A method and system for predicting lithium battery capacity attenuation trend

Abstract

The present invention provides a method and system for predicting the capacity decay trend of a lithium battery in the field of lithium battery detection technology. The method comprises the following steps: step S10, creating a normal decay prediction model based on CNN and LSTM, and training the normal decay prediction model; step S20, creating a capacity proliferation prediction model based on DNN, and training the capacity proliferation prediction model; step S30, generating a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity proliferation prediction model; step S40, using the capacity decay trend prediction model to predict the capacity decay trend of a lithium battery. The advantage of the present invention is that it greatly improves the accuracy of the capacity decay trend prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery detection, and in particular to a method and system for predicting the capacity attenuation trend of a lithium battery. Background Art

[0002] Lithium batteries have the advantages of high energy density, high cycle life, and no memory effect, and are widely used in the fields of ships, automobiles, aerospace, etc. However, after a long period of charge and discharge cycles, the available capacity of lithium batteries will decrease, the performance will decline, and there may be safety hazards. Therefore, it is very important to predict the capacity attenuation trend of lithium batteries.

[0003] In the accelerated test of the cycle life of lithium batteries, different test conditions (such as temperature, rate, etc.), test conditions, environmental switching, and lithium battery static conditions will all affect the capacity decay trend. Due to the relaxation effect of lithium batteries, the battery capacity often reverses during the intermediate process. How to accurately analyze the influencing factors and predict the decay trend is of great significance for the state estimation and aging prediction of lithium batteries.

[0004] Traditionally, there are two methods for predicting the battery capacity attenuation trend: model-based prediction methods and data-driven prediction methods. Model-based prediction methods require an in-depth understanding of the degradation mechanism, fully reflecting the characteristics of the degradation process or empirical knowledge of the degradation model, which limits the practical application of the model; data-driven prediction methods use machine learning or deep learning to analyze and extract features from lithium battery big data, and then predict the battery capacity attenuation trend. However, this method does not consider the impact of capacity proliferation, resulting in a certain deviation between the predicted results and the actual results.

[0005] Therefore, how to provide a method and system for predicting the capacity decay trend of a lithium battery to improve the accuracy of the capacity decay trend prediction has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for predicting the capacity decay trend of a lithium battery, so as to improve the accuracy of the capacity decay trend prediction.

[0007] In a first aspect, the present invention provides a method for predicting the capacity attenuation trend of a lithium battery, comprising the following steps:

[0008] Step S10, creating a normal attenuation prediction model based on CNN and LSTM, and training the normal attenuation prediction model;

[0009] Step S20: creating a capacity growth prediction model based on DNN, and training the capacity growth prediction model;

[0010] Step S30, generating a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model;

[0011] Step S40: using the capacity decay trend prediction model to predict the capacity decay trend of the lithium battery.

[0012] Furthermore, the step S10 specifically includes:

[0013] Step S11, creating a normal attenuation prediction model based on CNN and LSTM; the CNN uses a conv2d convolution kernel to convolve the input data into the data dimension required by LSTM;

[0014] Step S12, acquiring a large amount of first charge and discharge data including at least voltage, current, resistance, temperature, power, and charge and discharge time;

[0015] Step S13, setting a duration threshold, sequentially intercepting the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and filling the charge and discharge electronic data whose duration does not meet the duration threshold with 0;

[0016] Step S14: training a normal attenuation prediction model using the charge and discharge electronic data.

[0017] Furthermore, the step S20 is specifically as follows:

[0018] Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

[0019] Furthermore, the calculation formula of the capacity growth amplitude is:

[0020] Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth;

[0021] The capacity growth period represents the equivalent number of charge and discharge cycles between two adjacent capacity growths.

[0022] Furthermore, in step S30, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity growth amplitude and the capacity growth period predicted by the capacity growth prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity growth amplitude and the capacity growth period.

