Forecasting methods, forecasting programs, products, and information processing devices

Abstract

This invention provides a method for predicting the lifespan of lithium-ion batteries that eliminates the need for re-preparing learning data for machine learning even when materials and designs change, and where cycle life is represented as a probability distribution. The lithium-ion battery lifespan prediction method of this invention is executed by a computer as follows: Step (a) acquiring learning data including battery cycle measurement data and lifespan data; Step (b) applying the learning data to a lifespan prediction model for one or more prediction execution cycles, obtaining a set of learned lifespan prediction models corresponding to each prediction execution cycle; Step (c) sequentially acquiring prediction cycle measurement data for the battery as the prediction target up to the prediction execution cycle; and Step (d) inputting the prediction cycle measurement data acquired up to the prediction execution cycle into the learned lifespan prediction model for the corresponding prediction execution cycle, obtaining the probability distribution of lifespan for each prediction execution cycle as the output.

Description

Technical Field

[0001] This invention relates to a method for predicting the lifespan of lithium-ion batteries, a method for predicting discharge capacity retention, a lifespan prediction program product, a discharge capacity retention prediction program product, and an information processing device. Background Technology

[0002] In the research and development of lithium-ion batteries, the materials and design that make up the battery can be evaluated by measuring the cycle life of the battery cell through charge-discharge cycle testing. Figure 1 The 1A figure is a schematic diagram of cycle life obtained through charge-discharge cycle testing. Cycle life is obtained as the number of cycles before the discharge capacity retention rate falls below a threshold. A large portion of battery development time is spent on measuring cycle life. To accelerate battery development, a method is needed that can directly and early predict cycle life or predict cycle life early through discharge capacity retention rate.

[0003] Among them, as a method for directly and early predicting the cycle life of battery cells, a method using machine learning technology to predict cycle life based on the initial results of charge-discharge cycle tests is disclosed (Patent Documents 1 and 2). Furthermore, a method for predicting cycle life without performing charge-discharge cycle tests but using machine learning technology, which uses data including design factors, process factors, and formation factors determined during the design of the battery cell as learning data, is also disclosed (Patent Document 3).

[0004] [Cited Documents]

[0005] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2010-539473

[0007] [Patent Document 2] Japanese Patent Application Publication No. 2019-113524

[0008] [Patent Document 3] Japanese Patent Application Publication No. 2013-217897 Summary of the Invention

[0009] [Technical problem to be solved]

[0010] From the perspective of improving development efficiency, the cycle life prediction method used in the research and development of lithium-ion batteries needs to meet the following two requirements.

[0011] First, materials and designs frequently change during research and development. Each time such a change occurs, the learning data used for machine learning needs to be prepared anew. To eliminate this preparation work, the learning data should not depend on the materials and design that make up the battery.

[0012] Second, the current battery cells have a cycle life of thousands of cycles. Since the error in the prediction of such a long period is relatively large, it is desirable for the predicted cycle life to be in the form of a probability distribution.

[0013] The methods disclosed in Patent Documents 1 and 2 use measurement data such as voltage, current, and charge / discharge capacity acquired at the beginning of the charge / discharge cycle test as learning data. Since these methods do not contain information about the materials and design of the battery, they satisfy the first requirement mentioned above. However, the neural network used in the method disclosed in Patent Document 1 outputs a single value, and the method disclosed in Patent Document 2 does not specify a step for predicting cycle life in the form of a probability distribution. Therefore, neither of these methods satisfies the second requirement mentioned above.

[0014] Furthermore, regarding the method disclosed in Patent Document 3, since it uses data dependent on the materials and design constituting the battery as learning data, it does not meet the first requirement mentioned above.

[0015] Therefore, the object of the present invention is to provide a predictive technique that can improve the development efficiency of lithium-ion batteries.

[0016] [Technical Solution]

[0017] The present invention has the following configuration.

[0018] [1] A method for predicting the lifespan of a lithium-ion battery, characterized in that it is executed by a computer:

[0019] Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data;

[0020] Step (b): For more than one prediction execution loop, use the learning data to learn the lifetime prediction model and obtain a set of learned lifetime prediction models corresponding to each prediction execution loop.

[0021] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0022] Step (d) involves inputting the prediction cycle measurement data obtained up to the number of prediction execution cycles into the learned lifetime prediction model for the corresponding number of prediction execution cycles, and obtaining the probability distribution of lifetime for each prediction execution cycle as the output.

[0023] [2] The lithium-ion battery life prediction method as described in [1] is characterized in that,

[0024] The lifetime prediction model has

[0025] The data shaping unit shapes the cyclic measurement data into fixed-length data;

[0026] The feature extraction unit compresses the fixed-length data into compressed data; and

[0027] The nonlinear transformation unit converts the compressed data into nonlinear feature data mapped to a high-dimensional space to serve as nonlinear feature data.

[0028] It is connected to the regression unit in this order, wherein the regression unit takes the nonlinear feature data as input and outputs the probability distribution of lifetime.

[0029] [3] The lithium-ion battery life prediction method as described in [2] is characterized in that a dimension reduction method is used as the data compression method in the feature extraction unit.

[0030] [4] The life prediction method for lithium-ion batteries as described in [3] is characterized in that the dimension reduction method is principal component analysis.

[0031] [5] A method for predicting the discharge capacity retention rate of a lithium-ion battery, characterized in that it is executed by a computer:

[0032] Step (a): Acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle;

[0033] Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained.

[0034] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0035] Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime.

[0036] [6] A life prediction program for use by a computer:

[0037] Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data;

[0038] Step (b): For more than one prediction execution loop, use the learning data to learn the lifetime prediction model and obtain a set of learned lifetime prediction models corresponding to each prediction execution loop.

[0039] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0040] Step (d) involves inputting the prediction cycle measurement data obtained up to the number of prediction execution cycles into the learned lifetime prediction model for the corresponding number of prediction execution cycles, and obtaining the probability distribution of lifetime for each prediction execution cycle as the output.

