Energy storage battery life prediction method and device and energy storage system
By calculating the Euclidean distance and clustering the capacity decay curves of energy storage batteries, the prediction model was optimized, which solved the problem of insufficient accuracy in predicting the lifespan of energy storage batteries and achieved higher prediction accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG JINKO ENERGY STORAGE CO LTD
- Filing Date
- 2025-12-03
- Publication Date
- 2026-05-12
AI Technical Summary
Existing methods for predicting the lifespan of energy storage batteries cannot accurately reflect the degradation patterns of batteries. In particular, when battery types are complex and diverse, the training and updating of prediction models are difficult, resulting in insufficient prediction accuracy.
By calculating the distance matrix between the capacity decay curves of batteries in the training set and the capacity decay curves of the target battery, a training subset with a similar decay pattern to the target battery is obtained through clustering. The initial prediction model is then adjusted to optimize the target prediction model and improve prediction accuracy.
Without increasing the amount of training data, the accuracy of battery life prediction is improved, especially under complex operating conditions, the prediction accuracy can be improved by 3% to 25%.
Smart Images
Figure CN121254089B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of battery life prediction technology, and in particular to a method, device and energy storage system for predicting the life of energy storage batteries. Background Technology
[0002] The lifespan of energy storage batteries is closely related to their performance. If the battery is used beyond its lifespan, it may affect power supply or damage electrical equipment, potentially leading to safety issues. Therefore, predicting the lifespan of energy storage batteries is beneficial for improving the stability and reliability of power supply.
[0003] However, due to the complex aging mechanism of energy storage batteries and the significant changes in battery operating conditions over time, current prediction methods cannot accurately reflect the degradation patterns of batteries. Improving prediction accuracy relies on a large amount of historical data and frequent data updates. With the complexity and diversity of battery types, the required data increases exponentially, making the training and updating of prediction models quite challenging. Summary of the Invention
[0004] The purpose of this application is to provide a method, device and energy storage system for predicting the lifespan of energy storage batteries. By clustering the original data in the training set, a training subset that better matches the degradation law of the target battery is selected. The initial prediction model is then optimized using the training subset to obtain the target prediction model, thereby improving the accuracy of the target prediction model in predicting the lifespan of the target battery.
[0005] To address the aforementioned technical problems, embodiments of this application provide a method for predicting the lifespan of an energy storage battery, comprising: determining a target battery to be predicted and an initial prediction model pre-trained using a training set; obtaining a distance matrix calculated from capacity decay curves; wherein the capacity decay curves include: capacity decay curves of the training set batteries and the capacity decay curve of the target battery; clustering a target battery device containing the target battery based on the distance matrix, wherein the capacity decay curves of the other batteries in the target battery device besides the target battery together form a training subset; adjusting the initial prediction model using the training subset to obtain a target prediction model; and using the target prediction model to predict the battery lifespan of the target battery.
[0006] Embodiments of this application also provide an energy storage battery life prediction device, comprising: a determination module, an acquisition module, a clustering module, an adjustment module, and a prediction module; the determination module is used to determine the target battery to be predicted and an initial prediction model pre-trained using a training set; the acquisition module is used to acquire a distance matrix calculated from the capacity decay curves; wherein, the capacity decay curves include: the capacity decay curves of the training set batteries and the capacity decay curve of the target battery; the clustering module is used to cluster target battery devices containing the target battery according to the distance matrix, wherein the capacity decay curves of the target battery devices other than the target battery together form a training subset; the adjustment module is used to adjust the initial prediction model using the training subset to obtain a target prediction model; the prediction module is used to predict the battery life of the target battery using the target prediction model.
[0007] Embodiments of this application also provide an energy storage system, including: an energy storage battery life prediction device as described above, or an energy storage battery life prediction device that performs the energy storage battery life prediction method described above.
[0008] Compared to related technologies, this embodiment of the application, after determining the target battery to be predicted and the initial prediction model pre-trained using the training set, obtains the capacity decay curves of the batteries in the training set and the target battery, and calculates a distance matrix based on the capacity decay curves of the batteries. Based on the distance matrix, a target battery device containing the target battery is obtained through clustering. The capacity decay curves of other batteries in the target battery device, excluding the target battery, together form a training subset. The training set is then filtered through clustering to obtain a training subset that has a certain correlation with the capacity decay pattern of the target battery. The initial prediction model is adjusted using this training subset to obtain the target prediction model. The target prediction model is then used to predict the battery life of the target battery. Without increasing the amount of data in the training set, clustering is used to filter and optimize the data in the training set, making the filtered training subset more closely match the capacity decay pattern of the target battery, thereby making the model's prediction of the target battery's life more accurate.