[0023] In a second aspect, the present invention provides a lithium battery capacity attenuation trend prediction system, comprising the following modules:

[0024] A normal attenuation prediction model training module, used for creating a normal attenuation prediction model based on CNN and LSTM, and training the normal attenuation prediction model;

[0025] A capacity growth prediction model training module, used to create a capacity growth prediction model based on DNN and train the capacity growth prediction model;

[0026] A capacity decay trend prediction model generation module, used to generate a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model;

[0027] The capacity decay trend prediction module is used to predict the capacity decay trend of the lithium battery using the capacity decay trend prediction model.

[0028] Furthermore, the normal attenuation prediction model training module specifically includes:

[0029] A normal attenuation prediction model creation unit, used to create a normal attenuation prediction model based on CNN and LSTM; the CNN uses a conv2d convolution kernel to convolve input data into a data dimension required by LSTM;

[0030] A first charge-discharge data acquisition unit, used to acquire a large amount of first charge-discharge data including at least voltage, current, resistance, temperature, power and charge-discharge time;

[0031] A charge and discharge data interception unit, used to set a duration threshold, sequentially intercept the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and fill the charge and discharge electronic data whose duration does not meet the duration threshold with 0;

[0032] The model training unit is used to train the normal attenuation prediction model using the charge and discharge electron data.

[0033] Furthermore, the volume growth prediction model training module is specifically used for:

[0034] Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

[0035] Furthermore, the calculation formula of the capacity growth amplitude is:

[0036] Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth;

[0037] The capacity growth period represents the equivalent number of charge and discharge cycles between two adjacent capacity growths.

[0038] Furthermore, in the capacity decay trend prediction model generation module, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity increase amplitude and the capacity increase period predicted by the capacity growth prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity growth amplitude and the capacity growth period.

[0039] The advantages of the present invention are:

[0040] By creating a normal attenuation prediction model for predicting the main attenuation trend of lithium battery capacity, creating a capacity proliferation prediction model for predicting the capacity proliferation amplitude and capacity proliferation cycle, and training the normal attenuation prediction model and the capacity proliferation prediction model, and then generating a capacity attenuation trend prediction model based on the trained normal attenuation prediction model and the capacity proliferation prediction model, and finally using the capacity attenuation trend prediction model to predict the capacity attenuation trend of lithium batteries, that is, comprehensively considering the main attenuation trends and capacity proliferation in the degradation process of lithium batteries, reducing the error of capacity attenuation trend prediction, and thus greatly improving the accuracy of capacity attenuation trend prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0042] Figure 1 The present invention is a flow chart of a method for predicting the capacity attenuation trend of a lithium battery.

[0043] Figure 2 It is a structural schematic diagram of a lithium battery capacity attenuation trend prediction system of the present invention.

[0044] Figure 3 It is a schematic diagram of the process of the present invention.

[0045] Figure 4 It is a structural schematic diagram of the normal attenuation prediction model of the present invention.

[0046] Figure 5 It is a structural schematic diagram of the capacity growth prediction model of the present invention.

[0047] Figure 6 It is a schematic diagram of the structure of the LSTM of the present invention.

[0048] Figure 7 It is a curve schematic diagram of normal attenuation of the present invention.

[0049] Figure 8 It is a curve schematic diagram of capacity growth of the present invention.

[0050] Fig. 9 It is a curve schematic diagram of the present invention for predicting the capacity attenuation trend of the battery 1.

[0051] Fig.10 It is a curve schematic diagram of the present invention for predicting the capacity attenuation trend of the battery 2. DETAILED DESCRIPTION

[0052] The technical solution in the embodiments of the present application has the following overall idea: creating a normal attenuation prediction model and a capacity proliferation prediction model, generating a capacity attenuation trend prediction model based on the trained normal attenuation prediction model and the capacity proliferation prediction model, and then using the capacity attenuation trend prediction model to predict the capacity attenuation trend of the lithium battery, comprehensively considering the main attenuation trends and capacity proliferation in the degradation process of the lithium battery to improve the accuracy of the capacity attenuation trend prediction.