[0041] [7] A discharge capacity maintenance prediction program for use by a computer:

[0042] Step (a): Acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle;

[0043] Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained.

[0044] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0045] Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime as the output.

[0046] [8] An information processing device, performing:

[0047] Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data;

[0048] Step (b): For more than one prediction execution loop, use the learning data to learn the lifetime prediction model and obtain a set of learned lifetime prediction models corresponding to each prediction execution loop.

[0049] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0050] Step (d) involves inputting the prediction cycle measurement data obtained up to the number of prediction execution cycles into the learned lifetime prediction model for the corresponding number of prediction execution cycles, and obtaining the probability distribution of lifetime for each prediction execution cycle as the output.

[0051] [9] An information processing device, performing:

[0052] Step (a): Acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle;

[0053] Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained.

[0054] Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles; and

[0055] Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime as the output.

[0056] [Beneficial Effects]

[0057] First, even if materials and design change, there is no need to re-prepare the learning data for machine learning. Second, the cycle life or discharge capacity maintenance rate is in the form of a probability distribution, which can suppress the impact of errors in long-term predictions.

[0058] In other words, according to the present invention, a predictive technique that can improve the development efficiency of lithium-ion batteries can be provided. Attached Figure Description

[0059] [ Figure 1 ] Figure 1 The 1A is a graph representing the relationship between discharge capacity retention and cycle life obtained through charge-discharge cycle testing. Figure 1 1B is a graph representing the discharge capacity retention rate up to the predicted number of cycles and the probability distribution of cycle lifetime obtained by using the combination of cycle measurements up to the predicted number of cycles and the present invention.

[0060] [ Figure 2 ] Figure 2 This is a flowchart of the lifetime prediction method according to the first embodiment of the present invention.

[0061] [ Figure 3 ] Figure 3 This is a diagram illustrating the composition of the cyclic measurement data and lifetime data used in the first embodiment of the present invention.

[0062] [ Figure 4 ] Figure 4 This is a schematic block diagram illustrating the configuration of the lifetime prediction model used in the first embodiment of the present invention.

[0063] [ Figure 5 ] Figure 5 This is a diagram illustrating the hardware configuration of an information processing device that executes the lifetime prediction model used in this invention.

[0064] [ Figure 6 ] Figure 6 This is a graph that selects and depicts a portion of the cyclic measurement data used in the embodiments.

[0065] [ Figure 7 ] Figure 7 7A is a graph depicting the current capacity relative to voltage in the constant current discharge step of the charge-discharge cycle before the processing of step (4) of the data shaping unit C-1. Figure 7 7B is a graph depicting the current capacity relative to voltage obtained by resampling at equally spaced voltage sampling points.

[0066] [ Figure 8 ] Figure 8 This is a graph representing the lifetime prediction results obtained in the embodiments.

[0067] [ Figure 9 ] Figure 9 It is a graph showing the discharge capacity retention rate up to the predicted number of execution cycles and the predicted discharge capacity retention rate after the predicted number of execution cycles.

[0068] [ Figure 10 ] Figure 10 This is a flowchart of the lifetime prediction method according to the second embodiment of the present invention.

[0069] [ Figure 11 ] Figure 11 This is a diagram showing the composition of the cyclic measurement data and discharge capacity maintenance rate data used in the second embodiment of the present invention.

[0070] [ Figure 12 ] Figure 12This is a schematic block diagram illustrating the configuration of the lifetime prediction and discharge capacity maintenance prediction models used in the second embodiment of the present invention. Detailed Implementation

[0071] The embodiments will now be described with reference to the accompanying drawings. To facilitate understanding, the same symbols are used to label the same components as much as possible in each figure, and repeated descriptions are omitted.

[0072] [First Implementation]

[0073] The lithium-ion battery life prediction method of the present invention includes: step (a) acquiring learning data (also called training data) including battery cycle measurement data and life data; step (b) applying the learning data to train the life prediction model for one or more prediction execution cycles (i.e., training the life prediction model) to obtain a set of learned life prediction models (also called trained life prediction models) corresponding to each prediction execution cycle number; step (c) sequentially acquiring prediction cycle measurement data of the battery as the prediction object up to the prediction execution cycle number; and step (d) inputting the prediction cycle measurement data acquired up to the prediction execution cycle number into the learned life prediction model of the corresponding prediction execution cycle number to obtain the probability distribution of life of each prediction execution cycle number as output.

[0074] Figure 2 This is a flowchart for providing a detailed explanation of steps (a) to (d), where step (a) corresponds to step S1, step (b) corresponds to steps S31 to S32, step (c) corresponds to step S41, and step (d) corresponds to step S42. The following is based on... Figure 2 The life prediction method for lithium-ion batteries of the present invention will be described.

[0075] like Figure 2As shown, the lithium-ion battery life prediction method of the present invention is executed according to the following steps: Step S1: acquiring learning cycle measurement data A and learning life data B; Step S2: determining one or more prediction execution cycle numbers ni (i = 1, 2, ...); Step S31: cutting out (extracting) initial ni cycle portions of measurement data Ani from the learning cycle measurement data A according to each of the one or more prediction execution cycle numbers ni determined in Step S2, and Step S32: applying the cut-out measurement data Ani and the learning life data B to learn the life prediction model C to obtain the learned life prediction model Cni; and Step S41: acquiring prediction cycle measurement data Dni up to the prediction execution cycle number ni according to each of the one or more prediction execution cycle numbers ni determined in Step S2, and Step S42: inputting the prediction cycle measurement data Dni into the learned life prediction model Cni to obtain the probability distribution Eni of the predicted life.

[0076] In step S1, charge-discharge cycle tests are performed on the cells of multiple batteries, and learning cycle measurement data A and learning lifespan data B are obtained. The number of cells used for learning data acquisition is preferably 50 or more. Furthermore, to obtain the learning lifespan data B, the charge-discharge cycle tests are preferably performed continuously until each cell reaches its cycle life.

