Method and system for predicting the remaining life of lithium-ion batteries based on long-term and short-term models

Abstract

The present invention relates to the technical field of life prediction of energy storage systems, and provides a method and system for predicting the remaining life of lithium-ion batteries based on long-term and short-term models. The method comprises the following steps: establishing a lithium-ion battery database; obtaining a short-term model and a health value scale through the data in the lithium-ion battery database; using the data in the lithium-ion battery database, the health value scale and the long-term iterative model; importing the data of the battery to be tested into the short-term model to obtain the prediction result of the remaining life of the battery, and determining the health value scale of the battery to be tested; when the health value scale of the battery to be tested reaches the limit value, importing the data of the battery to be tested into the long-term iterative model to obtain the prediction result of the remaining life of the battery to be tested. The method solves the problem that the practical lithium battery prediction model is generally not accurate, is subject to many restrictions, and is difficult to achieve accurate prediction within a wide cycle range, and achieves the excellent effect of an average prediction error of the test battery of less than 5%.

Description

Technical Field

[0001] The present invention relates to the technical field of life prediction of energy storage systems, and in particular to a method and system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model. Background Art

[0002] Lithium battery energy storage systems are widely used in the field of deep peak regulation after new energy sources are connected to the power system due to their high energy density and long cycle life. However, due to certain safety hazards of lithium-ion batteries at the end of their life, timely and reliable life prediction methods are particularly important in the field of early warning. At present, the practical lithium battery prediction models are generally not very accurate and are subject to many restrictions, making it difficult to achieve accurate predictions within a wide cycle range. The data processing capabilities of lithium battery remaining life prediction models are generally weak and cannot achieve fast and high-precision predictions under large battery samples. The advance amount of lithium battery remaining life prediction is generally insufficient. In order to achieve an accurate remaining life (RUL) prediction, it is usually necessary to cycle more than half and reach the inflection point of lithium battery capacity decay before a more accurate prediction can be achieved. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the related art. To this end, the present invention provides a method and system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, so as to achieve accurate and rapid prediction of the life of a lithium-ion battery.

[0004] The present invention provides a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, comprising the following steps:

[0005] S1: Collect the capacity and time variation curves of lithium-ion batteries and establish a lithium-ion battery database;

[0006] S2: Perform CNN-LSTM training using the data from the lithium-ion battery database to obtain a short-term model and health indicators;

[0007] S3: Perform LSTM training based on the data and health indicators of the lithium-ion battery database to obtain a determination limit, a long-term iteration model, and a training result of the long-term iteration model;

[0008] S4: Import the data of the battery to be tested into the short-term model to obtain the short-term prediction result of the remaining life of the battery and the health value scale. When the health value scale is greater than or equal to the judgment limit, execute step S5. When the health value scale is less than the judgment limit, continue to execute step S4 to monitor the remaining life of the battery.

[0009] S5: Importing the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtaining a long-term prediction result of the remaining life of the battery according to the first prediction result combined with the training result of the long-term iterative model.

[0010] According to a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention, step S2 comprises:

[0011] S21: fitting, discretizing and segmenting the data of the lithium-ion battery database to obtain data sequence fragments;

[0012] S22: Conduct a preliminary experiment on the data of the lithium-ion battery database to obtain the optimal sequence fragment part;

[0013] S23: trimming the data sequence segments according to the optimal sequence segment part to obtain the optimal data sequence segment;

[0014] S24: Use the best data sequence fragment to train CNN-LSTM and obtain a short-term training model;

[0015] S25: Input the best data sequence fragment into the short-term training model, and obtain the health index through model mapping.

[0016] According to a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention, step S22 includes:

[0017] S221: Performing a grid search on a portion of the data in the lithium-ion battery database using a convolutional neural network to obtain input sequence fragments and convolutional network model parameters;

[0018] S222: Use K-fold cross validation to ensure the robustness of convolutional network model parameters;

[0019] S223: Save the best input sequence fragment and convolutional network model parameters to obtain the best sequence fragment part.