[0053] Please refer to Figures 1 to 10 As shown, a preferred embodiment of a method for predicting the capacity attenuation trend of a lithium battery of the present invention comprises the following steps:

[0054] Step S10, creating a normal attenuation prediction model based on CNN (convolutional neural network) and LSTM (long short-term memory network), and training the normal attenuation prediction model;

[0055] Step S20: creating a capacity growth prediction model based on DNN (deep neural network), and training the capacity growth prediction model;

[0056] Step S30, generating a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model;

[0057] Step S40: using the capacity decay trend prediction model to predict the capacity decay trend of the lithium battery.

[0058] The step S10 specifically includes:

[0059] Step S11, creating a normal attenuation prediction model based on CNN and LSTM; the CNN adopts a conv2d convolution kernel, which can extract the features themselves and the effective information between the features, and can change the data dimension at the same time, so as to convolve the input data into the data dimension required by LSTM; the normal attenuation prediction model is used to predict the main attenuation trend of the lithium battery capacity;

[0060] The structure of the normal attenuation prediction model is as follows Figure 4As shown in the figure, CNN is used to extract features from time series information, LSTM is used to extract features and model the correlation between time information, and finally the features extracted by the network are fully connected to output the predicted number of cycles corresponding to multiple specific attenuation values, thereby fitting a curve of the normal degradation trend of lithium batteries;

[0061] The input data can be charging data, discharging data or complete charging and discharging data of 1 or N consecutive cycles. Since the time series lengths of different cycles are different, and the normal attenuation prediction model can only process fixed-length inputs, it is necessary to truncate the data into a uniform length based on the time length threshold and perform corresponding normalization processing; the final data dimension input to the normal attenuation prediction model is NxCxHxW; where N represents batch_size; C represents the number of feature dimensions, such as only inputting V, I, and T, then C=3; H represents the number of input cycles; W represents the time series length of each feature of each cycle;

[0062] The formula for conv2d input and output is as follows:

[0063]

[0064] Among them, W intput Indicates the size of the input data; W output Indicates the size of the output data; K filter Indicates the size of the convolution kernel; Padding indicates the size of the padding; Stride indicates the step size of the convolution kernel;

[0065] LSTM has two transmission states, namely neuron state C t and the hidden state h t , LSTM first uses the current input x t and the previous hidden state h t-1 The splicing calculation results in four states, namely f t , i t , o t , the calculation formula is as follows:

[0066] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0067]

[0068] i t =σ(W t ·[h t-1 ,x t ]+b t);

[0069] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0070] Among them, f t Represents the forget gate, which is used to control which of the previous neuron states (cell states) need to be forgotten; represents the state update gate; i t Represents the input gate, used to control Which features of are used to update C t ;o t represents the output gate; σ() represents the activation function; W f , W c , W i , W o Both represent weight; b f , b c , b t , b o All represent bias; tanh() represents the activation function;

[0071] The output is calculated from four states, including the hidden layer:

[0072] h t =o t *tanh(C t ),

[0073] Step S12, acquiring a large amount of first charge and discharge data including at least voltage, current, resistance, temperature, power, and charge and discharge time;

[0074] Step S13, setting a duration threshold, sequentially intercepting the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and filling the charge and discharge electronic data whose duration does not meet the duration threshold with 0;

[0075] Step S14: training a normal attenuation prediction model using the charge and discharge electronic data.

[0076] The step S20 is specifically as follows:

[0077] Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

[0078] Since the rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy will affect the capacity growth, the second charge and discharge data is used to train the capacity growth prediction model.

[0079] The structure of the volume growth prediction model is as follows: Figure 5 As shown, it includes input layer, hidden layer and output layer; assuming that the input vector is <x 1 ,x 2 ,x 3 >, the hidden layer nodes are 4, 5, 6, 7, the output layer nodes are 8, 9, and the formula for calculating the hidden layer 4 from the input layer nodes is a4 = δ(W 14 *x 1 +W 24 *x 2 +W 34 *x 3 +b 1 ), where W 14 represents the weight between nodes 1 and 4; b 1 represents the bias of node 4; δ() represents the activation function, usually the sigmoid function, expressed as y = 1 / (1+e -x ), in order to introduce nonlinearity to increase the network's expressive power; the calculation method of the nodes in the output layer is similar to that of the hidden layer, except that the input to the output layer becomes the nodes in the hidden layer.