[0009] Furthermore, the step of clustering to obtain a target battery device containing the target battery based on the distance matrix includes: determining the similarity between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries based on the Euclidean distance between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries; and clustering the target battery device based on the similarity. By calculating the Euclidean distance to determine the similarity of the capacity decay curves of different batteries, a capacity decay curve with a similar capacity decay pattern to the target battery can be determined. This data processing method is simple and easy to operate.
[0010] In addition, the Euclidean distance between the two capacity decay curves is determined by: detecting the capacity of the battery at several preset period points during a periodic battery aging test; the capacity decay curve is composed of the capacity corresponding to each preset period point; and the sum of the squares of the differences between the same preset period points of the two capacity decay curves is taken as the Euclidean distance between the two capacity decay curves.
[0011] Furthermore, the distance matrix is as follows:
[0012] ;
[0013] Where i and j represent the cell numbers in the total training set, 0 < i ≤ (n+1), 0 < j ≤ (n+1), and n is an integer greater than 1; D represents the cell number distance matrix; d ij x represents the Euclidean distance between the capacity decay curves of battery i and battery j; i (k) represents the capacity decay curve of battery number i in the kth period; x j (k) represents the capacity decay curve of battery number j in the kth period, 0 < k ≤ N.
[0014] Furthermore, the step of adjusting the initial prediction model using the training subset includes: determining the temporal features of the training subset; and adjusting the weight parameters and bias parameters of the fully connected layers of the initial prediction model based on the temporal features. Optimizing the initial prediction model by adjusting only the fully connected layers improves the efficiency of model adjustment.
[0015] In addition, before obtaining the distance matrix calculated from the capacity decay curves, the method further includes: filtering the capacity decay curves of the target battery and the training set batteries; obtaining the distance matrix calculated from the capacity decay curves involves obtaining the distance matrix calculated from the filtered capacity decay curves. Filtering reduces noise in the data, thereby improving the accuracy of subsequent calculations.
[0016] In addition, the filtering process for the capacity decay curves of the target battery and the training set batteries includes: filtering the capacity decay curves using a moving average filter with a preset window size.
[0017] In addition, the initial prediction model is trained in the following way: determining the degradation characteristic data of several batteries based on preset battery aging tests; wherein, the degradation characteristic data includes: battery degradation characteristics corresponding to several preset period points during the periodic battery aging test; using the degradation characteristic data as the training set to train the initial prediction model.
[0018] In addition, the battery degradation characteristics include any one or a combination of the following indicators: capacity retention rate, capacity degradation rate, internal resistance growth rate, peak charge / discharge position, charge / discharge amplitude, voltage plateau change rate, and temperature change trend.
[0019] In addition, the step of determining the degradation characteristic data of several batteries based on the preset battery aging test includes: constructing an accelerated aging experiment matrix by using multiple controlled conditions according to the orthogonal principle or the full factorial principle; wherein the controlled conditions include any one or a combination of ambient temperature, charge / discharge rate, state of charge range and cycle depth; and using the accelerated aging experiment matrix to determine the degradation characteristic data of several batteries. Attached Figure Description
[0020] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments, and the figures in the drawings are not to be limited by scale.
[0021] Figure 1 This is a flowchart of a lifespan prediction method for an energy storage battery according to an embodiment of this application;
[0022] Figure 2 This is a flowchart of a sub-step of step 101 in a life prediction method for an energy storage battery according to an embodiment of this application;
[0023] Figure 3 This is a schematic diagram of the structure of the initial prediction model of a life prediction method for an energy storage battery according to an embodiment of this application;
[0024] Figure 4 This is a schematic diagram of the clustering principle in a life prediction method for an energy storage battery according to an embodiment of this application;
[0025] Figure 5 This is a schematic diagram of the structure of a life prediction device for an energy storage battery according to an embodiment of this application;
[0026] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation
[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been presented in the various embodiments of this application to enable readers to better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments.
