Secondary battery degradation prediction system and secondary battery degradation prediction method

Abstract

This invention provides a secondary battery degradation prediction system and method that can predict degradation with high accuracy even when the time-series data of secondary batteries is short or the usage period is short. [Solution] The secondary battery degradation prediction system 2 is a secondary battery degradation prediction system 2 that predicts the degradation of a secondary battery 13 to be diagnosed based on the usage history information of other secondary batteries, and comprises a learning model construction unit 222 that constructs a learning model 223 using characteristic data of other secondary batteries and features that are finally determined during the learning process as input variables, a feature acquisition unit 224 that acquires features that are finally determined during the learning process of the secondary battery 13 to be diagnosed from the usage history information of the secondary battery 13 to be diagnosed, and a degradation prediction unit 225 that inputs the usage history information of the secondary battery to be diagnosed and the features that are finally determined during the learning process into the learning model 223 and estimates the degradation of the secondary battery to be diagnosed.

Description

Technical Field

[0001] The present invention relates to a secondary battery degradation prediction system and a secondary battery degradation prediction method.

Background Art

[0002] As a method for predicting the degradation of a secondary battery, there is a technique described in Patent Document 1. In Patent Document 1, a plurality of first log data indicating the state of a secondary battery during charging or discharging, acquired from a device equipped with a rechargeable battery (secondary battery), and a plurality of first degradation degrees indicating the degradation degree of the secondary battery calculated by a degradation degree estimation method using each first log data are acquired. Based on the content of a first charge / discharge that is a charge or discharge corresponding to each first log data, a reliability indicating the likelihood of each first degradation degree corresponding to each first log data is calculated, and a first learned model is generated by machine learning the relationship between the plurality of first degradation degrees and the plurality of first log data. A second learned model is generated by machine learning the relationship between a first degradation degree among the plurality of first degradation degrees whose reliability is equal to or higher than a predetermined value and the first log data corresponding to the first degradation degree. The estimation accuracy of the degradation degree of the secondary battery in each of the first learned model and the second learned model is evaluated, and it is disclosed that a learned model evaluated to have the best estimation accuracy among the first learned model and the second learned model is output.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in Patent Document 1, it is not assumed that the time-series data of the acquired information is short, and there is a risk that the accuracy of the degradation degree will decrease when the time-series data is short.

[0005] Therefore, the present invention provides a secondary battery degradation prediction system and a secondary battery degradation prediction method that can predict degradation with high accuracy even when the time series data of the secondary battery is short or the usage period is short. [Means for solving the problem]

[0006] To solve the above problems, the secondary battery degradation prediction system according to the present invention is a secondary battery degradation prediction system that predicts the degradation of a secondary battery to be diagnosed based on usage history information of other secondary batteries, and is characterized by comprising: a model construction unit that constructs a learning model using characteristic data of other secondary batteries and features that are finally determined in the learning process as input variables; a feature acquisition unit that acquires features that are finally determined in the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed; and a degradation prediction unit that inputs the usage history information of the secondary battery to be diagnosed and the features that are finally determined in the learning process into the learning model and estimates the degradation of the secondary battery to be diagnosed.

[0007] Furthermore, the secondary battery degradation prediction method according to the present invention is a secondary battery degradation prediction method that predicts the degradation of a secondary battery to be diagnosed based on usage history information of other secondary batteries, and is characterized by comprising: a step in which a model construction unit constructs a learning model using characteristic data of other secondary batteries and features finalized during the learning process as input variables; a step in which a feature acquisition unit acquires features finalized during the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed; and a step in which a degradation prediction unit inputs the usage history information of the secondary battery to be diagnosed and the features finalized during the learning process into the learning model and estimates the degradation of the secondary battery to be diagnosed. [Effects of the Invention]

[0008] According to the present invention, it is possible to provide a secondary battery degradation prediction system and a secondary battery degradation prediction method that can predict degradation with high accuracy even when the time series data of the secondary battery is short or the usage period is short. Other issues, configurations, and effects not mentioned above will be clarified by the following description of the embodiments. [Brief explanation of the drawing]