[0080] The calculation formula of the volume growth amplitude is:

[0081] Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth;

[0082] The capacity growth cycle represents the equivalent charge and discharge cycle number between two adjacent capacity growths, that is, the capacity growth cycle represents the equivalent charge and discharge cycle number from the capacity growth start cycle to the SOH closest to but not higher than the capacity growth start cycle.

[0083] In step S30, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity increase amplitude and capacity increase period predicted by the capacity increase prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity increase amplitude and capacity increase period.

[0084] A preferred embodiment of a lithium battery capacity attenuation trend prediction system of the present invention includes the following modules:

[0085] A normal attenuation prediction model training module, used to create a normal attenuation prediction model based on CNN (convolutional neural network) and LSTM (long short-term memory network), and train the normal attenuation prediction model;

[0086] A capacity growth prediction model training module, used to create a capacity growth prediction model based on DNN (deep neural network) and train the capacity growth prediction model;

[0087] A capacity decay trend prediction model generation module, used to generate a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model;

[0088] The capacity decay trend prediction module is used to predict the capacity decay trend of the lithium battery using the capacity decay trend prediction model.

[0089] The normal attenuation prediction model training module specifically includes:

[0090] A normal attenuation prediction model creation unit is used to create a normal attenuation prediction model based on CNN and LSTM; the CNN adopts a conv2d convolution kernel, which can extract the effective information of the feature itself and between the features, and can change the data dimension at the same time, so as to convolve the input data into the data dimension required by LSTM; the normal attenuation prediction model is used to predict the main attenuation trend of the lithium battery capacity;

[0091] The structure of the normal attenuation prediction model is as follows Figure 4 As shown in the figure, CNN is used to extract features from time series information, LSTM is used to extract features and model the correlation between time information, and finally the features extracted by the network are fully connected to output the predicted number of cycles corresponding to multiple specific attenuation values, thereby fitting a curve of the normal degradation trend of lithium batteries;

[0092] The input data can be charging data, discharging data or complete charging and discharging data of 1 or N consecutive cycles. Since the time series lengths of different cycles are different, and the normal attenuation prediction model can only process fixed-length inputs, it is necessary to truncate the data into a uniform length based on the time length threshold and perform corresponding normalization processing; the final data dimension input to the normal attenuation prediction model is NxCxHxW; where N represents batch_size; C represents the number of feature dimensions, such as only inputting V, I, and T, then C=3; H represents the number of input cycles; W represents the time series length of each feature of each cycle;

[0093] The formula for conv2d input and output is as follows:

[0094]

[0095] Among them, W intput Indicates the size of the input data; W output Indicates the size of the output data; K filterIndicates the size of the convolution kernel; Padding indicates the size of the padding; Stride indicates the step size of the convolution kernel;

[0096] LSTM has two transmission states, namely neuron state C t and the hidden state h t , LSTM first uses the current input x t and the previous hidden state h t-1 The splicing calculation results in four states, namely f t , i t , o t , the calculation formula is as follows:

[0097] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0098]

[0099] i t =σ(W t ·[h t-1 ,x t ]+b t );

[0100] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0101] Among them, f t Represents the forget gate, which is used to control which of the previous neuron states (cell states) need to be forgotten; represents the state update gate; i t Represents the input gate, used to control Which features of are used to update C t ;o t represents the output gate; σ() represents the activation function; W f , W c , W i , W o Both represent weight; b f , b c , b t , b o All represent bias; tanh() represents the activation function;

[0102] The output is calculated from four states, including the hidden layer:

[0103] h t =o t *tanh(C t ),

[0104] A first charge-discharge data acquisition unit, used to acquire a large amount of first charge-discharge data including at least voltage, current, resistance, temperature, power and charge-discharge time;

[0105] A charge and discharge data interception unit, used to set a duration threshold, sequentially intercept the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and fill the charge and discharge electronic data whose duration does not meet the duration threshold with 0;

[0106] The model training unit is used to train the normal attenuation prediction model using the charge and discharge electron data.