[0028] The division of the following embodiments is for ease of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.
[0029] Embodiments of this application relate to a method for predicting the lifespan of energy storage batteries, such as... Figure 1 As shown, it includes:
[0030] Step 101: Determine the target battery to be predicted and the initial prediction model pre-trained using the training set.
[0031] Specifically, the target battery to be predicted refers to a single battery cell. The initial prediction model is a model trained using a training set, and it is used to predict the battery's lifespan. The training set includes: degradation characteristic data of several batteries determined by preset aging tests, or capacity degradation curves calculated based on the battery's degradation characteristics, etc.
[0032] like Figure 2 As shown, the initial prediction model can be trained in the following way:
[0033] Step 1011: Determine the degradation characteristic data of several batteries based on the preset battery aging test; wherein, the degradation characteristic data includes: battery degradation characteristics corresponding to several preset period points during the periodic battery aging test.
[0034] Specifically, battery aging test rules are designed based on the actual working requirements of the research battery. The design of these rules can involve constructing an accelerated aging test matrix under controlled experimental conditions, using orthogonal or full-factor principles. The controlled conditions include any or a combination of ambient temperature (T), charge / discharge rate (C-rate), state of charge window (SOC window), and depth of cycle (DOD). Other environmental conditions affecting the battery may also be included. The accelerated aging test matrix represents a combination of these controlled conditions. The purpose of designing the battery aging test is to accelerate battery capacity decay and performance degradation within a finite timeframe, and to record the battery's degradation characteristics at each stage of performance degradation, thereby accelerating the characteristic data recording process and improving characteristic data collection efficiency.
[0035] Battery aging testing is a periodic cycle experiment for batteries. One cycle consists of discharging the battery from full capacity to zero capacity and then recharging it back to full capacity. The battery degradation characteristic data includes a feature set consisting of battery degradation characteristics corresponding to several preset cycle points. Battery degradation characteristics include any one or a combination of the following: capacity retention rate, capacity decay rate, internal resistance growth rate, peak charge / discharge position (dQ), charge / discharge amplitude (dV), voltage plateau change rate, and temperature change trend. These battery degradation characteristics are feature vectors obtained by calculating, extracting, or fitting the battery's basic operating data. They are used to comprehensively characterize the battery's degradation state. In practical applications, degradation characteristic data can be any indicator that characterizes battery performance degradation over time, and is not limited to the aforementioned feature types.
[0036] Specifically, degradation characteristic data can also be presented as a capacity degradation curve. This curve represents the dynamic sequence of capacity changes with the number of cycles or operating time throughout the cyclic testing of the battery's aging process. The capacity degradation curve can characterize the battery's basic degradation characteristics and also provide the basis for extracting higher-order degradation characteristics. These higher-order characteristics include the battery's degradation rate, inflection point location, and curve fitting parameters. Compared to basic degradation characteristics, higher-order degradation characteristics can more accurately reflect the battery's lifespan degradation.
[0037] Step 1012: Use the decay feature data as the training set to train the initial prediction model.
[0038] Specifically, before training the initial prediction model, the decay feature data can be filtered to improve the accuracy of the initial prediction model. For example, when filtering the capacity decay curve, a moving average filter can be applied to the curve with a preset window size to remove occasional noise.
[0039] The specific process of average filtering is as follows:
[0040] Assume the initial capacity of the i-th capacity decay curve in the k-th period is x. i (k) has N periodic points. If a simple moving average filter is performed using a sliding window of length M, and M is an odd number, then the half-window width is: h = (M-1) / 2.
[0041] Filtered capacity value:
[0042] = ;
[0043] Specifically, for the first h periods of the capacity decay curve:
[0044] ;
[0045] For the last h periods in the capacity decay curve,
[0046] .
[0047] In practical applications, the value of n ranges from 20 to 100, usually determined based on the number of batteries tested and the operating conditions. The value of N ranges from 500 to 2000, determined by the scale of the battery aging test. The value of M can be an odd number such as 5, 7, or 9.
[0048] When filtering the capacity decay curve, in addition to the moving average filtering method mentioned above, smoothing algorithms such as exponential weighted average (EWMA), Gaussian filtering, wavelet denoising, or empirical mode decomposition (EMD) can also be used for filtering.
