electrolyte leakage or the like is confirmed in the storage battery 40, cracks or deformations are confirmed on the surface, and the storage battery 40 is likely to break down in the future. Alternatively, Level B indicates a state in which the internal positive and negative plates are severely damaged, and the storage battery 40 is likely to become inoperable in the future. Alternatively, Level B indicates a state in which the electrolyte has changed in quality or the liquid level has decreased, and the storage battery 40 is likely to become inoperable in the future.

[0033] Level C indicates a state in which it is necessary to reduce the frequency of inspections of the storage battery 40. For example, measures such as increasing the frequency of inspections of the determined parts from once a year to three times a year are necessary. Level C indicates a state in which, for example, although no electrolyte leakage, cracks, deformation, or damage is found on the surface of the storage battery 40, it is confirmed that the external shape of the storage battery 40 has deformed from a normal storage battery, and there is a possibility that deterioration of the storage battery 40 has occurred. Alternatively, Level C indicates a state in which damage, deformation, or alteration is found on the internal positive and negative electrode plates, and there is a possibility that deterioration has occurred. Alternatively, Level C indicates a state in which slight changes such as alteration of the electrolyte or a decrease in the liquid level are found, and there is a possibility that deterioration has occurred.

[0034] Level D is a state in which the surface and interior of the battery case and lid of the storage battery 40 are in the same state as a normal storage battery, and no abnormality has occurred in the storage battery 40.

[0035] The first learning model construction unit 113 constructs a first learning model for determining an abnormality in the storage battery 40 included in the actual photo information by performing supervised learning using a pair of the first input data and the first label as training data. The constructed first learning model is used for determination by the storage battery state determination device 20.

[0036] The first learning model construction unit 113 according to this embodiment constructs the first learning model as a trained model by machine learning. For example, the first learning model construction unit 113 constructs the first learning model by performing deep learning using a neural network with a multi-layer structure (input layer, output layer, intermediate layer).

[0037] Note that the algorithm used by the first learning model construction unit 113 to construct the first learning model is not limited to this. For example, the first learning model construction unit 113 may perform supervised learning using a support vector machine (also referred to as SVM) to construct the first learning model, or may perform learning using other algorithms to construct the first learning model.

[0038] In this case, the first learning model construction unit 113 uses a binarized label indicating whether or not the input data corresponds to a specific degree of abnormality as the first label, and calculates a hyperplane that separates the space including the input data so as to maximize the margin regarding whether or not the input data corresponds to the specific degree of abnormality. Furthermore, the first learning model construction unit 113 can use the coefficients of this hyperplane as parameters of a learning model that the battery state determination device 20 (described later) uses to determine the battery state.

[0039] The first learning model storage unit 114 executes a process of storing the first learning model constructed by the first learning model construction unit 113 in a storage device of the machine learning device 10. For example, when the first learning model construction unit 113 constructs the first learning model, the first learning model storage unit 114 executes a process of storing the constructed first learning model in a storage device of the machine learning device 10.

[0040] The second learning unit 12 constructs a second learning model for determining the location of an abnormality such as a crack in the storage battery 40 contained in new actual photograph information that includes the storage battery 40 as a subject. Fig. 3 is a functional block diagram of the second learning unit 12. The second learning unit 12 includes a second input data acquisition unit 121, a second label acquisition unit 122, a second learning model construction unit 123 as a model construction unit, and a second learning model storage unit 124.

[0041] The second input data acquisition unit 121 acquires, as second input data, an actual photograph including the storage battery 40 as a subject. The second label acquisition unit 122 acquires, as a second label, a determination result of an abnormal position, such as a crack, in the storage battery 40, which is included in actual photograph information including the storage battery 40 as a subject. In this embodiment, the determination result of the abnormal position is obtained by a skilled inspector directly checking the actual photograph or the actual battery and making a determination.

[0042] The abnormality location determination result is not limited to the visual determination by a skilled inspector, and includes not only the location of the crack but also the location of deformation, damage, and the location of electrolyte leakage from the crack.

[0043] The second learning model construction unit 123 constructs a second learning model for determining the location of an abnormality, such as a crack, in the storage battery 40 contained in new actual photograph information that includes the storage battery 40 as a subject, by performing supervised learning using a pair of the second input data and the second label as training data using a method similar to that used by the first learning model construction unit 113. The constructed second learning model is used for determination by the storage battery state determination device 20.