[0107] The volume growth prediction model training module is specifically used for:

[0108] Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

[0109] Since the rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy will affect the capacity growth, the second charge and discharge data is used to train the capacity growth prediction model.

[0110] The structure of the volume growth prediction model is as follows: Figure 5 As shown, it includes input layer, hidden layer and output layer; assuming that the input vector is <x 1 ,x 2 ,x 3 >, the hidden layer nodes are 4, 5, 6, 7, the output layer nodes are 8, 9, and the formula for calculating the hidden layer 4 from the input layer nodes is a4 = δ(W 14 *x 1 +W 24 *x 2 +W 34 *x 3 +b 1 ), where W 14 represents the weight between nodes 1 and 4; b 1 represents the bias of node 4; δ() represents the activation function, usually the sigmoid function, expressed as y = 1 / (1+e -x), in order to introduce nonlinearity to increase the network's expressive power; the calculation method of the nodes in the output layer is similar to that of the hidden layer, except that the input to the output layer becomes the nodes in the hidden layer.

[0111] The calculation formula of the volume growth amplitude is:

[0112] Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth;

[0113] The capacity growth cycle represents the equivalent charge and discharge cycle number between two adjacent capacity growths, that is, the capacity growth cycle represents the equivalent charge and discharge cycle number from the capacity growth start cycle to the SOH closest to but not higher than the capacity growth start cycle.

[0114] In the capacity decay trend prediction model generation module, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity increase amplitude and the capacity increase period predicted by the capacity increase prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity increase amplitude and the capacity increase period.

[0115] Experimental verification:

[0116] The lithium battery cycle life test was carried out using different models of battery 1 and battery 2. The normal attenuation prediction model was used when capacity proliferation was not considered, and the capacity attenuation trend prediction model was used when capacity proliferation was considered. The test index used was MAPE:

[0117]

[0118] Among them, y i Indicates the actual capacity value of the i-th charge and discharge cycle from the cycle starting point of capacity prediction; It represents the predicted capacity value of the ith charge and discharge cycle from the starting point of the capacity prediction cycle; n represents the total number of predicted cycles.

[0119] refer to Fig. 9 , the MAPE of battery 1 using the normal attenuation prediction model for inference is 0.17%, and the MAPE of battery 1 using the capacity attenuation trend prediction model for inference is 0.10%; Fig.10 , the MAPE of battery 2 using the normal attenuation prediction model for reasoning is 0.57%, and the MAPE of battery 2 using the capacity attenuation trend prediction model for reasoning is 0.45%.

[0120] It can be seen from the above test results that the normal attenuation prediction model that does not consider the battery capacity growth can only predict its normal degradation cycle, while the capacity decay trend prediction model adopted by the present invention can not only predict the conventional degradation trend, but also accurately predict the capacity growth amplitude and cycle of the battery according to its test standing time, capacity, temperature, SOH and other factors, which is more consistent with its actual attenuation process and achieves a certain improvement in the test indicator MAPE.

[0121] In summary, the advantages of the present invention are:

[0122] By creating a normal attenuation prediction model for predicting the main attenuation trend of lithium battery capacity, creating a capacity proliferation prediction model for predicting the capacity proliferation amplitude and capacity proliferation cycle, and training the normal attenuation prediction model and the capacity proliferation prediction model, and then generating a capacity attenuation trend prediction model based on the trained normal attenuation prediction model and the capacity proliferation prediction model, and finally using the capacity attenuation trend prediction model to predict the capacity attenuation trend of lithium batteries, that is, comprehensively considering the main attenuation trends and capacity proliferation in the degradation process of lithium batteries, reducing the error of capacity attenuation trend prediction, and thus greatly improving the accuracy of capacity attenuation trend prediction.