[0049] After filtering the capacity decay curves corresponding to each battery in the above manner, the filtered capacity decay curves are used as the training set to pre-train the initial prediction model.
[0050] The aforementioned filtering process can occur at any stage before obtaining the distance matrix calculated from the capacity decay curve. Filtering during the initial prediction model generation stage can reduce noise and improve the accuracy of the initial prediction model. Filtering the capacity decay curve during the initial prediction model update stage can improve the accuracy of the updated target prediction model.
[0051] The initial prediction model can be a sequence-to-sequence (Seq2Seq) model, with the model structure as follows: Figure 3 As shown, the system consists of an encoder and a decoder. The training set is input into the encoder, which converts the decaying feature data of the input into a vector through an embedding layer. The encoder then interprets the input sequence through LSTM layer 1, Dropout layer, and LSTM layer 2, converting the input data into a fixed-size context vector. The decoder inputs the initial state into LSTM layer 1 through the embedding layer. Based on the context vector from the encoder, it gradually generates the output content through LSTM layer 1, Dropout layer, LSTM layer 2, and a fully connected layer.
[0052] After identifying the target battery to be predicted and using an initial prediction model pre-trained from the training set, the following steps are performed:
[0053] Step 102: Obtain the distance matrix calculated from the capacity decay curves; wherein, the capacity decay curves include: the capacity decay curves of the training set of batteries and the capacity decay curves of the target batteries.
[0054] Specifically, the distance matrix can be determined by calculating the Euclidean distance between the capacity decay curves of any two batteries. The Euclidean distance reflects the similarity between the two capacity decay curves. The Euclidean distance between two capacity decay curves is calculated as follows:
[0055] During the periodic battery aging test, the capacity of the battery is measured at several preset period points; the capacity decay curve is composed of the capacity corresponding to each preset period point; the sum of the squares of the differences between the same preset period points of the two capacity decay curves is taken as the Euclidean distance between the two capacity decay curves.
[0056] Assume the training set contains n battery degradation characteristic data points, meaning it includes n capacity degradation curves, each with N cycle points (cycle number). The target battery's capacity degradation curve is denoted as the (n+1)th capacity degradation curve. Then, the capacity of the i-th capacity degradation curve in the k-th cycle is denoted as... Let i = 1, 2, 3, ..., n+1. n is an integer greater than 1.
[0057] The Euclidean distance between the i-th capacity decay curve and the j-th capacity decay curve can be expressed as:
[0058] .
[0059] To facilitate understanding of the Euclidean distance calculation method described above, a specific example will be used for explanation below:
[0060] Assume that the capacity decay curves of the i-th and j-th lines are in the most recent The capacity at each cycle point (capacity unit: % capacity retention rate) is:
[0061] ;
[0062] The squares of the differences between the cycle points of the two capacity decay curves are as follows:
[0063] ;
[0064] ;
[0065] ;
[0066] ;
[0067] ;
[0068] Summing and taking the square root of the above squared terms, we obtain the Euclidean distance between the i-th and j-th capacity decay curves as follows:
[0069]
[0070] Then, the n+1 capacity decay curves are paired, and the Euclidean distance of each pair of paired capacity decay curves is calculated, resulting in the distance matrix of the n+1 capacity decay curves:
[0071] ;
[0072] Where i and j represent the cell numbers in the total training set, 0 < i ≤ (n+1), 0 < j ≤ (n+1); D represents the cell number distance matrix; d ij x represents the Euclidean distance between the capacity decay curves of battery i and battery j; i (k) represents the capacity decay curve of battery number i in the kth period; x j (k) represents the capacity decay curve of battery number j in the kth period, 0 < k ≤ N.
[0073] The above uses % capacity as the unit of the capacity decay curve. The unit of the capacity decay curve can also be Ah. In order to avoid the influence of the dimensions and magnitude on the prediction results, the dimensions and magnitude of the capacity decay curve can be normalized. For example, the standardization can be performed according to the first cycle capacity or the maximum capacity of each curve. Then, the Euclidean distance is calculated based on the normalized capacity decay curve.
[0074] Step 103: Based on the distance matrix, cluster the target battery devices containing the target battery. The capacity decay curves of other batteries in the target battery devices, excluding the target battery, together form a training subset.