[0009] [Figure 1] This is a diagram illustrating the overall system configuration according to Embodiment 1 of the present invention. [Figure 2] This is a diagram illustrating an example of a learning model. [Figure 3] This is a flowchart showing the processing flow for degradation prediction. [Figure 4] This flowchart shows the processing flow for building a learning model using characteristic data of secondary batteries other than the secondary battery you want to diagnose. [Figure 5] This flowchart shows the processing flow for acquiring feature quantities of a secondary battery to be diagnosed, using the characteristic data of the secondary battery to be diagnosed. [Figure 6] This flowchart shows the processing flow for predicting the degradation of a secondary battery using its characteristic data. [Figure 7] This figure shows an example of a screen displaying the predicted results on a vehicle equipped with a secondary battery and an output device. [Figure 8] This is an example of a schematic graph showing the results of obtaining the final determined feature quantities during the learning process of secondary batteries other than the secondary battery being diagnosed. [Figure 9] This is a schematic diagram showing the route operation status used for creating a route plan according to Embodiment 2 of the present invention. [Figure 10] This flowchart shows the processing flow for predicting bus deterioration when the route is changed midway through the journey. [Modes for carrying out the invention]

[0010] In this specification, "features that are finally determined during the learning process" refers to the PB (Parametric Bias) value, and is described as "features that are finally determined during the learning process (PB value)." Furthermore, in this specification, "secondary battery" may refer to, for example, a lithium-ion battery, a lead-acid battery, a sodium battery, a fluoride battery, a magnesium battery, a redox flow battery, and the like. The following describes embodiments of the present invention with reference to the drawings. [Examples]

[0011] Figure 1 is a configuration diagram showing an overview of the entire system according to Embodiment 1 of the present invention. The secondary battery-equipped vehicle 1 and the secondary battery degradation prediction system 2 are connected via a network 3. The secondary battery-equipped vehicle 1 includes a communication unit 11, a memory 12, and a rechargeable secondary battery 13. The communication unit 11 is an electronic circuit that sends and receives information to and from the secondary battery degradation prediction system 2 via the network 3. The information to be transmitted includes specification information and usage history information regarding the secondary battery 13. The specification information includes at least one of the following: the number of cells in the secondary battery, rated capacity, size, and weight. The usage history information includes at least one of the following: time, battery current, battery voltage, and battery temperature (or temperature of the ambient environment around the battery). The memory 12 is a storage device capable of storing information, such as ROM or RAM. The secondary battery-equipped vehicle 1 is, for example, an electric vehicle, bus, or train equipped with a secondary battery 13. Alternatively, instead of the secondary battery-equipped vehicle 1, a stationary storage battery equipped with a secondary battery 13 may be used, not limited to a vehicle.

[0012] The secondary battery degradation prediction system 2 comprises a communication unit 21, a calculation unit 22, and a memory 23. The communication unit 21 is an electronic circuit that transmits and receives information to and from the vehicle 1 equipped with the secondary battery via the network 3. The information transmitted is degradation information about the secondary battery 13 calculated by the calculation unit 22. The calculation unit 22 comprises an input data conversion unit 221, a model construction unit 222, a learning model 223, a feature acquisition unit 224, and a degradation prediction unit 225. Here, the input data conversion unit 221, the model construction unit 222, the learning model 223, the feature acquisition unit 224, and the degradation prediction unit 225 are implemented using, for example, a processor such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a GPGPU (General-purpose computing on graphics processing units) (not shown), a ROM for storing various programs, a RAM for temporarily storing data during the calculation process, and an external storage device. The processor, such as the CPU, reads and executes the various programs stored in the ROM, and stores the calculation results, which are the execution results, in the RAM or external storage device.

[0013] The input data conversion unit 221 converts the information received from the communication unit 11 of the secondary battery-equipped vehicle 1 into characteristic data and stores it in the database of memory 23. Details of the data conversion will be described later. Characteristic data is time-series data that represents the characteristics of each charge and discharge cycle, or time-series data that represents the degradation of the secondary battery. Examples include SOC (State of Charge), discharge current value, time derivative of charging voltage, and maximum charge / discharge capacity. The received data does not necessarily have to be converted by the input data conversion unit 221; it may be stored directly in the database. In that case, when constructing the learning model 223, the data stored in the database needs to be converted by the input data conversion unit 221.