[0123] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the capacity attenuation trend of lithium batteries. Features: The steps include: Step S10, creating a normal attenuation prediction model based on CNN and LSTM, and training the normal attenuation prediction model; Step S20: creating a capacity growth prediction model based on DNN, and training the capacity growth prediction model; Step S30, generating a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model; Step S40, predicting the capacity decay trend of the lithium battery using the capacity decay trend prediction model; The step S10 specifically includes: Step S11, creating a normal attenuation prediction model based on CNN and LSTM; the CNN uses a conv2d convolution kernel to convolve the input data into the data dimension required by LSTM; Step S12, acquiring a large amount of first charge and discharge data including at least voltage, current, resistance, temperature, power, and charge and discharge time; Step S13, setting a duration threshold, sequentially intercepting the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and filling the charge and discharge electronic data whose duration does not meet the duration threshold with 0; Step S14: training a normal attenuation prediction model using the charge and discharge electronic data.

2. A method for predicting the capacity attenuation trend of a lithium battery as claimed in claim 1, Features: The step S20 is specifically as follows: Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

3. A method for predicting the capacity attenuation trend of a lithium battery as claimed in claim 2, Features: The calculation formula of the volume growth amplitude is: Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth; The capacity growth period represents the equivalent number of charge and discharge cycles between two adjacent capacity growths.

4. A method for predicting the capacity attenuation trend of a lithium battery as claimed in claim 1, Features: In step S30, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity increase amplitude and capacity increase period predicted by the capacity increase prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity increase amplitude and capacity increase period.

5. A lithium battery capacity attenuation trend prediction system, Features: Includes the following modules: A normal attenuation prediction model training module, used for creating a normal attenuation prediction model based on CNN and LSTM, and training the normal attenuation prediction model; A capacity growth prediction model training module, used to create a capacity growth prediction model based on DNN and train the capacity growth prediction model; A capacity decay trend prediction model generation module, used to generate a capacity decay trend prediction model based on the trained normal decay prediction model and the capacity growth prediction model; A capacity decay trend prediction module, used to predict the capacity decay trend of the lithium battery using the capacity decay trend prediction model; The normal attenuation prediction model training module specifically includes: A normal attenuation prediction model creation unit, used to create a normal attenuation prediction model based on CNN and LSTM; the CNN uses a conv2d convolution kernel to convolve input data into a data dimension required by LSTM; A first charge-discharge data acquisition unit, used to acquire a large amount of first charge-discharge data including at least voltage, current, resistance, temperature, power and charge-discharge time; A charge and discharge data interception unit, used to set a duration threshold, sequentially intercept the first charge and discharge data based on the duration threshold to obtain a plurality of charge and discharge electronic data, and fill the charge and discharge electronic data whose duration does not meet the duration threshold with 0; The model training unit is used to train the normal attenuation prediction model using the charge and discharge electron data.

6. A lithium battery capacity attenuation trend prediction system as claimed in claim 5, Features: The volume growth prediction model training module is specifically used for: Based on DNN, a capacity growth prediction model for predicting the capacity growth amplitude and capacity growth cycle is created; a large amount of second charge and discharge data including at least rest time, ambient temperature, SOH, battery capacity and charge and discharge strategy is obtained, and the capacity growth prediction model is trained using each of the second charge and discharge data.

7. A lithium battery capacity attenuation trend prediction system as claimed in claim 6, Features: The calculation formula of the volume growth amplitude is: Volume growth amplitude = (SOH after growth - SOH before growth) / SOH before growth; The capacity growth period represents the equivalent number of charge and discharge cycles between two adjacent capacity growths.

8. A lithium battery capacity attenuation trend prediction system as claimed in claim 5, Features: In the capacity decay trend prediction model generation module, the capacity decay trend prediction model is used to integrate the capacity predicted by the normal decay prediction model with the relative values ​​of the capacity increase amplitude and the capacity increase period predicted by the capacity increase prediction model, and then fit the output capacity decay curve with the absolute values ​​of the capacity increase amplitude and the capacity increase period.

Images