[0020] According to a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention, step S24 includes:

[0021] The model structure of the CNN-LSTM training includes: an input layer, a convolution layer, a pooling layer, an expansion layer, a verification layer, and an output layer, and a time-distributed packaging layer technology is used in the verification layer.

[0022] According to a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention, step S3 includes:

[0023] S31: Determine a determination limit value according to the health index;

[0024] S32: Smoothing the data of the lithium-ion battery database according to a Gaussian kernel smoothing formula to obtain smoothed data;

[0025] S33: Using LSTM to perform long-term iteration on the smoothed data to obtain a long-term iteration model;

[0026] S34: Using the long-term iterative model to train the data of the lithium-ion battery database to obtain a training result of the long-term iterative model.

[0027] According to a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention, step S31 includes:

[0028] The determination limit is a health value index, and the determination limit is required to be greater than or equal to 0.6 and less than or equal to 0.8; and the error of the absolute battery life cycle corresponding to the determination limit is less than 50 cycles, and the error of the relative battery life cycle corresponding to the determination limit is less than 0.1.

[0029] According to the present invention, a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model is provided, comprising: the iteration standard of the long-term iterative model is that RMSE is less than 0.1 and the R2 determination coefficient is greater than 0.9. When the conditions are not met, the model with the largest R2 determination coefficient is taken as the final retained model.

[0030] A method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention includes: when in step S4, the health value scale is greater than or equal to the judgment limit three times, executing step S5.

[0031] According to the present invention, a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model includes: in step S5, a first prediction result is compared with a training result of the long-term iterative model, and when the root mean square error between the first prediction result and the training result of the long-term iterative model is less than or equal to 0.05, the training result of the long-term iterative model is a similar training result,

[0032] When the number of similar training results is greater than or equal to a threshold, the number of the closest similar training results is determined according to the threshold, and the average value of the corresponding number of similar training results is calculated to obtain a long-term prediction result of the remaining life of the battery to be tested;

[0033] When the number of similar training results is less than the threshold, the similar training result with the smallest RMSE is taken as the long-term prediction result of the remaining life of the battery to be tested.

[0034] The present invention provides a system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, which is used to execute the system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, and includes:

[0035] Data collection module: collects lithium-ion battery capacity and time variation curves, and establishes a lithium-ion battery database;

[0036] Model training module: CNN-LSTM training is performed based on the data of the lithium-ion battery database to obtain a short-term model and health index; LSTM training is performed based on the data and health index of the lithium-ion battery database to obtain a determination limit, a long-term iterative model and a training result of the long-term iterative model;

[0037] Short-term model prediction module: import the data of the battery to be tested into the short-term model to obtain the short-term prediction result of the remaining battery life and the health value scale. When the health value scale is greater than or equal to the judgment limit, the long-term model prediction module is executed. When the health value scale is less than the judgment limit, the short-term model prediction module continues to be executed to monitor the remaining battery life.

[0038] Long-term model prediction module: import the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtain a long-term prediction result of the remaining life of the battery based on the first prediction result combined with the training result of the long-term iterative model.

[0039] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0040] The present invention provides a method and system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model. By combining a short-term model trained by a CNN-LSTM method and a long-term model trained by an LSTM method and introducing a health value scale, the accuracy of lithium battery life prediction is improved, the limitations are reduced, and accurate prediction over a wide range is achieved.

[0041] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A method flow chart of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0044] Figure 2 A short-term model flow chart of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0045] Figure 3 A schematic diagram of a lithium-ion battery capacity and time variation curve according to an embodiment of a method and system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0046] Figure 4 A schematic diagram of a long-term iterative model flow chart of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0047] Figure 5 Part of the short-term model prediction results of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0048] Figure 6 The present invention provides a method and system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, and provides some long-term iterative model prediction results.

[0049] Figure 7 The present invention provides a new battery prediction result of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model.

[0050] Figure 8 A system block diagram of a method and system embodiment for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model provided by the present invention.