[0044] The second learning model storage unit 124 executes a process of storing the second learning model constructed by the second learning model construction unit 123 in a storage device of the machine learning device 10. For example, when the second learning model construction unit 123 constructs a second learning model, the second learning model storage unit 124 executes a process of storing the constructed second learning model in a storage device of the machine learning device 10.

[0045] The third learning unit 13 constructs a third learning model for predicting the remaining life of the storage battery 40 from the actual data. As described above, historical information and life information of the actual data of the storage battery 40 whose life has ended are used as training data. FIG. 4 is a functional block diagram of the third learning unit 13. The third learning unit 13 includes a third input data acquisition unit 131, a third label acquisition unit 132, a third learning model construction unit 133 as a model construction unit, and a third learning model storage unit 134.

[0046] The third input data acquisition unit 131 acquires the above-mentioned actual data as the third input data from history data in the measurement result database 213 of the memory unit 21 of the battery state determination device 20 (described later). The third input data acquisition unit 131 may also calculate other data based on the actual data and acquire the calculated data as the third input data. For example, the third input data acquisition unit 131 may calculate the rate of decrease of the electrolyte from the level of the electrolyte surface acquired at a different time.

[0047] The third label acquisition unit 132 acquires, as a third label, the remaining life of the storage battery 40 indicated by the third input data. The remaining life of the storage battery 40 acquired by the third label acquisition unit 132 is calculated from the measurement date of the outside temperature and the like of the third input data read from the history data of the measurement result database 213 described below and the date and time when the life of the storage battery 40 ended.

[0048] The third learning model construction unit 133 constructs a third learning model for predicting the remaining life of the storage battery 40 by performing supervised learning using a pair of the third input data and the third label as training data in a manner similar to that of the first learning model construction unit 113. The constructed third learning model is used for the determination of the storage battery state determination device 20.

[0049] The third learning model storage unit 134 executes a process of storing the learning model constructed by the third learning model construction unit 133 in a storage device of the machine learning device 10. For example, when the third learning model construction unit 133 constructs a third learning model, the third learning model storage unit 134 executes a process of storing the constructed third learning model in a storage device of the machine learning device 10.

[0050] [1.3 Configuration of the battery state determination device] 5 is a functional block diagram of the battery state determination device 20. The battery state determination device 20 includes a storage unit 21, a control unit 22, a communication unit 23, and a display unit 24.

[0051] The storage unit 21 stores the learning model acquired from the machine learning device 10. The storage unit 21 further stores a facility information database 211, a learning model database 212, and a measurement result database 213.

[0052] Equipment data and the like are stored in the equipment information database 211. For example, the equipment information database 211 stores equipment information such as the date of start of use of the storage battery 40, the serial number, the manufacturer, and the date of manufacture.

[0053] The learning model database 212 stores multiple learning models transmitted from the first learning model construction unit 113, the second learning model construction unit 123, and the third learning model construction unit 133 of the machine learning device 10, and is read when the process of determining the state of the storage battery 40 is executed.

[0054] The measurement result database 213 receives as input the inspection results transmitted via the inspector's terminal device 30 during a patrol inspection of the storage battery 40. For example, the measurement result database 213 stores image data of the storage battery 40 and storage battery information such as the liquid level of the electrolyte, specific gravity, outside air temperature, voltage, and the temperature of the electrolyte in the cells. Note that the inspection results may be measured automatically by providing various sensors and communication devices in the storage battery 40, and transmitted to and stored in the measurement result database 213 via the network N.

[0055] The control unit 22 is a part that controls the entire storage battery state determination device 20, and realizes various functions in this embodiment by appropriately reading and executing various programs from a storage area such as a ROM, RAM, flash memory, or hard disk (HDD). The control unit 22 may be a CPU. The control unit 22 includes an actual object photo acquisition unit 221, an actual object data acquisition unit 222, a learning model acquisition unit 223, an abnormality determination unit 224, an abnormality position determination unit 225, a remaining life determination unit 226, and a storage battery state management unit 227 as a storage battery state determination unit.

[0056] The actual object photo acquisition unit 221 acquires new actual object photo information separately from the actual object photo information included in the training data used when learning was performed by the machine learning device 10. For example, an inspector operates the terminal device 30 during a patrol inspection to load the actual object photo information into the battery state determination device 20, and the actual object photo acquisition unit 221 acquires the actual object photo information.

