Health state estimation method of energy storage batteries with different aging paths based on deep learning

Abstract

The present invention specifically relates to a method for estimating the health status of energy storage batteries with different aging paths based on deep learning, including: obtaining battery charging voltage data under different working conditions of the energy storage battery, and inputting it into a battery health estimation model, and outputting the battery health status estimation value under the corresponding working condition; the battery health estimation model is constructed based on a deep neural network and trained twice; the first training: training the battery health estimation model through battery aging experimental data under a certain working condition; the second training: freezing the core network layer parameters of the pre-trained estimation model and calling its model parameters; then training and optimizing the estimation model through battery aging experimental data under different working conditions to adjust the parameters of its fully connected output layer, and obtaining the trained battery health estimation model. The present invention can be applied to battery health status estimation under a variety of working conditions without the need for large-scale training data, thereby reducing the difficulty and cost of training the estimation model and improving the scope of application of the estimation model.

Description

Technical Field

[0001] The present invention relates to the technical field of electrochemical energy storage management, and in particular to a method for estimating the health status of energy storage batteries with different aging paths based on deep learning. Background Art

[0002] Lithium-ion energy storage batteries have been widely used in the field of electrochemical energy storage due to their high energy density and wide temperature operating range. Since the decline of energy storage battery capacity and power performance is inevitable, the sudden change of energy storage battery performance has become the main cause of battery safety problems. Therefore, if the health status of lithium-ion energy storage batteries can be estimated more accurately, it will be of great significance for the management and protection of lithium-ion energy storage batteries.

[0003] The existing methods for estimating the health status of energy storage batteries mainly include estimation methods based on data-driven, model and model-data-driven hybrid. For example, the Chinese patent with publication number CN105891715A discloses "A method for estimating the health status of lithium-ion batteries", including: establishing a battery equivalent physical model; obtaining the battery SOC-OCV curve through battery pulse discharge, and obtaining the relationship between battery SOC and open circuit voltage; collecting battery voltage, current and temperature parameters; online identification of model parameters based on recursive least squares method to obtain battery ohmic internal resistance; and calculating battery health status using the estimated value of battery ohmic internal resistance. This existing solution is to realize the health status estimation of energy storage batteries based on the data-driven method.

[0004] The data-driven estimation method is a non-invasive method with low computational complexity and high precision. However, the data-driven estimation method requires high-precision and high-quality data, and there are problems such as model overfitting and underfitting. Therefore, most of the prior art adopts a hybrid estimation method based on models and model-data driven, and the model generally adopts a machine learning model. The applicant found that the working environment of the energy storage battery is complex and changeable, which requires the estimation model to be applicable to the energy storage battery health state estimation under various working conditions (aging path). However, when training an estimation model applicable to various working conditions, the prior art needs to obtain a large amount of historical data under different working conditions to train the estimation model in order to ensure the accuracy of the battery health state estimation of the estimation model, which makes the training of the estimation model difficult and the training cost high. Therefore, how to design a method for estimating the health state of an energy storage battery that can be applicable to battery health state estimation under various working conditions and can reduce the scale of training data for the estimation model is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a method for estimating the health status of energy storage batteries with different aging paths based on deep learning, so that it can be applicable to battery health status estimation under various working conditions without the need for large-scale training data, thereby reducing the training difficulty and cost of the estimation model, and increasing the scope of application of the estimation model, thereby improving the effect of energy storage battery health status estimation.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] The health status estimation method of energy storage batteries with different aging paths based on deep learning includes:

[0008] S1: Obtain battery charging voltage data under different working conditions of the energy storage battery;

[0009] S2: Input the battery charging voltage data of the energy storage battery under the corresponding working condition into the trained battery health estimation model, and output the estimated value of the battery health state under the corresponding working condition;

[0010] The battery health estimation model is built based on a deep neural network and trained twice;

[0011] First training: Train the battery health estimation model through battery aging experimental data under a certain working condition to obtain a pre-trained estimation model;

[0012] Second training: Freeze the core network layer parameters of the pre-trained estimation model and call its model parameters to obtain the optimized estimation model; then train the optimized estimation model with battery aging experimental data under different working conditions to adjust the parameters of its fully connected output layer to obtain the trained battery health estimation model;

[0013] S3: The estimated value of the battery health state under each working condition of the energy storage battery is used as its health state estimation result.

[0014] Preferably, in step S1, the operating conditions of the energy storage battery include different charge and discharge rates, different temperatures and different discharge depths.

[0015] Preferably, in step S2, the battery health estimation model is constructed based on a long short-term memory neural network, a gated recurrent neural network or a convolutional neural network.