[0014] The model construction unit 222 constructs a learning model 223 using the characteristic data converted by the input data conversion unit 221 of secondary batteries other than the secondary battery to be diagnosed and the feature amounts (PB values) finally determined in the learning process. For the construction of the learning model 223, the characteristic data at time t is used as the input, and the characteristic data at time t + 1 is output. The feature amounts (PB values) finally determined in the learning process are values obtained for each battery and are self-organized values corresponding to each secondary battery. When there is SOH (State of Health) representing the degree of deterioration in the characteristic data, the learning model 223 is constructed with SOH as the input. Also, when there is no SOH, it may be estimated from the usage history information and the learning model 223 may be constructed with SOH as the input, or the learning model 223 may be constructed with the capacity as the input instead of SOH. The input variables are changed according to the information that can be collected. However, the input variables used in the model construction unit 222, the input variables used in the feature amount acquisition unit 224, and the input variables used in the deterioration prediction unit 225 are common. The learning model 223 is a predictor that can handle time-series data capable of predicting from time t to time t + 1 constructed by the model construction unit 222. Here, the input variables are SOC, discharge current, time derivative of charge voltage, maximum discharge capacity, secondary battery temperature (or temperature of the surrounding environment of the secondary battery 13), etc.

[0015] The feature amount acquisition unit 224 acquires the feature amounts (PB values) finally determined in the learning process of the secondary battery 13 to be diagnosed using the learned learning model 223. The deterioration prediction unit 225 predicts the deterioration of the secondary battery 13 to be diagnosed at future times after the final time obtained from the information of the secondary battery 13 to be diagnosed using the learning model 223 and the feature amounts (PB values) finally determined in the learning process estimated by the feature amount acquisition unit 224. The memory 23 is a storage device capable of storing information such as ROM and RAM, and can store the programs and databases of the arithmetic unit 22 (specification information, usage history information, characteristic data of each secondary battery). The output device 4 is a device that receives and displays the deterioration and remaining life predicted by the secondary battery deterioration prediction system 2 via the network 3.

[0016] In this embodiment, the example described is one in which specification information and usage history information regarding the secondary battery 13 are received from the memory 12 via the communication unit 11 and network 3 by the communication unit 21 constituting the secondary battery degradation prediction system 2, but this is not the only example. For example, the memory 12 may be taken out of the vehicle 1 equipped with the secondary battery, and the input data conversion unit 221 constituting the secondary battery degradation prediction system 2 may read the specification information and usage history information regarding the secondary battery 13 from the memory 12 and convert the data. Alternatively, information may be extracted via a wired connection (including when connected by an external cable, or via a portable storage medium, etc.).

[0017] Here, we will explain the features (PB values) that are ultimately determined during the learning process and the direction of PB value calculation. The PB value is a learnable input variable that represents the characteristics of each secondary battery common to all time points. These characteristics include, for example, degradation trends and individual differences. Having the PB value as an input variable makes it possible to handle multiple different time-series data patterns with a single learning model. Furthermore, it allows for learning multiple patterns with fewer learning parameters, thus improving learning efficiency. The method for calculating the PB value is updated at each time t to minimize the evaluation function, similar to the weight thresholds of the neurons that make up the neural network, during the construction of the learning model 223. The PB value may or may not be normalized. However, normalization is preferable for reasons such as improving the convergence speed of the learning model 223 and preventing overfitting. Examples of normalization functions include the sigmoid function and the ReLU (Rectified Linear Unit) function.

[0018] Figure 2 is a diagram illustrating an example of a learning model. Learning model 223 is a neural network model with an input layer, an output layer, and hidden layers. The input layer contains nodes representing the feature quantities (PB values) that are ultimately determined during the learning process. The neural network may be a recurrent neural network. While the hidden layer is shown as a single layer with four nodes, it may have two or more layers, and each layer may have one or more nodes.