[0057] The actual object data acquisition unit 222 acquires new actual object data separately from the actual object data included in the training data used when machine learning was performed by the machine learning device 10. For example, an inspector operates the terminal device 30 to read the actual object data into the battery state determination device 20, and the actual object data acquisition unit 222 acquires the actual object photograph information.

[0058] The actual data acquisition unit 222 may acquire new actual data or actual photograph information from the battery state determination device 20 into which new actual photograph information or actual data has been input by an inspector operating an input device such as a keyboard or a mouse provided on the battery state determination device 20. The actual data acquisition unit 222 may also acquire the actual data or actual photograph information stored in the measurement result database 213 or the equipment information database 211 of the memory unit 21 of the battery state determination device 20 as new input data.

[0059] The learning model acquisition unit 223 executes a process of acquiring a learning model to be used for the determination when determining the state of the storage battery. For example, when determining an abnormality in the storage battery 40, the learning model acquisition unit 223 refers to the learning model database 212 to search for, read, and acquire a first learning model for determining an abnormality in the storage battery 40. If the first learning model is not present in the learning model database 212, the learning model acquisition unit 223 requests the machine learning device 10 to transmit the first learning model.

[0060] The abnormality determination unit 224 determines whether or not there is an abnormality in the storage battery 40 based on the actual photograph information acquired by the actual photograph acquisition unit 221 and the first learning model acquired by the learning model acquisition unit 223.

[0061] The abnormality position determination unit 225 determines the abnormality position in the storage battery contained in the new actual photo information based on the actual photo information acquired by the actual photo acquisition unit 221 and the second learning model acquired by the learning model acquisition unit 223.

[0062] The remaining life determination unit 226 determines the remaining life of the storage battery 40 based on the actual data acquired by the actual data acquisition unit 222 and the third learning model acquired by the learning model acquisition unit 223.

[0063] When the battery state management unit 227 acquires a battery state determination request, it causes the actual object photo acquisition unit 221 and the actual object data acquisition unit 222 to acquire various information, causes the abnormality determination unit 224, the abnormality position determination unit 225, and the remaining life determination unit 226 to perform determination processing, and outputs the results. For example, when the battery state management unit 227 acquires a battery state determination request received from the terminal device 30 via the communication unit 23, it instructs the actual object photo acquisition unit 221 and the actual object data acquisition unit 222 to acquire the necessary information. Next, the battery state management unit 227 causes the abnormality determination unit 224, the abnormality position determination unit 225, and the remaining life determination unit 226 to perform determination processing.

[0064] Next, the battery status management unit 227 transmits the battery status determination result performed by the battery status determination device 20, such as information on whether an investigation is necessary or the details of the abnormality, to the terminal device 30 via the communication unit 23 described below. The display unit of the terminal device 30 that receives the information displays whether an investigation of the battery 40 is necessary and the details of the abnormality, allowing the inspector to easily determine what to do about the battery 40.

[0065] The communication unit 23 is a communication interface used by the battery state determination device 20 to communicate with the machine learning device 10 and the terminal device 30.

[0066] The display unit 24 is a monitor that displays an operation screen for performing processing related to the battery management system S.

[0067] [2. Operation of the embodiment] Next, the operation of the storage battery management system S according to this embodiment will be described. The storage battery state determination process in the storage battery management system S according to this embodiment includes a learning stage in which a learning model is constructed by supervised learning using pairs of input data and labels acquired by the machine learning device 10 as training data. The training data used in the learning stage is a set of data consisting of actual photographic information or actual data of the storage battery 40 inspected by a skilled inspector and information on the assessment result of the storage battery 40 by the skilled inspector.

[0068] The battery state determination process also includes a utilization stage in which the battery state determination device 20 uses the learning model to determine the battery state from actual photograph information or actual data as new input data input from the terminal device 30, etc. Below, the operation of the learning stage by the machine learning device 10 and the operation of the utilization stage by the battery state determination device 20 will be described.

[0069] 2.1 Learning model construction process of the machine learning device 10 First, the learning model construction process of the machine learning device 10 will be described with reference to Figures 6 to 8. When the machine learning device 10 receives a request to execute machine learning, it causes the first learning unit 11, the second learning unit 12, and the third learning unit 13 to perform the learning model construction process.