[0077] Each cycle in the charge-discharge cycle test must include a constant current charging step and a subsequent constant current discharging step. A constant voltage charging step may also be included between the constant current charging and discharging steps. Furthermore, a pause (rest) step may be included between each charge-discharge step.

[0078] The cycle life is the number of cycles in which the discharge capacity falls below a predetermined (pre-determined) threshold. For example, this threshold is 80% of the discharge capacity measured in the first cycle.

[0079] Figure 3 The composition of cyclic measurement data and lifetime data is shown.

[0080] Cyclic measurement data is obtained by performing charge-discharge cycle tests on multiple battery cells and recording the sampling times t1, t2, ... and the physical property values ​​at these times during each cycle. The sampling time is preferably appropriately selected in a manner that adequately follows the changes in physical property values, and does not need to be a fixed interval. The recorded physical property values ​​must include the voltage applied to the battery cell and the current flowing through the battery cell, as well as measurable physical and chemical properties exhibited by at least one battery cell. Examples of recorded physical property values ​​include, for example, voltage, current, and current capacity.

[0081] Lifetime data refers to the cycle life of each cell from which the cycle measurement data was acquired.

[0082] In step S1, when acquiring the learning cycle measurement data A and the learning lifespan data B, if the charge-discharge cycle test has not been performed to reach the cycle life, the cycle life can be estimated (estimated) using a combination of the change (shift) in discharge capacity up to the completed charge-discharge cycle and a mathematical model or time series data analysis method. The estimated cycle life is then used as the learning lifespan data B. Examples of mathematical models include known root laws and power laws related to the capacity decay of lithium-ion batteries. Examples of time series data analysis methods include the Autoregressive Integrated Moving Average (ARIMA) model, Kalman filter, and Gaussian process regression.

[0083] The learning cycle measurement data A and learning lifetime data B obtained in step S1 can also be stock measurement data.

[0084] The predicted execution cycle number determined in step S2 is the number of charge-discharge cycles in which the charge-discharge cycle test is repeated until the predicted execution of the cell with the predicted lifespan, and the measurement data is accumulated. Multiple predicted execution cycle numbers can be selected. If a pre-prepared predicted execution cycle number is used, step S2 can be omitted.

[0085] For one or more predicted execution loop counts ni (i = 1, 2, ...) determined in step S2, repeat steps S31 and S32.

[0086] In step S31, the learning loop measurement data up to the predicted execution loop number ni is extracted from the learning loop measurement data A obtained by step S1 as learning loop measurement data Ani. If there are cells in the learning loop measurement data A and the learning lifetime data B whose measurements have ended before the predicted execution loop number ni, the data of that cell is excluded because it cannot be used.

[0087] In the next step S32, the cut-out learning cycle measurement data Ani is used as the explanatory variable, and the learning lifetime data B is used as the target variable, thereby enabling the lifetime prediction model C to learn and obtain the learned lifetime prediction model Cni that predicts the number of execution cycles ni.

[0088] In steps S31 and S32, a learned lifetime prediction model Cni is prepared, which has been trained on the cyclic measurement data up to the initial ni cycles, corresponding to the predicted execution cycle number ni. That is, when multiple prediction execution cycle numbers are selected, multiple lifetime prediction models Cni are trained for each cycle of the initial ni cycles corresponding to the predicted execution cycle number ni.

[0089] For the initial ni cycles, for example, in the case of the initial 200 cycles, this refers to cycles 1 through 200. The lifetime prediction model C200 is trained using the initial 200 cycles of learning cycle measurement data A200 (cycles 1 through 200) extracted from the learning cycle measurement data A, and the lifetime data B. The learned lifetime prediction model C200 outputs a probability distribution E200 for predicted lifetimes when the prediction cycle measurement data D200 up to the 200th prediction execution cycle is input.

[0090] Cyclic lifetime, as the target variable, is always a positive value. Therefore, in step S32, when training the lifetime prediction model C, the lifetime data B is not directly applied to train the lifetime prediction model C. Instead, the lifetime data B is preferably converted into the logarithm of B before training the lifetime prediction model C. In this way, the prediction result is also the logarithm of the lifetime, thereby ensuring that the predicted lifetime obtained through the inverse conversion is positive.

[0091] like Figure 4 As shown, the life prediction model used in the lithium-ion battery life prediction method of the present invention can be configured by sequentially connecting a data shaping unit C-1, a feature extraction unit C-2, a nonlinear transformation unit C-3, and a regression unit C-4. The life prediction model receives prediction cycle measurement data up to the prediction execution cycle number ni as input and outputs the probability distribution of lifespan. The elements constituting the life prediction model will be explained below as an example.

[0092] The data shaping unit C-1 receives the cyclic measurement data up to the predicted execution cycle number ni as input, and outputs fixed-length data in matrix form with the number of rows equal to the number of cells and the number of columns equal to the fixed length. The fixed-length data in matrix form can be generated by shaping the cyclic measurement data of each cell into data with a fixed number of columns by the shaping processes (1) to (7) below, and overlapping them along the row direction.

[0093] (1) The shaped data may include physical property values ​​at specific time points of charge-discharge cycle testing and values ​​calculated using these values. Items included in the shaped data may include, for example: charging start voltage of the first cycle, charging start voltage of the second cycle, ..., charging start voltage of the ni-th cycle, charging time of the first cycle, charging time of the second cycle, ..., charging time of the ni-th cycle, charging capacity of the first cycle, charging capacity of the second cycle, ..., charging capacity of the ni-th cycle, discharging capacity of the first cycle, discharging capacity of the second cycle, ..., discharging capacity of the ni-th cycle, initial coulombic efficiency, slope and intercept of the near-near number of cycles of the ni-th cycle – charging start voltage curve, slope and intercept of the near-near number of cycles of the ni-th cycle – charging time curve, slope and intercept of the near-near number of cycles of the ni-th cycle – charging capacity curve, and slope and intercept of the near-near number of cycles of the ni-th cycle – discharging capacity curve. These are all single values, and each item has a length of 1.