[0075] Specifically, the model trained using other batteries with similar degradation characteristics to the target battery has a higher accuracy in predicting battery life compared to the model trained using a training set containing multiple different degradation characteristics. Therefore, the purpose of screening the training set is to select target battery devices with high similarity to the target battery's degradation characteristics, and then determine the training subset. The target battery device can be an overall structure formed by combining multiple battery cells, such as battery modules, battery packs, and battery devices.
[0076] The target battery device is determined by: determining the similarity between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries based on the Euclidean distance between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries, and then clustering the target battery device based on the similarity.
[0077] In application, the target battery device can be determined through clustering calculations. The clustering calculation steps are as follows:
[0078] Taking n+1 capacity decay curves as an example, where the first to nth capacity decay curves represent the capacity decay curves of the batteries in the training set, and the (n+1)th capacity decay curve represents the capacity decay curve of the target battery, with each capacity decay curve having a period point of N. Initial clustering is performed on the n+1 capacity decay curves, with each curve forming an initial cluster. .
[0079] For any two initial clusters A and B The fully linked distance between two clusters is defined as: The fully linked distance represents the distance between the least dissimilar periodic points in two clusters.
[0080] Based on the fully linked distance, the initial clusters are merged. After t merges, the current cluster set is: Select the cluster pair with the smallest fully linked distance: ;merge Two clusters, the current cluster after the t-th merge is obtained. At this point, the target curve The cluster it belongs to is: , ,at this time, The first one containing the target battery One capacity decay curve and several capacity decay curves of training batteries, the number of which is denoted as . ,in 'm' is a preset value that can be set according to the number of capacity decay curves; for example, 'm' can be set to 10. When the number of capacity decay curves 'n' < 30, the range of 'm' can be [5, 8], or 'm' can be set according to the range [0.1n, 0.25n]. The final training subset set is as follows: .
[0081] The principle of merging the initial clusters mentioned above is as follows: Figure 4 As shown, each initial cluster is independent before merging. After one merging, b and c merge into one cluster, and d and e merge into another cluster. After two mergings, a, b, and c merge into one cluster. When the number of batteries in the current cluster containing the target battery after merging is greater than the preset number, it means that the number of training subsets has met the condition, and the selection of training subsets is completed.
[0082] Step 104: Adjust the initial prediction model using the training subset to obtain the target prediction model.
[0083] Specifically, adjusting the initial prediction model includes: determining the temporal features of the training subset; and adjusting the weight and bias parameters of the fully connected layers of the initial prediction model based on these temporal features. Optimizing the initial prediction model by adjusting only its fully connected layers improves the efficiency of model tuning.
[0084] by Figure 3 Taking the Sequence-to-Sequence (Seq2Seq) model as an example of the initial prediction model, when adjusting it, the parameters of the embedding layer, recurrent layers (LSTM layer 1 and LSTM layer 2), and dropout layer of the encoder and decoder are kept fixed. Only the fully connected layer at the output of the decoder is reconstructed. The parameters of the encoder and decoder, except for the fully connected layer, are frozen and not updated. Only the parameters of the fully connected layer are optimized, with the goal of minimizing the loss function between the predicted value and the true label. The optimization method is as follows:
[0085] Initialization: Fully Connected Layer Weights Uniform distribution or Xavier initialization can be used, with bias applied. Initialize to 0;
[0086] Iterative training: On the training subset data, the sequence of input capacity decay curves is processed by an Encoder-Decoder to extract temporal features, and finally, the prediction result is output by a fully connected layer; the prediction is updated according to the loss function. parameter;
[0087] Early stopping strategy: Monitor the mean squared error or remaining lifetime prediction error on the validation set, and stop training when there is no improvement for several consecutive rounds to prevent overfitting.
[0088] After being updated using the methods described above The parameters are used to update the initial prediction model to obtain the target prediction model.
[0089] Step 105: Use the target prediction model to predict the battery life of the target battery.
[0090] Specifically, by inputting the capacity decay curve of the target battery into the target prediction model, the model can output a predicted lifetime value or a future capacity change sequence, thereby achieving high-precision lifetime prediction. The target prediction model can determine when the target battery capacity will decrease below the lifetime threshold and output the remaining lifetime of the target battery.