[0016] Preferably, in step S2, the battery health estimation model is constructed based on a gated recurrent neural network, including a reset gate and an update gate;

[0017] The reset gate is represented as:

[0018] r t =σ(W r ·[h t-1 ,xt ]);

[0019]

[0020]

[0021]

[0022] Where: r t represents the output of the reset gate; σ represents the sigmoid activation function; W r Indicates resetting the gate weight; h t-1 represents the hidden state at time t-1; W represents the weight matrix; x t represents the input feature at time t; x represents the input feature; Represents the candidate hidden state of the input feature;

[0023] The update gate is expressed as:

[0024] z t =σ(W z ·[h t-1 ,x t ]);

[0025]

[0026] Where: z t represents the output of the update gate; W z represents the update gate weight; h t Indicates hidden state; h t-1 represents the hidden state at time t-1; x t Represents the input features at time t; Represents the candidate hidden state of the input feature.

[0027] Preferably, after the first training, the core network layer parameters of the pre-trained estimation model are frozen and the model parameters are called by the fine-tuning technology of transfer learning, and then the output of the pre-trained estimation model is connected to the fully connected layer to form a fully connected output layer and the activation of the hidden layer neurons is linearly changed by the following formula to obtain the optimized estimation model; finally, the optimized estimation model is trained for the second time;

[0028] y=aA T +b;

[0029] ReLU(y)=max(0,y);

[0030] Where: y represents the output of the fully connected output layer, that is, the estimated value of the battery health state; a represents the number of neurons; A represents the weight; T represents the transposed sign; b represents the bias; ReLU represents the linear rectified unit; max represents the maximum value.

[0031] Preferably, the loss function of the battery health estimation model adopts the following mean square error loss function;

[0032]

[0033] Where: y i Indicates the target value of the battery health status at time i; represents the estimated value of the battery health status at time i; N represents the number of samples.

[0034] Preferably, the optimizer of the battery health estimation model adopts the following adaptive moment estimation algorithm;

[0035] m t =β1m t-1 +(1-β1)g t

[0036]

[0037]

[0038]

[0039] Where: m t and m t-1 represents the sliding mean after smoothing of iterations t and t-1; t represents the number of iterations; β1 and β2 represent smoothing constants; g t represents the objective function gradient; V t represents the sliding mean of the squared gradient; w t and w t+1 Respectively represent the update parameters for iterations t and t+1; α t represents the learning rate for iteration t; ε = 10 -8 Indicates avoiding division by 0; and Represents the smoothing constant raised to the power of t.

[0040] Preferably, in step S2, the first training is performed by the following steps:

[0041] S201: Carry out battery cycle aging experiments under different working conditions and construct battery aging experimental data sets under various working conditions;

[0042] S202: Acquire battery charging voltage data from a battery aging experiment data set under a certain working condition as training data;

[0043] S203: training a battery health estimation model using primary training data;

[0044] S204: Calculate the training loss of the battery health state estimation value based on the battery health estimation model output: If the training loss reaches a preset value, save the model parameters to obtain a pre-trained estimation model; otherwise, return to step S202.

[0045] Preferably, in step S202, the battery charging voltage data of the first 20% capacity of the energy storage battery is used as training data.

[0046] Preferably, in step S2, the second training is performed by the following steps:

[0047] S211: Through the fine-tuning technology of transfer learning, freeze the core network layer parameters of the pre-trained estimation model and call its model parameters, then connect the output of the pre-trained estimation model to the fully connected layer to form a fully connected output layer and make linear changes to the hidden layer neuron activations to obtain an optimized estimation model;

[0048] S212: Acquire battery charging voltage data from a battery aging experiment data set under different working conditions as secondary training data;

[0049] S213: training the optimization estimation model through secondary training data to adjust the parameters of its fully connected output layer;

[0050] S214: Calculate the training loss of the battery health status estimation value based on the output of the optimized estimation model: If the training loss reaches a preset value, save the model parameters and obtain the trained battery health estimation model; otherwise, return to step S212.

[0051] The method for estimating the health status of energy storage batteries with different aging paths based on deep learning in the present invention has the following beneficial effects:

[0052] The battery health estimation model of the present invention is constructed based on a deep neural network and is trained twice. In the first training, a pre-trained estimation model is obtained by training battery aging experimental data under a specific working condition (aging path). Since it is for energy storage batteries of the same type and working under the same working condition, a pre-trained estimation model with higher accuracy can be obtained by training with relatively small-scale historical data (but the pre-trained estimation model at this time has poor versatility and cannot accurately estimate the health status of energy storage batteries under other chemical compositions or working conditions); in the second training, the pre-trained estimation model is used as the source model, and its core network layer parameters are frozen and its model parameters are called through the fine-tuning technology of transfer learning (the health status estimation between different working conditions of energy storage batteries has certain commonalities), and then the parameters of the fully connected output layer of the estimation model are adjusted through the battery aging experimental data under different working conditions. At this time, the estimation model can be adapted to a variety of working conditions by using relatively small-scale historical data, that is, the battery health estimation model of the present invention can be applicable to battery health status estimation under a variety of working conditions without the need for large-scale training data, which can reduce the training difficulty and cost of the estimation model, increase the scope of application of the estimation model, and thus improve the effect of energy storage battery health status estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0054] Figure 1 The logical block diagram of the health status estimation method of energy storage batteries with different aging paths based on deep learning;