[0094] (2) The shaped data may include physical property values ​​at specific time points in each charge-discharge cycle test. The values ​​included in the shaped data may include, for example, the voltage at the start of charging for each cycle, the charging time for each cycle, the charging capacity for each cycle, and the discharging capacity for each cycle. In this case, the length of each item is the predicted number of cycles, ni.

[0095] (3) The data from the constant current charging step of the charge-discharge cycle is treated as a function of voltage and processed. If a property value shows a meaningful change relative to voltage, it is resampled at a predetermined number of voltage resampling points. By performing this process on all cycles, a property can be converted into data with a length of "predicted number of cycles ni × number of voltage resampling points". Current capacity can be cited as an example of a property value showing a meaningful change relative to voltage.

[0096] (4) The data for the constant current discharge step of the charge-discharge cycle is processed in the same way as the data for the constant current charging step.

[0097] (5) In cases where a constant voltage charging step exists within a charge-discharge cycle, the data is treated as a function of current and processed. If a property value shows a meaningful change relative to the current, resampling is performed at a predetermined number of current resampling points. By performing this process on all cycles, for a property, it can be converted into data with a length of "predicted number of execution cycles ni × number of current resampling points". Time can be cited as an example of a property value showing a meaningful change relative to the current.

[0098] (6) If there is a pause step in the charge / discharge cycle, delete the data.

[0099] (7) Connect the values ​​of the items obtained in steps (1) to (6) above along the column direction to generate a row of fixed-length data in matrix form.

[0100] In the resampling method performed by the data shaping unit C-1, the simplest linear interpolation can be used, or polynomial interpolation of quadratic or higher, spline interpolation, etc. can be used.

[0101] Feature extraction unit C-2 receives fixed-length data output from data shaping unit C-1 as input, compresses the fixed-length data into compressed data using a data compression method, and then outputs it. Examples of data compression methods include dimensionality reduction methods, with Principal Component Analysis (PCA) being a preferred method. When using PCA, it is preferable to compress the data for each item connected along the column direction within the matrix-form data. However, items with a column length of 1 are not compressed and are left as is. PCA extracts principal components from the fixed-length data. The number of principal components extracted needs to be predetermined for each item. The number of principal components extracted is preferably chosen to be large enough that the cumulative contribution rate of the extracted principal components is 90% or more.

[0102] The nonlinear transformation unit C-3 receives the compressed data output from the feature extraction unit C-2 as input, applies a nonlinear transformation to convert it into nonlinear feature data mapped to a high-dimensional space, and then outputs it. The subsequent regression unit C-4 receives the nonlinear feature data as input and outputs it as a probability distribution of lifetimes.

[0103] Although the methods used in the nonlinear transformation section C-3 and the regression section C-4 can be set separately, it is preferable to apply Gaussian process regression, which has two functions, to C-3 and C-4. When set separately, the kernel method is preferably used for the nonlinear transformation section C-3. The output of the regression section C-4 should be a probability distribution; in addition to Gaussian process regression, Bayesian ridge regression, etc., can also be used.

[0104] In cases where Gaussian process regression is applied in nonlinear transformation section C-3 and regression section C-4, and in cases where kernel methods are applied in nonlinear transformation section C-3, nonlinear kernels such as radial basis function kernels can be used to achieve nonlinear transformation.

[0105] By repeatedly executing steps S31 and S32, a set of learned lifetime prediction models Cni corresponding to one or more prediction execution loop counts ni (i = 1, 2, ...) can be obtained. Then, for these one or more ni and Cni, steps S41 and S42 as described below are repeated.

[0106] In step S41, charge-discharge cycle tests are performed on the cells of the battery with the predicted lifespan up to the predicted number of cycles ni, obtaining the predicted cycle measurement data Dni. The composition of the physical property values ​​of the predicted cycle measurement data Dni includes the composition of the physical property values ​​of the learned cycle measurement data Ani.

[0107] In the next step S42, the prediction cycle measurement data Dni is input into the learned lifetime prediction model Cni, which corresponds to the prediction execution cycle number ni, to obtain the lifetime probability distribution Eni as the output. If the prediction cycle measurement data Dni contains items with physical property values ​​that are not in the learned cycle measurement data Ani, these items are removed from Dni before being input into the lifetime prediction model Cni.

[0108] Steps S41 and S42 can be repeated for more than one prediction execution cycle number ni. However, if the implementer determines, based on the lifetime prediction results obtained from the earlier prediction execution cycle number, that sufficient confidence has been obtained regarding the lifetime of the target cell, subsequent predictions based on the prediction execution cycle number ni can be omitted. In this case, the implementer can terminate the charge-discharge cycle test of the target cell early.

[0109] Figure 5 The hardware configuration of the information processing device used to execute the above-described lifetime prediction model is shown. For example... Figure 5 As shown, the information processing device 500 includes a processor 501, a memory 502, an auxiliary storage device 503, an I / F (Interface) device 504, a communication device 505, and a drive device 506. It should be noted that the various hardware components of the information processing device 500 are interconnected via a bus 507.

[0110] The processor 501 has various computing devices such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 501 can load various programs (e.g., a life prediction program) into the memory 502 and execute them.

[0111] The memory 502 has main storage devices such as ROM (Read Only Memory) and RAM (Random Access Memory). The processor 501 and the memory 502 form a so-called computer, which can perform the above-mentioned functions by executing various programs read from the memory 502 by the processor 501.

[0112] The auxiliary storage device 503 can store various programs and various data used by the processor 501 when the programs are executed.

[0113] I / F device 504 is a connection device for connecting operating device 511 and display device 512 to information processing device 500. Communication device 505 is a communication device for communicating with an external device (not shown) via a network. Drive device 506 is a device for inserting storage medium 513.