[0091] The energy storage battery life prediction method proposed in this application improves the accuracy and adaptability of battery life prediction under varying operating conditions by introducing cluster analysis based on capacity decay characteristics and combining it with a sequence-to-sequence deep learning model. Furthermore, by training the method with data most similar to the target battery in the offline training set, not only is the model's predictive ability for different battery types enhanced, but the method also becomes better able to adapt to complex degradation mechanisms, thus improving the accuracy of life prediction for different battery types.
[0092] In addition, this application also designs an accelerated aging experiment matrix based on the actual working requirements of the research battery, constructs a battery life degradation feature set, and improves the versatility of the model.
[0093] Embodiments of this application relate to a device for predicting the lifespan of an energy storage battery, such as... Figure 5 As shown, it includes: a determination module, an acquisition module, a clustering module, an adjustment module, and a prediction module. The determination module is used to determine the target battery to be predicted and the initial prediction model pre-trained using the training set. The acquisition module is used to acquire the distance matrix calculated from the capacity decay curves. The capacity decay curves include the capacity decay curves of the batteries in the training set and the capacity decay curve of the target battery. The clustering module is used to cluster the target battery devices containing the target battery based on the distance matrix. The capacity decay curves of other batteries in the target battery devices, excluding the target battery, together form the training subset. The adjustment module is used to adjust the initial prediction model using the training subset to obtain the target prediction model. The prediction module is used to predict the battery life of the target battery using the target prediction model.
[0094] In one embodiment, the clustering module is further used to determine the similarity between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries based on the Euclidean distance between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries; and to obtain the target battery device by clustering according to the similarity.
[0095] In one embodiment, the energy storage battery life prediction device further includes: an Euclidean distance calculation module; the Euclidean distance calculation module is used to detect the capacity of the battery corresponding to several preset period points during the periodic battery aging test; the capacity decay curve is composed of the capacity corresponding to each preset period point; the sum of the squares of the differences between the same preset period points of the two capacity decay curves is used as the Euclidean distance between the two capacity decay curves.
[0096] In one embodiment, the Euclidean distance calculation module is specifically used to calculate the Euclidean distance according to the following formula.
[0097] ;
[0098] Where i and j represent the cell numbers in the total training set, 0 < i ≤ (n+1), 0 < j ≤ (n+1); D represents the cell number distance matrix; d ij x represents the Euclidean distance between the capacity decay curves of battery i and battery j; i (k) represents the capacity decay curve of battery number i in the kth period; x j (k) represents the capacity decay curve of battery number j in the kth period, 0 < k ≤ N.
[0099] In one embodiment, the adjustment module is used to determine the temporal features of the training subset; based on the temporal features, the weight parameters and bias parameters of the fully connected layers of the initial prediction model are adjusted.
[0100] In one embodiment, the energy storage battery life prediction device further includes a filtering module; the filtering module is used to filter the capacity decay curves of the target battery and the training set of batteries before obtaining the distance matrix calculated from the capacity decay curves.
[0101] In one embodiment, the filtering module is specifically used to filter the capacity decay curve using a moving average filtering method with a preset window size.
[0102] In one embodiment, the energy storage battery life prediction device further includes: a training module; the training module is used to determine the degradation characteristic data of a number of batteries according to a preset battery aging test; wherein, the degradation characteristic data includes: battery degradation characteristics corresponding to a number of preset period points during the periodic battery aging test; the degradation characteristic data is used as a training set to train an initial prediction model.
[0103] In one embodiment, battery degradation characteristics include any one or a combination of the following indicators: capacity retention rate, capacity degradation rate, internal resistance growth rate, peak charge / discharge position, charge / discharge amplitude, voltage plateau change rate, and temperature change trend.
[0104] In one embodiment, the training module is specifically used to construct an accelerated aging experiment matrix by means of multiple controlled conditions according to the orthogonal principle or the full factorial principle; wherein, the controlled conditions include any one or a combination of ambient temperature, charge / discharge rate, state of charge range and cycle depth; and the accelerated aging experiment matrix is used to determine the degradation characteristic data of several batteries.