[0019] Figure 3 is a flowchart showing the processing flow for degradation prediction. As shown in Figure 3, in step S1, the model building unit 222, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, constructs a learning model 223 using characteristic data of secondary batteries other than the secondary battery to be diagnosed and the feature quantity (PB value) that is finally determined during the learning process as input variables. In step S2, the feature acquisition unit 224, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, uses the learning model 223 obtained in step S1 to acquire the feature quantities (PB values) that are finally determined in the learning process of the secondary battery 13 to be diagnosed from the usage history information of the secondary battery to be diagnosed. In step S3, the degradation prediction unit 225, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, inputs the usage history information of the secondary battery to be diagnosed and the feature quantity (PB value) that is finally determined during the learning process obtained in step S2 into the learning model 223 to estimate the degradation of the secondary battery to be diagnosed.

[0020] Figure 4 is a flowchart showing the processing flow for building a learning model using characteristic data of secondary batteries other than the secondary battery to be diagnosed. Figure 4 is a detailed flow of step S1 in Figure 3. As shown in Figure 4, in step S11, the model building unit 222, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, divides the characteristic data converted by the input data conversion unit 221. In step S12, the model building unit 222, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, arbitrarily sets the initial values ​​of the feature quantities (PB values) that are finally determined during the learning process for each secondary battery data other than the secondary battery to be diagnosed. In step S13, the model building unit 222, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, uses the characteristic data and the initial values ​​of the feature quantities (PB values) that are finally determined during the learning process as input variables, and constructs a learning model 223 that predicts the characteristic data for the next time step. The learning model 223 is constructed by setting the feature quantities that are finally determined during the learning process and the parameters of the learning model 223 to minimize the evaluation function determined by the predicted value and the correct value.

[0021] In this way, by splitting the data, it is possible to artificially increase the amount of data with different initial states. Differences in initial states include differences in initial quality even among rechargeable batteries of the same model number, differences in initial states between rechargeable batteries of different models, and the fact that characteristic data is not available from the brand-new state (SOH=100%). By obtaining the final determined feature (PB value) for each split data set during the learning process, the differences in initial states can be incorporated into the features. The split can be made by dividing the entire period into two, three, or more parts. The number of splits can also be determined by the length of the time series of the usage history information, and some data may not be split. The initial value of the final determined feature (PB value) during the learning process can be 0 or a random number. Also, there may be one or more final determined features (PB values) during the learning process. Input variables may be not only characteristic data but also values ​​calculated based on usage history information (such as SOH and SOC). Output variables should be set to include the characteristic data and calculated values ​​of the input variables. The evaluation function in step S13 may use either the mean squared error or the mean absolute error. Furthermore, the minimization method may be either gradient descent or backpropagation.

[0022] Figure 5 is a flowchart showing the processing flow for acquiring feature quantities of a secondary battery to be diagnosed using characteristic data of the secondary battery to be diagnosed. Figure 5 is a detailed flow of step S2 in Figure 3. As shown in Figure 5, in step S21, the feature acquisition unit 224, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, arbitrarily sets the initial value of the feature (PB value) that is finally determined during the learning process for the secondary battery 13 to be diagnosed. In step S22, the feature acquisition unit 224, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, takes characteristic data and initial values ​​of the feature quantities (PB values) that are finally determined during the learning process as input variables, and uses the learning model 223, which uses the parameters of the learning model 223 determined in step S13 of Figure 4, to acquire the feature quantities (PB values) that are finally determined during the learning process of the secondary battery 13 to be diagnosed. The acquisition is performed in a manner that minimizes the evaluation function determined by the predicted value and the correct value.

[0023] In this way, by setting the feature quantities (PB values) that are finally determined during the learning process for the secondary battery 13 to be diagnosed, the characteristics of the secondary battery 13 to be diagnosed can be quantified, and degradation estimation corresponding to the secondary battery to be diagnosed can be performed. Even if the time series of characteristic data obtained from the secondary battery 13 to be diagnosed is short, by obtaining the feature quantities (PB values) that are finally determined during the learning process for the short time series of characteristic data, it is possible to estimate which secondary batteries other than the secondary battery to be diagnosed have similar characteristics, and predictions can be made while utilizing information on secondary batteries other than the secondary battery to be diagnosed. The initial value of the feature quantities (PB values) that are finally determined during the learning process may be 0, or may be given as random numbers. The number of feature quantities (PB values) that are finally determined during the learning process should be the same as the number set when constructing the learning model 223.