[0114] It should be noted that the various programs installed in the auxiliary storage device 503 can be installed, for example, by inserting the distributed storage medium 513 into the drive device 506, and having the drive device 506 read the various programs stored in the storage medium 513 for installation. Alternatively, the various programs installed in the auxiliary storage device 503 can also be installed by downloading them from the network via the communication device 505.

[0115] [Example]

[0116] The embodiments based on this implementation method will be described below.

[0117] In this embodiment, step S1 involves performing charge-discharge cycle tests on 104 lithium-ion battery cells, obtaining learning cycle measurement data A and learning lifespan data B. The charge-discharge cycle test consists of five steps: constant current charging, constant voltage charging, rest, constant current discharging, and rest. As physical property values ​​included in the cycle measurement data, three items—the voltage applied to the cell, the current flowing through the cell, and the current capacity—were recorded (and saved).

[0118] Figure 6 This is a graph plotting a portion of the cyclic measurement data used in the embodiments. Specifically, data from the start of the first cycle to the end of the second cycle for a specific cell was extracted from the learning cyclic measurement data A obtained in the embodiments, and plotted with time on the horizontal axis and voltage, current, and current capacity on the vertical axis. Figure 6As shown, in the charge-discharge cycle test of this embodiment, the voltage range is 2.8V to 4.2V, and the current range is -50mA to 50mA. Therefore, the curve is a smooth curve or a straight line within each step, and the sampling time is appropriately selected to fully follow the changes in physical property values.

[0119] In this embodiment, the number of cycles in which the discharge capacity of a battery cell is less than 80% of that of the first cycle is defined as cycle life. For the charge-discharge cycle tests used to obtain learning cycle life data B in step S1, the shortest cycle life was 2,000 cycles, and the longest was 4,000 cycles. Furthermore, for some cells that did not reach their cycle life, the cycle life was estimated using a combination of the shift in discharge capacity obtained as a measurement and a power law. The estimated cycle life was used as learning cycle life data B, provided that the root mean square error of the curve fitting based on the power law was less than 0.001 mAh. Therefore, the number of battery cells included in the learning cycle measurement data A and the learning cycle life data B in this embodiment is 70, and the cycle life range in the learning cycle life data B is approximately 500 to 6,000 cycles.

[0120] As the predicted execution loop number ni determined in step S2, in this embodiment, 10 loops were selected in units of 100 loops within the range of the 100th loop to the 1,000th loop.

[0121] Steps S31 and S32 were repeatedly executed for the 10 predicted execution loop counts ni determined in step S2. Here, in step S31, the cyclic measurement data from the learning cyclic measurement data A to the predicted execution loop count ni was extracted as the learning cyclic measurement data Ani. In step S32, the extracted learning cyclic measurement data Ani was used as the explanatory variable, the data after converting the learning lifetime data B into logarithmic data was used as the target variable, and the lifetime prediction model C was trained to obtain the learned lifetime prediction model Cni for the predicted execution loop count ni.

[0122] The details of the lifetime prediction model in this embodiment are as follows.

[0123] The data shaping unit C-1 receives the cyclic measurement data up to the predicted execution cycle number ni as input, and outputs fixed-length data in matrix form with the number of rows equal to the number of cells and the number of columns equal to the number of columns. In this embodiment, the fixed-length data can be formed by shaping the cyclic measurement data of each cell into data with a fixed number of columns by the shaping processes (1) to (7) below, and overlapping them along the row direction.

[0124] (1) The physical property values ​​at specific time points during charge-discharge cycle testing and the values ​​calculated using these physical property values ​​include the following 33 items: the charging start voltage of the first cycle, the charging start voltage of the second cycle, the charging start voltage of the ni cycle, the total charging time of the first cycle, the total charging time of the second cycle, the total charging time of the ni cycle, the constant voltage charging time of the first cycle, the constant voltage charging time of the second cycle, the constant voltage charging time of the ni cycle, the total charging capacity of the first cycle, the total charging capacity of the second cycle, the total charging capacity of the ni cycle, the constant voltage charging capacity of the first cycle, the constant voltage charging capacity of the second cycle, and the constant voltage charging capacity of the ni cycle. The discharge capacity of the first cycle, the discharge capacity of the second cycle, the discharge capacity of the nith cycle, the discharge capacity maintenance rate of the nith cycle, the slope and intercept of the voltage curve at the start of charging (near the nith cycle), the slope and intercept of the total charging time curve (near the nith cycle), the slope and intercept of the constant voltage charging time curve (near the nith cycle), the slope and intercept of the total charging capacity curve (near the nith cycle), the slope and intercept of the constant voltage charging capacity curve (near the nith cycle), the slope and intercept of the discharge capacity curve (near the nith cycle), and the slope and intercept of the discharge capacity maintenance rate curve (near the nith cycle). These values ​​are all single values, and each item has a length of 1.

[0125] (2) The physical property values ​​at specific time points in each charge-discharge cycle test include the following seven items: the charging start voltage of each cycle, the total charging time of each cycle, the constant voltage charging time of each cycle, the total charging capacity of each cycle, the constant voltage charging capacity of each cycle, the discharge capacity of each cycle, and the discharge capacity maintenance rate of each cycle. The length of each item is the predicted number of cycles, ni.

[0126] (3) The current capacity data in the constant current charging step of the charge-discharge cycle is regarded as a function of voltage. In this embodiment, 101 voltage samples are taken at equal intervals within the voltage range of 2.8V to 4.2V, and these sample points are resampled by linear interpolation. By performing this process on all cycles, the data is converted into data with a length of "predicted number of execution cycles ni × number of voltage resampling points 101".

[0127] (4) The current capacity data in the constant current discharge step of the charge-discharge cycle was processed in the same way as the current capacity data in the constant current charging step.

[0128] (5) The current capacity and time data in the constant voltage charging step of the charge-discharge cycle are considered as functions of current. In this embodiment, 101 current samples are taken at equal intervals within the current range of -50mA to 50mA, and these sample points are resampled by linear interpolation. By performing this process on all cycles, the current capacity and time are converted into data with a length of "predicted number of execution cycles ni × number of voltage resampling points 101".