[0051] Reference numerals:

[0052] 101. Data collection module; 102. Model training module; 103. Short-term model prediction module; 104. Long-term model prediction module. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme in the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0054] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0055] Example

[0056] Combine the following Figures 1 to 8 Describing the Invention:

[0057] This embodiment provides a method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model. Figure 1 Shown include:

[0058] S1: Collect the capacity and time variation curves of lithium-ion batteries and establish a lithium-ion battery database;

[0059] S2: Perform CNN-LSTM (Convolutional Neural Networks-Long Short Term Memory) training using the data from the lithium-ion battery database to obtain a short-term model and health indicators;

[0060] S3: Perform LSTM (Long short-term memory) neural network training based on the data and health indicators of the lithium-ion battery database to obtain a determination limit, a long-term iteration model, and a training result of the long-term iteration model;

[0061] S4: Import the data of the battery to be tested into the short-term model to obtain the short-term prediction result of the remaining life of the battery and the health value scale. When the health value scale is greater than or equal to the judgment limit, execute step S5. When the health value scale is less than the judgment limit, continue to execute step S4 to monitor the remaining life of the battery.

[0062] S5: Importing the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtaining a long-term prediction result of the remaining life of the battery according to the first prediction result combined with the training result of the long-term iterative model.

[0063] Specifically, in step S1, this embodiment uses the MIT battery database: the MIT battery data set includes full cycle data of 124 ternary lithium-ion batteries, the battery model is A123 APR18650M1A, and the default initial maximum capacity is 1.1Ah. The batteries in the MIT battery database all use step-by-step charging and constant current discharge, and there is a corresponding constant voltage trickle charge at the end of charging. In order to ensure a wide coverage of the overall battery data and ensure the continuity and continuity of the cycle data changes. For the lithium-ion battery data collection work, it is necessary to ensure that the collected charge and discharge termination conditions are consistent. If an unexpected situation occurs and the data is interrupted in advance, human intervention should be made to ensure the accuracy of the data. In the MIT data set, according to the above-mentioned restrictions, 26 groups of lithium-ion battery capacity and time change curves with excessive noise are discarded, and only 124 groups of lithium-ion battery capacity and time change curves with less noise are optimized. In this embodiment, four lithium-ion battery capacity and time change curves are randomly selected for reservation to act as new batteries.

[0064] Specifically, step S2 is as follows: Figure 2 As shown, the specific steps include:

[0065] S21: fitting, discretizing and segmenting the data of the lithium-ion battery database to obtain data sequence fragments;

[0066] S22: Conduct a preliminary experiment on the data of the lithium-ion battery database to obtain the optimal sequence fragment part;

[0067] S23: trimming the data sequence segments according to the optimal sequence segment part to obtain the optimal data sequence segment;

[0068] S24: Use the best data sequence fragment to train CNN-LSTM and obtain a short-term training model;

[0069] S25: Input the best data sequence fragment into the short-term training model, and obtain the health index through model mapping.

[0070] like Figure 3 As shown, Figure 3 A schematic diagram of a lithium-ion battery capacity and time variation curve of a method and system embodiment of the present invention for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model. In step S21, the lithium-ion battery capacity and time variation curve of the lithium-ion battery database are fitted. The embodiment of the present invention adopts the common nonlinear least squares fitting method to obtain the fitted curve, and then the fitted curve is discretized according to the formula:

[0071]

[0072] in, It is the upper cut-off voltage of the lithium battery; is the lower cut-off voltage of the lithium battery, and All are determined by the type of lithium battery; It is the data partitioning index, which is determined by the sampling rate; is the sampling sequence number, ; is the discrete collected voltage value. Then the discrete collected voltage value is segmented to obtain data sequence fragments.

[0073] In step S22, the lithium-ion battery noise analysis shows that the middle stage of the lithium battery cycle has the best robustness, and the corresponding noise and measurement errors are significantly smaller than those in the early and late stages. Therefore, a small number of lithium-ion batteries can be pre-tested to find the best sequence fragment part, and a common convolutional neural network can be used for grid search, and K-fold cross validation is used to ensure the robustness of the parameters, and finally the best input sequence fragment and CNN-LSTM model parameters are saved.