[0024] Figure 6 is a flowchart showing the processing flow for predicting the degradation of a secondary battery using its characteristic data. Figure 6 is a detailed flow of step S3 in Figure 3. As shown in Figure 6, in step S31, the degradation prediction unit 225, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, takes the characteristic data and the feature quantity (PB value) that is finally determined during the learning process of the secondary battery to be diagnosed as input variables, and uses the learning model 223, which uses the parameters of the learning model 223 determined in step S13 in Figure 4, to predict up to the final time of the characteristic data. In step S32, the degradation prediction unit 225, which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, predicts the final time for the characteristic data, and then uses the output characteristic data or the characteristic data to be input as input variables to predict degradation up to a specified future period.

[0025] In this way, by using predicted characteristic data (output variables) as input variables to predict future degradation, it is possible to predict the future degradation of characteristic data that does not have a time series. If you have characteristic data that you want to predict, you can also make predictions by using usage patterns as input variables instead of output variables.

[0026] Figure 7 shows an example of the screens displaying the predicted results on the secondary battery-equipped vehicle 1 and the output device 4. Screen 5 shows the screens on the secondary battery-equipped vehicle 1 and the output device 4 in Figure 1. As shown in Figure 7, the prediction result display unit 51, which serves as the first display area, displays a graph showing the relationship between the prediction result time and SOH. The solid line of the graph represents the SOH calculated from the usage history information, and the dashed line represents the predicted SOH. As an example, the time when SOH = 80% is displayed as the lifespan, but SOH = 70% is also acceptable, and it can be set arbitrarily. The lifespan display unit 52, which serves as the second display area, may display the period from the initial time until the lifespan is reached as the lifespan, or the period from the time of diagnosis until the lifespan is reached as the remaining lifespan. Alternatively, it may display the SOH at a set time. The feature quantity space display unit 53, which serves as the third display area, displays the feature quantities (PB values) that are finally determined during the learning process of secondary batteries other than the secondary battery to be diagnosed, which were used in building the learning model, and also displays the feature quantities (PB values) that are finally determined during the learning process of the secondary battery 13 to be diagnosed. This makes it possible to visually grasp secondary battery data with similar features. Next, the display unit as the fourth display area when there are two feature quantities is shown. The feature quantity 1 display unit 54, which serves as the fourth display area, is one of the features acquired in step S22 of Figure 5. The feature quantity 2 display unit 55, which serves as the fourth display area, is one of the features acquired in step S22 of Figure 5. The feature quantity 1 display unit 54 and the feature quantity 2 display unit 55 numerically represent the features that are finally determined during the learning process of the secondary battery to be diagnosed in the feature quantity spatial display unit 53. An example of displaying them as black circles in the feature quantity spatial display unit 53 is also shown. The predicted degradation results, lifespan, and features (PB values) that are finally determined during the learning process may be output to the secondary battery-equipped vehicle 1 in Figure 1, or they may be output to the output device 4.

[0027] The following describes a practical example of the secondary battery degradation prediction system 2 according to this embodiment. Using a database of training data containing multiple secondary battery model numbers and multiple usage history information (with different charge / discharge conditions, secondary battery temperature, etc.), a learning model 223 is constructed according to the flow in Figure 4. Next, following the flow in Figure 5, the feature quantity (PB value) that is finally determined during the learning process of the secondary battery to be diagnosed is obtained. Then, following the flow in Figure 6, the degradation of the secondary battery 13 to be diagnosed is predicted. If the battery model number and usage history information (hereinafter referred to as usage conditions) of the secondary battery to be diagnosed are sufficient to be learned in the database, the degradation can be predicted according to the flow in Figure 6 using the feature quantity (PB value) that is finally determined during the learning process obtained according to the flow in Figure 5. Furthermore, even if the usage conditions of the secondary battery to be diagnosed are not in the database or are not sufficient to be learned, the degradation can be predicted according to the flow in Figure 6 by using the feature quantity obtained according to the flow in Figure 5 to estimate from short time-series data what temperature conditions and characteristics are similar to those of other secondary batteries, thereby utilizing data on other secondary batteries and temperature conditions.