[0105] Compared to related technologies, this application's embodiments, after determining the target battery to be predicted and an initial prediction model pre-trained using a training set, obtain the capacity decay curves of the batteries in the training set and the target battery's capacity decay curve, and calculate a distance matrix based on these battery capacity decay curves. Based on the distance matrix, clustering is used to obtain target battery devices containing the target battery. The capacity decay curves of other batteries in the target battery device, excluding the target battery, together form a training subset. Clustering is used to filter the training set to obtain a training subset that has a certain correlation with the capacity decay pattern of the target battery. This training subset is used to adjust the initial prediction model to obtain the target prediction model. The target prediction model is then used to predict the battery life of the target battery. Without increasing the amount of data in the training set, clustering is used to filter and optimize the data in the training set, making the filtered training subset more closely match the capacity decay pattern of the target battery, thus making the model's prediction of the target battery's life more accurate. Typically, the accuracy of predicting the life of the target battery using the target prediction model is 3% to 8% higher than that using the initial prediction model. When the training set contains a large amount of data, and the training subset determined based on the set has a large amount of data, the target prediction model improves the accuracy of predicting target battery life by 10% to 20% compared to the prediction accuracy using the initial prediction model. When the data in the training set is more dispersed, the accuracy of the target prediction model in predicting target battery life can even be improved by more than 25%.
[0106] It is not difficult to see that this embodiment is related to the method embodiment, and this embodiment can be implemented in conjunction with the method embodiment. The relevant technical details mentioned in the method embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the method embodiment.
[0107] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.
[0108] Embodiments of this application also provide an energy storage system, including: an energy storage battery life prediction device as described above, or an energy storage battery life prediction device that performs the energy storage battery life prediction method described above.
[0109] Compared with related technologies, the energy storage system provided in this application embodiment is equipped with the life prediction device for the energy storage battery provided in the aforementioned embodiment, or includes a life prediction device for the corresponding energy storage battery that performs the above-mentioned energy storage battery life prediction method. Therefore, it also has the technical effects provided by the aforementioned method embodiment or device embodiment, which will not be elaborated here.
[0110] This application also relates to an electronic device, such as... Figure 6 As shown, it includes at least one processor 601; and a memory 602 communicatively connected to at least one processor 601; wherein the memory 602 stores instructions executable by at least one processor 601, the instructions being executed by at least one processor 601 to enable at least one processor 601 to execute the energy storage battery life prediction method in the above embodiments.
[0111] The memory 602 and processor 601 are connected via a bus, which may include any number of interconnecting buses and bridges. The bus connects various circuits of one or more processors 601 and memory 602 together. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. A bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by processor 601 is transmitted over a wireless medium via an antenna, which further receives data and transmits it to processor 601.
[0112] Processor 601 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory 602 can be used to store data used by processor 601 during operation.
[0113] This application also relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the above-described method embodiments.
[0114] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0115] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.
Claims
1. A method for predicting the lifespan of an energy storage battery, characterized in that, include: Identify the target battery to be predicted and an initial prediction model pre-trained using the training set; The training set includes: degradation characteristic data of several batteries determined by preset aging tests, or capacity degradation curves calculated based on the degradation characteristics of the batteries. Obtain the distance matrix calculated from the capacity decay curves; wherein the capacity decay curves include: the capacity decay curves of the training set batteries and the capacity decay curves of the target batteries. The training set contains n capacity decay curves, and the capacity decay curve of the target battery is denoted as the (n+1)th capacity decay curve; combine the (n+1)th capacity decay curves pairwise, and calculate the Euclidean distance of all pairwise combinations of capacity decay curves to obtain the distance matrix of the (n+1)th capacity decay curves. Based on the distance matrix, clustering is used to obtain target battery devices that include the target battery. The capacity decay curves of other cells in the target battery device, excluding the target battery, together form a training subset. The target battery device is an integral structure formed by combining multiple battery cells. The initial prediction model is adjusted using the training subset to obtain the target prediction model; The target prediction model is used to predict the battery life of the target battery. The step of adjusting the initial prediction model using the training subset includes: Determine the temporal features of the training subset; Based on the temporal characteristics, the weight parameters and bias parameters of the fully connected layer of the initial prediction model are adjusted.
2. The method for predicting the lifespan of an energy storage battery according to claim 1, characterized in that, The step of clustering to obtain a target battery device containing the target battery based on the distance matrix includes: Based on the Euclidean distance between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries, the similarity between the capacity decay curve of the target battery and the capacity decay curves of the remaining batteries is determined. The target battery device is obtained by clustering based on the similarity.