[0070] 6 is a flowchart showing the learning model construction process of the first learning unit 11 of the machine learning device 10. First, the first input data acquisition unit 111 and the first label acquisition unit 112 acquire, for a storage battery 40 inspected by a skilled inspector, actual object photo information as first input data and judgment result information of the storage battery 40 by the skilled inspector as a first label (step S10). Next, the first learning model construction unit 113 sets the acquired set of actual object photo information and judgment result information as training data (step S11). Next, the first learning model construction unit 113 performs machine learning using the training data (step S12).

[0071] Next, the first learning model construction unit 113 determines whether or not to end the machine learning (step S13). If the first learning model construction unit 113 determines to repeat the machine learning (step S13: NO), it shifts the process to step S10. Then, the machine learning device 10 repeats the same operation. On the other hand, if the first learning model construction unit 113 determines to end the machine learning (step S13: YES), it shifts the process to step S14. Note that the condition for ending the machine learning can be set arbitrarily. For example, the machine learning may be ended when the machine learning has been repeated a predetermined number of times. Furthermore, the accuracy of the constructed learning model improves as the machine learning is repeated.

[0072] Next, the first learning model construction unit 113 executes a process of transmitting the constructed learning model to the battery state determination device 20 (step S14), and ends the process.

[0073] 7 is a flowchart showing the learning model construction process of the second learning unit 12 of the machine learning device 10. First, the second input data acquisition unit 121 and the second label acquisition unit 122 acquire actual object photo information of the object to be judged by an experienced inspector as second input data and anomaly position information as a second label (step S20). Next, the second learning model construction unit 123 sets the acquired pair of actual object photo information and anomaly position information as training data (step S21). Next, the second learning model construction unit 123 performs machine learning using the training data (step S22).

[0074] Next, the second learning model construction unit 123 determines whether or not to end the machine learning (step S23). If the second learning model construction unit 123 determines to repeat the machine learning (step S23: NO), it shifts the process to step S20. Then, the machine learning device 10 repeats the same operation. On the other hand, if the second learning model construction unit 123 determines to end the machine learning (step S23: YES), it shifts the process to step S24. Note that the condition for ending the machine learning can be set arbitrarily. For example, the machine learning may be terminated when the machine learning has been repeated a predetermined number of times.

[0075] 8 is a flowchart showing the learning model construction process of the third learning unit 13 of the machine learning device 10. First, the third input data acquisition unit 131 and the third label acquisition unit 132 acquire actual object data as third input data and remaining life determination result information as a third label (step S30). Next, the third learning model construction unit 133 sets the acquired pair of actual object data and remaining life determination result information as training data (step S31). Next, the third learning model construction unit 133 performs machine learning using the training data (step S32).

[0076] Next, the third learning model construction unit 133 determines whether to end the machine learning (step S33). If the third learning model construction unit 133 determines to repeat the machine learning (step S33: NO), it shifts the process to step S30. Then, the machine learning device 10 repeats the same operation. On the other hand, if the third learning model construction unit 133 determines to end the machine learning (step S33: YES), it shifts the process to step S34. Note that the condition for ending the machine learning can be set arbitrarily. For example, the machine learning may be terminated when the machine learning has been repeated a predetermined number of times.

[0077] When the battery state determination device 20 receives each learning model transmitted from the machine learning device 10 through this process, it stores the model in a learning model database 212 in the storage device. Each learning model stored in the learning model database 212 is used to determine the state of a battery through the operation of the battery state determination device 20, which will be described below. Furthermore, when new training data is acquired, the machine learning device 10 can perform further machine learning on the learning model.

[0078] 2.2 Battery state determination process by the battery state determination device 20 Next, the battery state determination process of the battery state determination device 20 will be described with reference to Fig. 9. When the battery state determination device 20 receives a battery state determination request, it executes the battery state determination process.

[0079] 9 is a flowchart showing the battery state determination process of the battery state determination device 20 according to an embodiment of the present invention. First, the control unit 22 of the battery state determination device 20 acquires either or both of actual object photograph information and actual object data as new input data (step S40). Next, the control unit 22 checks whether the new input data includes actual object photograph information that includes the storage battery 40 as a subject (step S41).

[0080] If the new input data does not include actual item photo information (step S41: NO), the control unit 22 proceeds to step S47. On the other hand, if the new input data includes actual item photo information (step S41: YES), the control unit 22 refers to the learning model database 212 to acquire a first learning model (step S42). Next, the control unit 22 determines whether there is an abnormality in the storage battery 40 based on the actual item photo information included in the new input data acquired in step S40 and the first learning model (step S43).