[0129] (6) Data for the pause steps of the charge-discharge cycle has been removed.

[0130] (7) Connect the 33 items of length 1, the 7 items of length ni, and the 4 items of length ni×101 obtained by the shaping process of (1) to (6) above along the column direction to generate a row of fixed-length data in matrix form.

[0131] Figure 7 7A is a graph depicting the current capacity relative to voltage during the constant current discharge step of the charge-discharge cycle before the data shaping process (4) of the data shaping unit C-1. The three curves are based on the current capacity of different cycle numbers, and the number and sampling position of voltage samples are different between the curves. Figure 7 7B is a response to... Figure 7 The current capacity is plotted relative to voltage by performing shaping (4) on the data shown in 7A and resampling at a common voltage sampling point.

[0132] The feature extraction unit C-2 receives fixed-length data output from the data shaping unit C-1 as input, applies a data compression method to it to compress the data, and then outputs it. In this embodiment, principal component analysis (PCA) is used as the data compression method, and PCA is applied to each item connected along the column direction within the fixed-length data in matrix form, thereby obtaining compressed data. At this time, for the 33 items with a column length of only 1, PCA is not applied, and they are left as is. The number of principal components extracted by PCA is selected to be sufficiently large so that the cumulative contribution rate of the extracted components is greater than 90%. Specifically, for the 7 items with a column length of ni (the number of predicted execution loops), 10 principal components are extracted respectively, and for the 4 items with a column length of ni × 101 (the number of predicted execution loops), 20 principal components are extracted respectively.

[0133] The nonlinear transformation unit C-3 receives the compressed data output from the feature extraction unit C-2 as input, applies a nonlinear transformation to it to convert it into nonlinear feature data, and then outputs it. The subsequent regression unit C-4 receives the nonlinear feature data as input and outputs it as a probability distribution of lifetimes. In this embodiment, the method used in the nonlinear transformation unit C-3 and the regression unit C-4 employs a Gaussian process regression with two functions, using the sum of the radial basis function kernel, the constant kernel, and the white kernel as the kernel.

[0134] By repeatedly executing steps S31 and S32, a set of learned lifetime prediction models Cni corresponding to 10 prediction execution loop counts ni is obtained. In this embodiment, as described below, steps S41 and S42 are repeatedly executed for these 10 ni and Cni.

[0135] In step S41 of this embodiment, charge-discharge cycle tests are performed on the two cells whose lifespan is predicted up to the predicted number of execution cycles ni, thereby obtaining the prediction cycle measurement data Dni. The structure of the prediction cycle measurement data Dni is the same as the structure of the learning cycle measurement data Ani in this embodiment.

[0136] In the next step S42, the predicted cyclic measurement data Dni is input into the learned lifetime prediction model Cni with the corresponding prediction execution cycle number ni, thereby obtaining the probability distribution Eni of the logarithm of the lifetime as the output.

[0137] Figure 8 The graph shown was obtained by calculating the mean and 95% confidence interval based on the probability distribution Eni of the logarithm of the obtained lifetime, performing an inverse logarithmic transformation, and plotting the graphs for the two cells separately. "□" represents the mean lifetime, and the vertical lines represent the 95% confidence interval of the lifetime. For the two cells used as the prediction targets, in... Figure 8 8A and Figure 8 They are represented separately in 8B. and Figure 8 The 8A-rated battery cell reached its lifespan after the 668th charge-discharge cycle test; therefore, predictions were not made after the 700th cycle. The 668 cycles represent the actual lifespan. Figure 8 8A is represented by a dashed line. (And...) Figure 8The 8B-corresponding cell did not reach its cycle life at the final predicted cycle count of 1,000. To verify this, charge-discharge cycle tests were subsequently conducted up to the 1,400th cycle, and the actual cycle life was estimated to be 4,650 cycles using the obtained discharge capacity and a power law combination. The root mean square error of the curve fitting was less than 0.001 mAh. The estimated actual cycle life of 4,650 cycles obtained through this estimation... Figure 8 The 8B is represented by a dashed line.

[0138] It can be seen that, Figure 8 8A and Figure 8 In section 8B, the average predicted lifetime "□" is close to the actual or estimated actual lifetime of the target cell. Regarding the predicted 95% confidence interval, it converges to the actual or estimated actual lifetime as the number of prediction execution cycles increases. If the implementer determines that the 95% confidence interval has become sufficiently narrow, subsequent predictions for the number of prediction execution cycles can be omitted to end the charge-discharge cycle test early.

[0139] [Second Implementation]

[0140] In the first embodiment described above, the probability distribution of the predicted lifetime was explained as a prediction item based on the predicted cycle measurement data. However, the prediction item is not limited to the probability distribution of lifetime; for example, it may be configured to further predict the discharge capacity maintenance rate in each cycle from the start of the prediction execution cycle until the predicted lifetime is reached, or until a certain point in time exceeding the predicted lifetime. The second embodiment will now be described, and it should be noted that the second embodiment will be described below focusing on the differences from the first embodiment described above.

[0141] First, the new prediction item in the second embodiment, namely the discharge capacity maintenance rate in each cycle, will be explained. Figure 9 This is a schematic diagram of the discharge capacity maintenance rate up to the predicted number of execution cycles and the predicted discharge capacity maintenance rate after the predicted number of execution cycles.

[0142] and Figure 1 same, Figure 9 The horizontal axis represents the number of cycles, and the vertical axis represents the discharge capacity retention rate.

[0143] Figure 9 In the figure, symbol 910 represents the curve of the prediction cycle measurement data obtained up to the number of prediction execution cycles.

[0144] In addition, symbol 920 represents the plotting point of the predicted discharge capacity maintenance rate at any number of cycles after the predicted execution cycle (e.g., cycle number = 200, 300, 400, etc.) and the broken line connecting each plotting point.