3. The method for predicting the lifespan of an energy storage battery according to claim 2, characterized in that, The Euclidean distance between the two capacity decay curves is determined as follows: During a periodic battery aging test, the capacity of the battery is measured at several preset period points; the capacity decay curve is composed of the capacity corresponding to each preset period point. The sum of the squares of the differences between the same preset period points of the two capacity decay curves is taken as the Euclidean distance between the two capacity decay curves.
4. The method for predicting the lifespan of an energy storage battery according to claim 2, characterized in that, The distance matrix is as follows: ； Where i and j represent the cell numbers in the total training set, 0 < i ≤ (n+1), 0 < j ≤ (n+1), and n is an integer greater than 1; D represents the cell number distance matrix; d ij x represents the Euclidean distance between the capacity decay curves of battery i and battery j; i (k) represents the capacity decay curve of battery number i in the kth period; x j (k) represents the capacity decay curve of battery number j in the kth period, 0 < k ≤ N.
5. The method for predicting the lifespan of an energy storage battery according to claim 1, characterized in that, Before obtaining the distance matrix calculated from the capacity decay curve, the method further includes: The capacity decay curves of the target battery and the training set batteries are filtered. The distance matrix obtained from the capacity decay curve is as follows: Obtain the distance matrix calculated from the filtered capacity attenuation curve.
6. The method for predicting the lifespan of an energy storage battery according to claim 5, characterized in that, The filtering process for the capacity decay curves of the target battery and the training set batteries includes: The capacity decay curve is filtered using a moving average filter with a preset window size.
7. The method for predicting the lifespan of an energy storage battery according to any one of claims 1 to 6, characterized in that, The initial prediction model was trained in the following manner: The degradation characteristic data of several batteries are determined according to a preset battery aging test; wherein, the degradation characteristic data includes: battery degradation characteristics corresponding to several preset period points during the periodic battery aging test. The decay feature data is used as the training set to train the initial prediction model.
8. The method for predicting the lifespan of an energy storage battery according to claim 7, characterized in that, The battery degradation characteristics include any one or a combination of the following indicators: capacity retention rate, capacity degradation rate, internal resistance growth rate, peak charge / discharge position, charge / discharge amplitude, voltage plateau change rate, and temperature change trend.
9. The method for predicting the lifespan of an energy storage battery according to claim 7, characterized in that, The step of determining the degradation characteristic data of several batteries based on a preset battery aging test includes: An accelerated aging experiment matrix is constructed by using multiple controlled conditions according to the orthogonal principle or the full factorial principle; wherein the controlled conditions include any or a combination of ambient temperature, charge / discharge rate, state of charge range, and cycle depth. The degradation characteristic data of several batteries are determined using the accelerated aging experiment matrix.
10. A device for predicting the lifespan of an energy storage battery, characterized in that, include: The module includes a determination module, an acquisition module, a clustering module, an adjustment module, and a prediction module. The determining module is used to determine the target battery to be predicted and the initial prediction model pre-trained using the training set. The training set includes: degradation characteristic data of several batteries determined by preset aging tests, or capacity degradation curves calculated based on the degradation characteristics of the batteries. The acquisition module is used to acquire the distance matrix calculated from the capacity decay curves; wherein, the capacity decay curves include: the capacity decay curves of the training set batteries and the capacity decay curves of the target batteries, the training set contains n capacity decay curves, and the capacity decay curve of the target battery is denoted as the (n+1)th capacity decay curve; the (n+1)th capacity decay curves are combined pairwise, and the Euclidean distance of all the pairwise combinations of capacity decay curves is calculated to obtain the distance matrix of the (n+1)th capacity decay curves; The clustering module is used to cluster target battery devices containing the target battery according to the distance matrix. The capacity decay curves of the target battery devices other than the target battery together form a training subset. The target battery device is an integral structure formed by combining multiple battery cells. The adjustment module is used to adjust the initial prediction model using the training subset to obtain the target prediction model; The step of adjusting the initial prediction model using the training subset includes: Determine the temporal features of the training subset; Based on the aforementioned temporal characteristics, the weight parameters and bias parameters of the fully connected layer of the initial prediction model are adjusted. The prediction module is used to predict the battery life of the target battery using the target prediction model.
11. An energy storage system, characterized in that, include: The energy storage battery life prediction device as described in claim 10, or the energy storage battery life prediction device that performs the energy storage battery life prediction method as described in any one of claims 1 to 9.