[0081] Next, the control unit 22 checks whether or not there is an abnormality in the storage battery 40 (step S44). If there is no abnormality in the storage battery 40 (step S44: NO), the control unit 22 shifts the process to step S47. On the other hand, if there is an abnormality in the storage battery 40 (step S44: YES), the control unit 22 refers to the learning model database 212 and acquires a second learning model (step S45). Next, the control unit 22 determines the location of the abnormality in the storage battery 40 based on the acquired actual photo information and the second learning model (step S46).

[0082] Next, the control unit 22 acquires a third learning model by referring to the learning model database 212 (step S47). Next, the control unit 22 acquires actual data included in the new input data acquired in step S40, and determines the remaining life of the storage battery 40 based on the actual data and the third learning model (step S48). Next, the control unit 22 transmits the result of this determination to the terminal device 30 (step S49), and ends the process.

[0083] The machine learning device 10 configured as described above includes a first input data acquisition unit 111 that acquires actual photographic information of at least one of the exterior and interior of the storage battery 40 as input data, a first label acquisition unit 112 that acquires actual data indicating the state of the storage battery 40 as a label, and a first learning model construction unit 113 that constructs a first learning model as a learning model for determining the state of the storage battery 40 by performing supervised learning using a pair of the actual photographic information and the actual data as training data.

[0084] As a result, the machine learning device 10 according to this embodiment can make more accurate determinations regardless of the experience or ability of the inspector.

[0085] Furthermore, the first label acquisition unit 112 acquires degree information indicating the degree of abnormality, and the first learning model construction unit 113 constructs a learning model that outputs the degree information according to the input image data.

[0086] As a result, the machine learning device 10 according to this embodiment can more accurately determine whether or not a storage battery has an abnormality, regardless of the experience or ability of the inspector.

[0087] Furthermore, in the machine learning device 10 according to this embodiment, the second label acquisition unit 122 acquires, as status information, abnormality location information indicating the location of the abnormality in the storage battery 40 corresponding to the on-site photo information, and the second learning model construction unit 123 constructs a second learning model as a learning model for determining the location of the abnormality by performing supervised learning using a pair of actual photo information and abnormality location information as training data.

[0088] As a result, the machine learning device 10 according to this embodiment can easily identify the location where an abnormality has occurred, and can deal with the abnormality more efficiently.

[0089] In addition, in the machine learning device 10 according to this embodiment, the third input data acquisition unit 131 acquires, as input data, specific gravity information of the electrolyte at a predetermined timing before the end of the life of the storage battery 40, the third label acquisition unit 132 acquires, as status information, remaining life period information indicating the remaining life of the storage battery 40 from the predetermined timing to the time when the life of the storage battery 40 ends, and the third learning model construction unit 133 constructs a third learning model as a learning model for determining the remaining life by performing supervised learning using a pair of specific gravity information and remaining life period information as teaching data.

[0090] As a result, with the machine learning device 10 according to this embodiment, even an inexperienced inspector can more accurately determine the remaining life of the storage battery 40, enabling efficient operation of the storage battery 40, such as by using it as long as it has life.

[0091] [3. Modifications] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments. Furthermore, the effects described in the present embodiments are merely a list of the most preferable effects resulting from the present invention, and the effects of the present invention are not limited to those described in the present embodiments.

[0092] In the storage battery management system S according to this embodiment, the determination result is output after determining the state of the storage battery using a learning model, but this is not limiting, and the determination result may be further weighted based on the facility information stored in the facility information database 211. For example, the remaining life information of the storage battery included in the determination result information may be shortened or extended based on the materials used for the storage battery stored in the facility information database 211.

[0093] In the battery management system S according to this embodiment, measurement data and the like of the storage battery are transmitted from the terminal device 30 to the battery state determination device 20 through input operations by an inspector, but this is not limited thereto, and for example, the battery state determination device 20 may acquire the data directly from the storage battery 40 via a network. In this case, the storage battery 40 includes a communication unit capable of communicating with the battery state determination device 20 via the network, a measurement unit that acquires data from various measurement devices possessed by the storage battery 40, and a control unit that performs processing to transmit the measurement data acquired by the measurement unit to the battery state determination device 20 via the communication unit.