[0074] In step S23, the embodiment of the present invention uses 20 lithium-ion battery data to perform grid search using a convolutional neural network to obtain input sequence fragments and convolutional network model parameters, uses 5-fold cross validation to ensure the robustness of the optimization, and obtains the best sequence fragment part. According to the obtained best sequence fragment part, the remaining data sequence fragments are trimmed to obtain the best data sequence fragment.

[0075] Table 1 Short-term model architecture parameters

[0076]

[0077] The CNN-LSTM model parameters of the embodiment of the present invention are shown in Table 1. In step S24 and step S25, the best data sequence fragment is trained with the CNN-LSTM model to obtain a short-term training model; the best sequence fragment is input into the short-term training model, and the short-term training model is mapped to a health index in the range of greater than 0 and less than 1. (Wherein, the initial health is 1, and the end-of-life health is 0). In order to strengthen the extraction of hidden features, the model structure of the short-term model CNN-LSTM training in the embodiment of the present invention includes: input layer, convolution layer, pooling layer, expansion layer, verification layer, and output layer. Time-distributed packaging layer technology is used in the verification layer. It should be emphasized that for different types of battery data, a grid search method is required to obtain the most suitable network model parameters, and the network model parameters cannot be universal. In the embodiment of the present invention, the remaining 100 groups of best data sequence fragments are used as training and verification data, of which 20 groups are used as test data. The ratio of training and verification data is 4:1, and 5-fold cross-validation is also used to ensure the robustness of the model results. After using the above parameters for short-term model prediction, the predicted RMSE (Root Mean Square Error) is 0.0374, and the R2 determination coefficient is 0.965, and the model prediction effect is good. Some short-term prediction results are as follows Figure 5 As shown in (a) to (c).

[0078] Specifically, step S3 is as follows: Figure 4 As shown:

[0079] S31: Determine a determination limit value according to the health index;

[0080] S32: Smoothing the data of the lithium-ion battery database according to a Gaussian kernel smoothing formula to obtain smoothed data;

[0081] S33: Using LSTM to perform long-term iteration on the smoothed data to obtain a long-term iteration model;

[0082] S34: Using the long-term iterative model to train the data of the lithium-ion battery database to obtain a training result of the long-term iterative model.

[0083] Among them, step S31 determines the health value scale and determines the determination limit to improve the overall operating efficiency and ensure accuracy. For the determination limit, it is necessary to ensure that there is a small relative life cycle error at this time. and absolute life cycle error From the training effect of the short-term model, we can see that with the preset The reduction Gradually getting smaller, The starting point is between 0.6 and 0.8, and the short-term model corresponds to and are all small, and the data accuracy is high. In this range, the health index Less than 30 laps, Are all less than 20%. In addition, when the health index is selected in the range of 0.6~0.8, the advance prediction is better, which is more helpful for practical applications, and the amount of data input into the long-term iteration model is moderate. When the health index is less than 0.6, the amount of input data for long-term iteration increases exponentially, and the corresponding training time is greatly extended. In the actual model training, the prediction accuracy is not greatly improved after the health index is less than 0.6, and its prediction effect is close to that in the range of 0.6~0.8, but the training time is greatly increased. When the health index is greater than 0.8, the battery is relatively healthy, and there is no need to determine the long-term iteration model of the battery. Under normal circumstances, the health indicator node corresponds to Usually needs to be less than 0.1, Usually less than 50 turns are required. In the embodiment of the present invention, through comparison and optimization, the final As the judgment limit, refer to Table 2 for selection.

[0084] Table 2 Different health indicators of the present invention and surface

[0085]

[0086] Due to the differences in battery capacity changes, the overall noise situation is very serious, and a relatively reliable smoothing tool must be used for processing before long-term iterative prediction can be performed. Therefore, in step S32, the data of the lithium-ion battery database is smoothed by Gaussian kernel. The Gaussian kernel smoothing formula used in the embodiment of the present invention is as follows:

[0087]

[0088] in, is the coordinate of any point, represents the center of the kernel function, Represents any point arrive The square of the Euclidean distance, is the Gaussian kernel width, Depending on the required degree of smoothing, the effect of Gaussian kernel smoothing needs to take into account the actual amount of data and error conditions.