[0028] [Modified Example 1] In Example 1, a learning model 223 was constructed, and an example was described in which degradation could be predicted for a secondary battery under untrained usage conditions. In a modified version of Example 1, the handling of the feature quantities (PB values) that are finally determined during the learning process is described. In the following description, the same reference numerals are used for the same components as in Example 1, and their explanations are omitted.

[0029] Figure 8 is a schematic example of a graph showing the results of acquiring the final feature quantities determined during the learning process of secondary batteries other than the secondary battery to be diagnosed. Figure 8 is a schematic example of a graph showing the results of acquiring the final feature quantities (PB values) determined during the learning process of secondary batteries other than the secondary battery to be diagnosed, in the step of building the learning model 223 in step S1 of Figure 3. In this example, there are two final feature quantities (PB values) determined during the learning process, and four secondary battery model numbers were learned (secondary battery A, secondary battery B, secondary battery C, secondary battery D). The following shows what can be done by comparing this graph with the final feature quantities (PB values) determined during the learning process of the secondary battery to be diagnosed.

[0030] First, by graphing the feature quantities (PB values) that are ultimately determined during the learning process of secondary batteries other than the one to be diagnosed, it is possible to visually grasp the results of classifying the secondary batteries by feature quantity. These classification results may also be output to the secondary battery-equipped vehicle 1 or the output device 4. Next, by checking for the presence or absence of other secondary batteries that have similar characteristics to the secondary battery being diagnosed, it is possible to determine whether a prediction is possible. Next, if there are no secondary batteries with similar characteristics, we can determine what kind of features a secondary battery with similar characteristics would be desirable, and decide which data should be added to build the learning model 223.

[0031] As described above, this embodiment makes it possible to provide a secondary battery degradation prediction system and method that can predict secondary battery degradation with high accuracy even when the time series data of the secondary battery is short or the usage period is short. Furthermore, by obtaining the final determined feature quantities (PB values) for each secondary battery in the training data and for the secondary battery to be diagnosed during the learning process, it is possible to make predictions even for secondary batteries under untrained usage conditions by estimating which other secondary batteries share similar characteristics with the secondary battery to be diagnosed, thereby utilizing information from batteries other than the one being diagnosed.

[0032] Furthermore, if there are no secondary batteries with similar characteristics, it becomes possible to determine what kind of features a secondary battery with similar characteristics would be desirable, and to decide which data should be added to build the learning model 223. [Examples]

[0033] Figure 9 is a schematic diagram showing the route operation status used for creating an operation plan according to Embodiment 2 of the present invention. In Embodiment 1 described above, a learning model 223 was constructed, and an example was described in which degradation could be predicted for a secondary battery under unlearned operating conditions. In this embodiment, the degradation of the secondary battery to be diagnosed is predicted by changing the operating conditions midway through. In the following, the same reference numerals are used for the same components as in Embodiment 1, and their descriptions are omitted.

[0034] As shown in Figure 9, this embodiment demonstrates a practical example of applying the secondary battery degradation prediction system 2 to the creation of a vehicle operation plan. In Figure 9, for simplicity, we assume two routes, Route A and Route B, with buses a and b operating on each. We also assume that the operation plan is created in the 10th year, the route change occurs in the 15th year, and the operation plan is created up to the 30th year. Deterioration is predicted for 10-15 years under the assumption of continuous use with no change in usage, and for 15-30 years, predictions are made with the routes swapped. Following the flow in Figure 4, a learning model 223 is constructed using characteristic data that includes past operation history information for Routes A and B, excluding buses a61 and b63, which are to be predicted to deteriorate, or the usage history of buses operating on routes other than Routes A and B. Next, following the flow in Figure 5, the final determined feature quantities (PB values) are obtained using characteristic data 62 for bus a61 and characteristic data 64 for bus b63 during the learning process. Figure 10 is a flowchart showing the processing flow for predicting bus deterioration when the route is changed midway. Specifically, Figure 10 shows the procedure for predicting the deterioration of bus a61 when the route is changed midway, based on the flowchart. Bus b63 is predicted using the same procedure.