[0094] The above-described series of processes can be executed by hardware or software. In other words, the functional configurations of Figures 1 to 4 are merely examples and are not particularly limited. That is, it is sufficient for the machine learning device 10 to be provided with a function that can execute the above-described series of processes as a whole, and the functional blocks used to realize this function are not particularly limited to the examples of Figures 1 to 4.

[0095] Furthermore, one functional block may be configured as a hardware unit, a software unit, or a combination thereof. The functional configuration in this embodiment is realized by a processor that executes arithmetic processing, and processors that can be used in this embodiment include those configured as various processing units such as single processors, multiprocessors, and multicore processors, as well as those that combine these various processing units with processing circuits such as ASICs (Application Specific Integrated Circuits) and FPGAs (Field-Programmable Gate Arrays).

[0096] When a series of processes is executed by software, the programs that make up the software are installed into a computer or the like from a network or a recording medium. The computer may be a computer built into dedicated hardware. Alternatively, the computer may be a computer that can execute various functions by installing various programs, such as a general-purpose personal computer.

[0097] Recording media containing such programs include not only removable media distributed separately from the device main body in order to provide the program to the user, but also recording media provided to the user in a state where they are pre-installed in the device main body. Removable media include, for example, magnetic disks (including floppy disks), optical disks, and magneto-optical disks. Optical disks include, for example, CD-ROMs (Compact Disk-Read Only Memory), DVDs (Digital Versatile Disks), and Blu-ray (registered trademark) Discs. Magneto-optical disks include, for example, MDs (Mini-Disks). Recording media provided to the user in a state where they are pre-installed in the device main body include, for example, ROMs on which the program is recorded, and hard disks as storage devices.

[0098] In this specification, the steps of describing a program to be recorded on a recording medium include not only processes that are performed chronologically in accordance with the order, but also processes that are not necessarily performed chronologically but are performed in parallel or individually.

[0099] Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. The present invention can take on various other embodiments, and various modifications such as omissions and substitutions can be made without departing from the spirit of the present invention. These embodiments and their modifications are included within the scope and spirit of the invention described in this specification, etc., and are also included in the invention described in the claims and their equivalents. [Explanation of symbols]

[0100] 10 Machine Learning Device 40 Storage battery 111 First input data acquisition unit 112 First label acquisition unit 113 First Learning Model Construction Unit

Claims

1. an input data acquisition unit that acquires image data of at least one of the exterior and interior of the storage battery as input data; a label acquisition unit that acquires, as a label, status information indicating a status of the storage battery corresponding to the image data; a model construction unit that constructs a first learning model for determining the state of the storage battery by performing supervised learning using a pair of the image data and the state information as training data, the label acquisition unit acquires, as the status information, abnormality position information indicating a position where an abnormality has occurred in the storage battery corresponding to the image data; The model construction unit is a machine learning device that further constructs a second learning model for determining the location of the abnormality by performing supervised learning using a pair of the image data and the abnormality location information as training data.

2. an input data acquisition unit that acquires image data of at least one of the exterior and interior of the storage battery as input data; a label acquisition unit that acquires, as a label, status information indicating a status of the storage battery corresponding to the image data; a model construction unit that constructs a first learning model for determining the state of the storage battery by performing supervised learning using a pair of the image data and the state information as training data, the input data acquisition unit acquires, as input data, information on the specific gravity of the electrolyte at a predetermined timing before the end of the life of the storage battery; the label acquisition unit acquires, as the status information, remaining life period information indicating a remaining life of the storage battery from the predetermined timing to a timing when the life of the storage battery ends; The model construction unit is a machine learning device that constructs a third learning model for determining the remaining life by performing supervised learning using a pair of the weight information and the remaining life period information as training data.

3. the label acquisition unit acquires, as the status information, degree information indicating the degree of abnormality; The machine learning device according to claim 1 , wherein the model construction unit constructs the first learning model that outputs the degree information in accordance with input image data.

4. a learning model acquisition unit that acquires the first learning model and the second learning model constructed by the machine learning device according to claim 1; a new input data acquisition unit that acquires, as new input data, image data of at least one of the exterior and interior of the storage battery or specific gravity information of the electrolyte of the storage battery; determining a state of the storage battery based on image data of at least one of the exterior and interior of the storage battery and the first learning model; A battery status determination device comprising: a battery status determination unit that, when an abnormality is determined in the determination of the battery status based on the first learning model, determines the location of the abnormality based on image data of at least one of the exterior and interior of the battery and the second learning model, and outputs the determination result.

Images