[0089] For the long-term iterative model, a variable sliding time window is used to ensure effective processing of wide cycle big data. In the long-term iterative model training of the embodiment of the present invention, the conventional iteration window length is 128. For numbered batteries with too little division data, window lengths of 64, 32, 16, etc. will be tested in turn until the final iteration procedure can proceed smoothly. In the embodiment of the present invention, the LSTM model architecture of the long-term iterative model is shown in Table 3 below:

[0090] Table 3 Long-term iterative model architecture parameters

[0091]

[0092] For long-term iterative models, multiple evaluation indicators must be met at the same time before time series prediction with a high degree of fit can be achieved. The main evaluation indicators used in this invention are as follows:

[0093]

[0094]

[0095] in, is the total number, is the true value, is the predicted value, is the average of the true values, is the individual ordinal number.

[0096] The iteration standard set in the embodiment of the present invention is , . For the embodiment of the present invention, the proportion of training batteries that meet the preset standard is 95%, among which the data of 6 batteries still fail to meet the preset standard after the maximum iteration of 10,000 times. For the battery set as a whole, the final average absolute life cycle error is 18.225, and the average relative life cycle error is 0.031. The training results of some long-term iterative models are as follows Figure 6 As shown in (a) to (d).

[0097] According to the actual application stage of steps S4 and S5, the corresponding real-time health acquisition and node judgment are mainly completed for new batteries. The real-time health index of the new battery can be directly obtained through the trained a priori short-term model, which is based on the preliminary estimate of the battery life in the early stage. When the starting point judgment condition is met, in order to ensure the stability of the model link, the matching long-term iterative model can be used for prediction only after the judgment conditions are met three times. For the matching problem of new batteries, the judgment nodes and the previous health changes are mainly used to match with the corresponding a priori models. The matching algorithm simultaneously considers multiple error judgment parameters and gives different matching ratios.

[0098] The first prediction result is compared with the training result of the long-term iterative model. When the root mean square error between the first prediction result and the training result of the long-term iterative model is less than or equal to 0.05, the training result of the long-term iterative model is a similar training result.

[0099] When the number of similar training results is greater than or equal to the threshold, the number of the closest similar training results is determined according to the threshold, and the average value is calculated to obtain the long-term prediction result of the remaining life of the battery to be tested;

[0100] When the number of similar training results is less than a threshold, the similar training result with the smallest root mean square error is taken as the long-term prediction result of the remaining life of the battery to be tested.

[0101] Finally, the three models with the highest matching degree are found for long-term iterative prediction. From the final long-term prediction results, if the root mean square error is small, the mean is used to improve the stability of iterative prediction; if the long-term prediction results have a large deviation, the similar training results with the smallest root mean square error are trusted first. After using this method, accurate predictions can be achieved for new batteries under wide cycle conditions.

[0102] In the embodiment of the present invention, for the application of brand new batteries, the real-time short-term model output health is used as a reference in the early stage to ensure the stable operation of the battery; when the judgment condition is triggered, the long-term iterative model can be applied to obtain the subsequent capacity changes and the final battery life. In the specific implementation, when the batteries ['b1c11'], ['b2c29'], ['b2c42'] and ['b3c31'] are taken as brand new batteries, the final life prediction error of the model is =0.042 and =18.75. The specific model prediction effect is as follows Figure 7 As shown. Among them, ['b1c11'] corresponds to Figure 7 In (a), ['b2c29'] corresponds to Figure 7 In (b), ['b2c42'] corresponds to Figure 7 In (c), ['b3c31'] corresponds to Figure 7 In (d), the first three batteries ['b1c11'], ['b2c29'], and ['b2c42'] take the average of the three groups of predictions as the final long-term prediction result. For the last battery ['b3c31'], due to the large difference in predictions, the prediction 1 with the highest matching degree is taken as the final result.