[0035] As shown in Figure 10, in step S41, the model building unit 222 (Figure 1), which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, constructs a learning model 223 that predicts the characteristic data for the next time step, using characteristic data of buses running on routes other than routes A and B, and the initial values ​​of the feature quantities (PB values) that are finally determined during the learning process as input variables. The feature quantity acquisition unit 224 (Figure 1), which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, constructs the learning model 223 by using the feature quantities (PB values) that are finally determined during the learning process and the parameters of the learning model 223 to minimize the evaluation function determined by the predicted value and the correct value.

[0036] In step S42, the degradation prediction unit 225 (Figure 1), which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, uses a learning model 223 that takes characteristic data 62 from when bus a61 operated on route A and the feature quantities (PB values) that are finally determined during the learning process of the acquired battery of bus a61 as input variables to predict up to the final time of the characteristic data. In step S43, the degradation prediction unit 225 (Figure 1), which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, uses the learning model 223 and the feature quantities (PB values) that are finally determined during the learning process of the battery of bus a61 as input variables to predict the degradation from the final time of the characteristic data until the change from route A to route B.

[0037] In step S44, the degradation prediction unit 225 (Figure 1), which constitutes the calculation unit 22 of the secondary battery degradation prediction system 2, after changing from route A to route B, uses the time-series data related to the operating conditions from the characteristic data of other vehicles that operated on route B (for example, bus b63) and the feature quantities (PB values) that are finally determined during the learning process of the acquired secondary battery of bus a61 as input variables to make predictions up to the planned end time.

[0038] In this way, at the time of switching routes, the time-series data related to operating conditions among the characteristic data used as input variables is replaced with time-series data related to operating conditions from the characteristic data of other vehicles that have previously operated on the new route, thereby allowing for the prediction of degradation when switching routes. Since the secondary battery remains installed in the bus when the route is switched, the features (PB values) that are finally determined during the learning process are not changed or updated and are used as they are. As an example, we used two routes, Route A and Route B, but predictions can be made even with multiple routes, and the learning model 223 is constructed using the battery characteristic data of buses that have operated on multiple routes.

[0039] By obtaining the final determined feature quantities (PB values) during the learning process for the secondary batteries of each bus in the training data and the secondary battery of the target bus to be diagnosed, it is possible to estimate which routes the buses operating on have similar characteristics, even when using the learning model 223 constructed with characteristic data that includes usage history information of buses operating on routes other than routes A and B. From the estimated route information, it is possible to predict the battery degradation of buses operating on routes A and B. Therefore, by using data from other routes, accurate predictions can be made without the need to prepare a large amount of characteristic data for routes A and B.

[0040] Furthermore, even if the secondary battery remains the same, a change in the route will alter the load pattern. In other words, a change in the route is equivalent to a change in the load pattern. And a change in the load pattern is equivalent to a change in the operating conditions described above.

[0041] As described above, this embodiment allows for prediction of changes in usage patterns midway through, in addition to the effects of Embodiment 1. Furthermore, by learning from operational information of routes other than the target route to be diagnosed, it becomes possible to reduce the amount of past operational data for routes A and B.

[0042] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are included. For example, the embodiments described above are explained in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. [Explanation of Symbols]

[0043] 1… Vehicles equipped with secondary batteries 11… Communications Department 12…Memory 13…Secondary battery 2…Secondary battery degradation prediction system 21... Communications Department 22...Arithmetic section 221...Input data conversion unit 222...Model Building Department 223...Learning Model 224...Feature extraction unit 225...Deterioration prediction section 23…Memory 3…Network 4…Output device 5…Screen 51...Prediction result display section 52...Life display section 53...Feature Spatial Display Unit 54…Feature Quantity 1 Display Section 55...Feature Quantity 2 Display Section 61... Bus A 62…Characteristic Data 63... Bus b 64…Characteristic Data

Claims

1. A secondary battery degradation prediction system that predicts the degradation of a secondary battery to be diagnosed based on the usage history information of other secondary batteries, A model building unit constructs a learning model using characteristic data of other secondary batteries and features that are finally determined during the learning process as input variables. A feature acquisition unit that obtains feature quantities that are ultimately determined during the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed, A secondary battery degradation prediction system comprising: a degradation prediction unit that inputs usage history information of a secondary battery to be diagnosed and a feature quantity finalized in the learning process into the learning model, and estimates the degradation of the secondary battery to be diagnosed.