[0145] In addition, symbols 921 and 922 represent the connecting line between the upper and lower limits of each 95% confidence interval when calculating the predicted discharge capacity maintenance rate based on any number of cycles after the predicted execution cycle.

[0146] In this way, by representing the predicted discharge capacity maintenance rate in each cycle during the period up to the predicted lifetime, the implementer can grasp the shift (change) of the predicted discharge capacity maintenance rate up to the predicted lifetime.

[0147] It should be noted that, Figure 9 In the example, the mean of the lifetime (symbol 930) and the 95% confidence interval of the lifetime (symbol 931) are represented together as the probability distribution of the lifetime.

[0148] Next, the flow of the lifetime prediction method of the second embodiment will be described. Figure 10 This is a flowchart of the lifetime prediction method according to the second embodiment of the present invention. Compared with the method described in the first embodiment, this method is based on... Figure 2 The difference in the flowcharts described lies in steps S12, S33, and S43.

[0149] In step S12, the discharge capacity retention rate B' obtained in step S11 by continuously performing charge-discharge cycle tests until the cycle life is reached is calculated.

[0150] In step S33, the cut-out (extracted) learning cycle measurement data Ani is used as the explanatory variable, the discharge capacity maintenance rate B' is used as the target variable, and the discharge capacity maintenance rate prediction model C' is trained to obtain the learned discharge capacity maintenance rate prediction model C'ni that predicts the number of execution cycles ni.

[0151] In steps S31 and S33, a learned discharge capacity maintenance prediction model C'ni is prepared, which has been trained on the cyclic measurement data up to the initial ni cycles corresponding to the predicted execution cycle number ni. That is, when multiple prediction execution cycle numbers are selected, multiple discharge capacity maintenance prediction models C'ni are trained for each cycle of the initial ni cycles corresponding to the predicted execution cycle number ni.

[0152] As described above, for the initial ni cycles, for example, if it is the initial 200 cycles, it refers to cycles 1 to 200. The discharge capacity maintenance prediction model C'200 is trained by applying the initial 200 cycles of training cycle measurement data A200 (cycles 1 to 200) cut from the training cycle measurement data A and the capacity maintenance rate B' of each cycle. After inputting the prediction cycle measurement data D200 up to the prediction execution cycle number 200 into the trained discharge capacity maintenance prediction model C'200, the trained discharge capacity maintenance prediction model C'200 can output the probability distribution E'200 of the predicted discharge capacity maintenance rate.

[0153] It should be noted that the predicted discharge capacity maintenance rate, which is the target variable, is usually greater than zero and less than the value of the discharge capacity maintenance rate at the predicted number of execution cycles. Therefore, when the discharge capacity maintenance rate prediction model C' is trained in step S33, it is not trained directly on the discharge capacity maintenance rate B', but rather preferably trained on the value transformed by the following equation (1). In this way, since the predicted discharge capacity maintenance rate is also a value transformed by the following equation (1), it can be ensured that the predicted discharge capacity maintenance rate obtained by the inverse transformation is greater than zero and less than the value of the discharge capacity maintenance rate at the predicted number of execution cycles.

[0154] Equation (1): Transformation value = log(ymi / (yni-ymi))

[0155] Here, ymi is the predicted discharge capacity retention rate at the number of cycles mi to be predicted, and yni is the predicted discharge capacity retention rate at the number of cycles ni to be predicted.

[0156] In step S43, the predicted cycle measurement data Dni is input into the learned discharge capacity maintenance rate prediction model C'ni for the corresponding prediction execution cycle number ni, thereby obtaining the probability distribution E'ni of the discharge capacity maintenance rate for each cycle as the output.

[0157] Furthermore, based on the probability distribution E'ni of the obtained discharge capacity maintenance rate transformed by Equation (1), the predicted discharge capacity maintenance rate and 95% confidence interval were calculated, and then inversely transformed to obtain the following results: Figure 9 The graph shown.

[0158] Figure 11 The composition of the cycle measurement data and discharge capacity retention rate data is shown. Regarding the details of the cycle measurement data, it has already been described in the first embodiment above. Figure 3 The description has been provided, therefore, the description is omitted here.

[0159] Discharge capacity retention rate is the discharge capacity retention rate of each cell in each cycle for which cyclic measurement data has been obtained. Figure 11 Examples show the discharge capacity retention rate of the m1th cycle, the discharge capacity retention rate of the m2th cycle, and so on.

[0160] Next, the configuration of the lifetime prediction and discharge capacity maintenance prediction models used in the second embodiment will be explained. Figure 12 This is a block diagram that roughly represents the structure of the lifetime prediction and discharge capacity maintenance prediction model used in the second embodiment of the present invention.

[0161] like Figure 12 As shown, the lifetime prediction and discharge capacity maintenance rate prediction model in the second embodiment includes a data shaping unit C-1, a feature extraction unit C-2, a nonlinear transformation unit C-3, a regression unit C-4, a nonlinear transformation unit C'-3, and a regression unit C'-4. The lifetime prediction and discharge capacity maintenance rate prediction model receives prediction cycle measurement data up to the prediction execution cycle number ni as input, and outputs the probability distribution of lifetime and the probability distribution of discharge capacity maintenance rate for each cycle.

[0162] Below, as an example, the elements constituting the lifetime prediction and discharge capacity maintenance prediction models will be explained respectively. However, since the first embodiment described above has already been based on... Figure 4 The data shaping section C-1 to the regression section C-4 have been explained, so only the nonlinear transformation section C'-3 and the regression section C'-4 will be explained here.

[0163] The nonlinear conversion unit C'-3 receives the compressed data output from the feature extraction unit C-2 as input and applies a nonlinear conversion to it to output nonlinear feature data. The subsequent regression unit C'-4 receives the nonlinear feature data as input and outputs the probability distribution of the discharge capacity maintenance rate for each cycle.