[0103] The comparison between the model proposed in the present invention and other models is shown in Table 4:

[0104] Table 4 Comparison table of the model proposed in this invention and other models

[0105]

[0106] The method and system for predicting the remaining life of lithium-ion batteries based on a long-term and short-term model proposed in the present invention can significantly improve the accuracy and versatility of the model. While ensuring sufficient accuracy, it can be applied to energy storage battery systems with a larger number of samples and a wider cycle range. The life prediction method has an improved lead time compared to traditional methods, is suitable for energy storage systems with large differences in battery cycle numbers, can significantly reduce noise interference, and improve economic benefits and safety.

[0107] The present invention provides a system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model. Figure 8 As shown, the method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model includes:

[0108] Data collection module 101: collects lithium-ion battery capacity and time variation curves, and establishes a lithium-ion battery database;

[0109] Model training module 102: Perform CNN-LSTM training based on the data of the lithium-ion battery database to obtain a short-term model and a health index; perform LSTM training based on the data of the lithium-ion battery database and the health index to obtain a determination limit, a long-term iterative model, and a training result of the long-term iterative model;

[0110] Short-term model prediction module 103: import the data of the battery to be tested into the short-term model, obtain the short-term prediction result of the remaining life of the battery and the health value scale, when the health value scale is greater than or equal to the judgment limit, execute the long-term model prediction module, when the health value scale is less than the judgment limit, continue to execute the short-term model prediction module to monitor the remaining life of the battery;

[0111] Long-term model prediction module 104: imports the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtains a long-term prediction result of the remaining life of the battery based on the first prediction result combined with the training result of the long-term iterative model.

[0112] The embodiment of the present invention utilizes the CNN-LSTM neural network model to process the capacity-voltage data of the battery, and utilizes the LSTM iterative model to process the maximum capacity-cycle data of the battery, and ultimately demonstrates an excellent effect of an average prediction error of the test battery of less than 5% on the MIT battery dataset (a wide cycle dataset of 150 to 2000 cycles).

[0113] The long-term and short-term fusion model used in the embodiment of the present invention can process up to 120 full-cycle battery data at the same time, and can still ensure high accuracy and good evaluation indicators for batteries with a wide cycle range. In terms of fast prediction, after using sufficient battery data for preliminary training, in actual applications, the matching time and response time of the corresponding model are both less than 10s.

[0114] In the neural network model of the present invention, the short-term model is used to roughly judge the current circulation state, and the short-term health index can be used to To identify and confirm the starting point of life prediction. For a brand new battery of the same type, the present invention predicts the future capacity change and the final remaining life cycle of the battery after collecting 30% to 40% of the data.

Claims

1. A method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, characterized in that: The following steps are involved: S1: Collect the capacity and time variation curves of lithium-ion batteries and establish a lithium-ion battery database; S2: Perform CNN-LSTM training using the data from the lithium-ion battery database to obtain a short-term model and health indicators; S3: LSTM training is performed using the data and health indicators of the lithium-ion battery database to obtain a determination limit, a long-term iteration model, and a training result of the long-term iteration model: S31: Determine a determination limit value according to the health index, the determination limit value is a health value index, the determination limit value is required to be greater than or equal to 0.6 and less than or equal to 0.8; and the determination limit value corresponds to the battery absolute life cycle error is less than 50 cycles, and the determination limit value corresponds to the battery relative life cycle error is less than 0.1; S32: Smoothing the data of the lithium-ion battery database according to a Gaussian kernel smoothing formula to obtain smoothed data; S33: Using LSTM to perform long-term iteration on the smoothed data to obtain a long-term iteration model, wherein the iteration criteria of the long-term iteration model are that RMSE is less than 0.1 and R2 determination coefficient is greater than 0.

9. When the conditions are not met, the model with the largest R2 determination coefficient is selected as the final retained model; S4: Import the data of the battery to be tested into the short-term model to obtain the short-term prediction result of the remaining life of the battery and the health value scale. When the health value scale is greater than or equal to the judgment limit, execute step S5. When the health value scale is less than the judgment limit, continue to execute step S4 to monitor the remaining life of the battery. S5: Importing the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtaining a long-term prediction result of the remaining life of the battery according to the first prediction result combined with the training result of the long-term iterative model.