2. A secondary battery degradation prediction system according to claim 1, The feature that is ultimately determined during the aforementioned learning process is a parametric bias value. The secondary battery degradation prediction system is characterized in that the aforementioned parametric bias value is a learnable input variable that represents the characteristics of each secondary battery common to all time points.

3. A secondary battery degradation prediction system according to claim 2, The secondary battery degradation prediction system is characterized in that the parametric bias value is updated at each time t to minimize the evaluation function.

4. A secondary battery degradation prediction system according to claim 3, The secondary battery degradation prediction system is characterized in that the characteristic data is time-series data showing the characteristics of the secondary battery for each charge and discharge, including discharge current or charge voltage, and is an input variable input to the learning model.

5. A secondary battery degradation prediction system according to claim 3, The aforementioned degradation prediction unit is characterized by estimating the degradation of a secondary battery to be diagnosed using the learning model, the feature quantities that are finally determined during the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed, and the usage history information of the secondary battery to be diagnosed.

6. A secondary battery degradation prediction system according to claim 3, The secondary battery degradation prediction system is characterized in that the degradation prediction unit estimates the similarity between batteries based on the feature quantities of each secondary battery other than the secondary battery to be diagnosed and the feature quantities of the secondary battery to be diagnosed, and estimates degradation based on the feature quantities of the secondary battery that has a high similarity to the secondary battery to be diagnosed.

7. A secondary battery degradation prediction system according to claim 6, The aforementioned degradation prediction unit is characterized by determining whether it can predict the secondary battery to be diagnosed based on the presence or absence of other secondary batteries that have similar characteristics to the secondary battery to be diagnosed.

8. A secondary battery degradation prediction system according to claim 3, A secondary battery degradation prediction system characterized by having an output device that outputs the result of predicting the degradation of a secondary battery to be diagnosed.

9. A secondary battery degradation prediction system according to claim 3, A secondary battery degradation prediction system characterized by using modified usage conditions as input variables and predicting degradation when the load pattern is changed during the operation of the secondary battery.

10. A secondary battery degradation prediction method that predicts the degradation of a secondary battery to be diagnosed based on the usage history information of other secondary batteries, The model building unit constructs a learning model using characteristic data from other secondary batteries and features that are ultimately determined during the learning process as input variables. The feature acquisition unit performs the step of acquiring the features that are ultimately determined during the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed, A secondary battery degradation prediction method characterized by comprising the steps of: a degradation prediction unit inputting usage history information of a secondary battery to be diagnosed and a feature quantity finalized in the learning process into the learning model, and estimating the degradation of the secondary battery to be diagnosed.

11. A method for predicting secondary battery degradation according to claim 10, The feature that is ultimately determined during the aforementioned learning process is a parametric bias value. A secondary battery degradation prediction method characterized in that the parametric bias value is a learnable input variable that represents the characteristics of each secondary battery common to all time points.

12. A method for predicting secondary battery degradation according to claim 11, A method for predicting secondary battery degradation, characterized in that the parametric bias value is updated at each time t to minimize the evaluation function.

13. A method for predicting secondary battery degradation according to claim 12, The method for predicting secondary battery degradation is characterized in that the characteristic data is time-series data showing the characteristics of the secondary battery for each charge and discharge cycle, including discharge current or charge voltage, and is an input variable input to the learning model.

14. A method for predicting secondary battery degradation according to claim 12, The degradation prediction unit is characterized by estimating the degradation of a secondary battery to be diagnosed using the learning model, the feature quantities that are finally determined during the learning process of the secondary battery to be diagnosed from the usage history information of the secondary battery to be diagnosed, and the usage history information of the secondary battery to be diagnosed.

15. A method for predicting secondary battery degradation according to claim 12, A secondary battery degradation prediction method characterized by using modified usage conditions as input variables and predicting degradation when the load pattern is changed during the operation of the secondary battery.

Images