[0164] The methods used in the nonlinear transformation section C'-3 and the regression section C'-4 can be set separately, but it is preferable to apply Gaussian process regression, which has two functions, to C'-3 and C'-4. It should be noted that, when setting them separately, it is preferable to use the kernel method in the nonlinear transformation section C'-3. On the other hand, the output of the regression section C'-4 needs to be a probability distribution; therefore, in addition to Gaussian process regression, Bayesian ridge regression, etc., can also be used.

[0165] It should be noted that when Gaussian process regression is applied in the nonlinear transformation section C'-3 and the regression section C'-4, or when the kernel method is applied in the nonlinear transformation section C'-3, a nonlinear kernel such as a radial basis function kernel can be used to make it a nonlinear transformation.

[0166] As described above, according to the second embodiment, predictions can be made based on the probability distribution of lifetime using predicted cycle measurement data, as well as the predicted discharge capacity maintenance rate and 95% confidence interval for each cycle during the period from the start of the predicted execution cycle to the arrival of the predicted lifetime, or until a point in time exceeding the predicted lifetime.

[0167] [Other Implementation Methods]

[0168] In the second embodiment described above, the prediction of the probability distribution of lifetime and the probability distribution of discharge capacity maintenance rate were achieved by using an integrated model (lifetime prediction model and discharge capacity maintenance rate prediction model). However, it is also possible to use separate independent models (lifetime prediction model and discharge capacity maintenance rate prediction model) to predict the probability distribution of lifetime and the probability distribution of discharge capacity maintenance rate. Specifically, the system can be configured to prepare a lifetime prediction program and a discharge capacity maintenance rate prediction program separately, and to have the information processing device 500 execute each program independently.

[0169] This application claims priority based on Japanese Patent Application No. 2020-174106, filed on October 15, 2020, the entire contents of which are incorporated herein by reference.

[0170] [Industrial Applicability]

[0171] The lithium-ion battery life prediction method of the present invention can be applied to life evaluation (assessment) in the development of lithium-ion batteries.

Claims

1. A method for predicting the lifespan of a lithium-ion battery, characterized in that, Executed by computer: Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data; Step (b): For more than one prediction execution loop number, the learning data is applied to enable the life prediction model to learn, and a set of learned life prediction models corresponding to each prediction execution loop number is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the prediction loop measurement data obtained up to the predicted execution loop number into the learned lifetime prediction model for the corresponding predicted execution loop number, and obtaining the probability distribution of lifetime for each predicted execution loop number as the output. The lifetime prediction model has The data shaping unit shapes the cyclic measurement data into fixed-length data; The feature extraction unit compresses the fixed-length data into compressed data. and The nonlinear transformation unit converts the compressed data into nonlinear feature data mapped to a high-dimensional space to serve as nonlinear feature data. It is connected to the regression unit in this order, wherein the regression unit takes the nonlinear feature data as input and outputs the probability distribution of lifetime.

2. The lithium-ion battery life prediction method as described in claim 1, characterized in that, As the data compression method in the feature extraction section, a dimension reduction method is used.

3. The lithium-ion battery life prediction method as described in claim 2, characterized in that, The dimension reduction method described is principal component analysis.

4. A method for predicting the discharge capacity retention rate of a lithium-ion battery, characterized in that, Executed by computer: Step (a) is to acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle; Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime as the output.

5. A life prediction program product for enabling a computer to perform: Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data; Step (b): For more than one prediction execution loop number, the learning data is applied to enable the life prediction model to learn, and a set of learned life prediction models corresponding to each prediction execution loop number is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the prediction loop measurement data obtained up to the predicted execution loop number into the learned lifetime prediction model for the corresponding predicted execution loop number, and obtaining the probability distribution of lifetime for each predicted execution loop number as the output. The lifetime prediction model has The data shaping unit shapes the cyclic measurement data into fixed-length data; The feature extraction unit compresses the fixed-length data into compressed data. and The nonlinear transformation unit converts the compressed data into nonlinear feature data mapped to a high-dimensional space to serve as nonlinear feature data. It is connected to the regression unit in this order, wherein the regression unit takes the nonlinear feature data as input and outputs the probability distribution of lifetime.

6. A discharge capacity maintenance prediction program product for enabling a computer to execute: Step (a) is to acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle; Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime as the output.

7. An information processing apparatus, performing: Step (a) involves acquiring learning data, including battery cycle measurement data and lifespan data; Step (b): For more than one prediction execution loop number, the learning data is applied to enable the life prediction model to learn, and a set of learned life prediction models corresponding to each prediction execution loop number is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the prediction loop measurement data obtained up to the predicted execution loop number into the learned lifetime prediction model for the corresponding predicted execution loop number, and obtaining the probability distribution of lifetime for each predicted execution loop number as the output. The lifetime prediction model has The data shaping unit shapes the cyclic measurement data into fixed-length data; The feature extraction unit compresses the fixed-length data into compressed data. and The nonlinear transformation unit converts the compressed data into nonlinear feature data mapped to a high-dimensional space to serve as nonlinear feature data. It is connected to the regression unit in this order, wherein the regression unit takes the nonlinear feature data as input and outputs the probability distribution of lifetime.

8. An information processing apparatus, performing: Step (a) is to acquire learning data including battery cycle measurement data and discharge capacity retention rate for each cycle; Step (b): For more than one prediction execution cycle, the learning data is used to learn the discharge capacity maintenance rate model for each cycle, and a set of learned discharge capacity maintenance rate prediction models corresponding to each prediction execution cycle is obtained. Step (c) involves sequentially acquiring prediction cycle measurement data for the battery being predicted up to the number of prediction execution cycles. and Step (d) involves inputting the predicted cycle measurement data obtained up to the predicted execution cycle number into the learned discharge capacity maintenance rate prediction model for the corresponding predicted execution cycle number, and obtaining the probability distribution of the discharge capacity maintenance rate for each cycle during the period from the start of each predicted execution cycle number to the end of the lifetime as the output.

Images