2. The method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model according to claim 1, characterized in that: Step S2 includes: S21: fitting, discretizing and segmenting the data of the lithium-ion battery database to obtain data sequence fragments; S22: Conduct a preliminary experiment on the data of the lithium-ion battery database to obtain the optimal sequence fragment part; S23: trimming the data sequence segments according to the optimal sequence segment part to obtain the optimal data sequence segment; S24: Use the best data sequence fragment to train CNN-LSTM and obtain a short-term training model; S25: Input the best data sequence fragment into the short-term training model, and obtain the health index through model mapping.

3. The method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model according to claim 2, characterized in that: Step S22 includes: S221: Performing a grid search on a portion of the data in the lithium-ion battery database using a convolutional neural network to obtain input sequence fragments and convolutional network model parameters; S222: Use K-fold cross validation to ensure the stability of convolutional network model parameters; S223: Save the best input sequence fragment and convolutional network model parameters to obtain the best sequence fragment part.

4. The method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model according to claim 2, characterized in that: Step S24 includes: The model structure of the CNN-LSTM training includes: an input layer, a convolution layer, a pooling layer, an expansion layer, a verification layer, and an output layer, and a time-distributed packaging layer technology is used in the verification layer.

5. The method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model according to claim 1, characterized in that: include: When the health value scale is greater than or equal to the determination limit three times in step S4, step S5 is executed.

6. The method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model according to claim 1, characterized in that: In step S5, the first prediction result is compared with the training result of the long-term iterative model. When the root mean square error between the first prediction result and the training result of the long-term iterative model is less than or equal to 0.05, the training result of the long-term iterative model is a similar training result. When the number of similar training results is greater than or equal to the threshold, the number of the closest similar training results is determined according to the threshold, and the average value of the corresponding number of similar training results is calculated to obtain the prediction result of the remaining life of the battery to be tested; When the number of similar training results is less than the threshold, the similar training result with the smallest RMSE is taken as the prediction result of the remaining life of the battery to be tested.

7. A system for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model, used to execute the method for predicting the remaining life of a lithium-ion battery based on a long-term and short-term model as claimed in any one of claims 1 to 6, characterized in that: include: Data collection module: collects lithium-ion battery capacity and time variation curves, and establishes a lithium-ion battery database; Model training module: CNN-LSTM training is performed through the data of the lithium-ion battery database to obtain a short-term model and health index; LSTM training is performed through the data and health index of the lithium-ion battery database to obtain a judgment limit, a long-term iterative model and the training results of the long-term iterative model: Through the data and health indicators of the lithium-ion battery database, LSTM training is performed to obtain the judgment limit, long-term iteration model and the training results of the long-term iteration model: Determine a determination limit value according to the health index, the determination limit value is a health value index, the determination limit value is required to be greater than or equal to 0.6 and less than or equal to 0.8; and the determination limit value corresponds to the battery absolute life cycle error is less than 50 cycles, and the determination limit value corresponds to the battery relative life cycle error is less than 0.1; Smoothing the data in the lithium-ion battery database according to a Gaussian kernel smoothing formula to obtain smoothed data; The smoothed data is subjected to long-term iteration using LSTM to obtain a long-term iteration model, wherein the iteration criteria of the long-term iteration model are that RMSE is less than 0.1 and R2 determination coefficient is greater than 0.

9. When the conditions are not met, the model with the largest R2 determination coefficient is taken as the final retained model; Short-term model prediction module: import the data of the battery to be tested into the short-term model to obtain the short-term prediction result of the remaining battery life and the health value scale. When the health value scale is greater than or equal to the judgment limit, the long-term model prediction module is executed. When the health value scale is less than the judgment limit, the short-term model prediction module continues to be executed to monitor the remaining battery life. Long-term model prediction module: import the data of the battery to be tested into the long-term iterative model to obtain a first prediction result of the remaining life of the battery to be tested, and obtain a long-term prediction result of the remaining life of the battery based on the first prediction result combined with the training result of the long-term iterative model.

Images