[0055] Figure 2 A modeling flowchart for the health status estimation method of energy storage batteries with different aging paths based on deep learning;

[0056] Figure 3 The battery health status estimation result obtained based on the battery health estimation model of the present invention;

[0057] Figure 4 Estimation results of the health status of energy storage batteries under different working conditions;

[0058] Figure 5 Statistics for error analysis of health status estimation of energy storage batteries under different working conditions. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but only represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0060] It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. In the description of the present invention, it should be noted that the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invention product is usually placed when used, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance. In addition, the terms "horizontal", "vertical", etc. do not mean that the components are absolutely horizontal or suspended, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted. In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0061] The following is a further detailed description through specific implementation methods:

[0062] Example:

[0063] This embodiment discloses a method for estimating the health status of energy storage batteries with different aging paths based on deep learning.

[0064] like Figure 1 and Figure 2 As shown in the figure, the health status estimation method of energy storage batteries with different aging paths based on deep learning includes:

[0065] S1: Obtain battery charging voltage data under different working conditions of the energy storage battery;

[0066] S2: Input the battery charging voltage data of the energy storage battery under the corresponding working condition into the trained battery health estimation model, and output the estimated value of the battery health state under the corresponding working condition;

[0067] The battery health estimation model is built based on a deep neural network and trained twice;

[0068] First training: Train the battery health estimation model through battery aging experimental data under a certain working condition to obtain a pre-trained estimation model;

[0069] Second training: Freeze the hidden layer parameters (kernel network layer parameters) of the pre-trained estimation model and adjust the parameters of its input layer and output layer to obtain an optimized estimation model; then train the optimized estimation model with battery aging experimental data under different working conditions to adjust the parameters of its fully connected output layer to obtain a trained battery health estimation model;

[0070] S3: The estimated value of the battery health state under each working condition of the energy storage battery is used as its health state estimation result.

[0071] In this embodiment, the battery health status refers to the ratio of the maximum available capacity of the energy storage battery to the rated capacity, which is used to reflect the aging degree of the energy storage battery.

[0072] Specifically, the first training is carried out through the following steps:

[0073] S201: Carry out battery cycle aging experiments under different working conditions and construct battery aging experimental data sets under various working conditions;

[0074] In this embodiment, the battery aging test data includes data such as battery test terminal voltage and capacity.

[0075] S202: Acquire battery charging voltage data from a battery aging experiment data set under a certain working condition as training data;

[0076] In this embodiment, the battery charging voltage data of the first 20% capacity of the energy storage battery is used as training data, in order to improve the training effect of the model.

[0077] S203: training a battery health estimation model using primary training data;

[0078] S204: Calculate the training loss of the battery health state estimation value based on the battery health estimation model output: If the training loss reaches a preset value, save the model parameters to obtain a pre-trained estimation model; otherwise, return to step S202.

[0079] After the first training, the core network layer parameters of the pre-trained estimation model are frozen and its model parameters are called through the fine-tuning technology of transfer learning. Then, the output of the pre-trained estimation model is connected to the fully connected layer to form a fully connected output layer, and the activation of the hidden layer neurons is linearly changed through the following formula to obtain the optimized estimation model.

[0080] y=aA T +b;

[0081] ReLU(y)=max(0,y);

[0082] Where: y represents the output of the fully connected output layer, that is, the estimated value of the battery health state; a represents the number of neurons; A represents the weight; T represents the transposed sign; b represents the bias; ReLU represents the linear rectified unit; max represents the maximum value.

[0083] The second training is carried out through the following steps:

[0084] S211: Through the fine-tuning technology of transfer learning (Fine-Tune), freeze the core network layer parameters of the pre-trained estimation model and call its model parameters, then connect the output of the pre-trained estimation model to the fully connected layer to form a fully connected output layer and make linear changes to the hidden layer neuron activations to obtain an optimized estimation model;

[0085] S212: Acquire battery charging voltage data from a battery aging experiment data set under different working conditions as secondary training data;

[0086] S213: training the optimization estimation model through secondary training data to adjust the parameters of its fully connected output layer;

[0087] S214: Calculate the training loss of the battery health status estimation value based on the output of the optimized estimation model: If the training loss reaches a preset value, save the model parameters and obtain the trained battery health estimation model; otherwise, return to step S212.

[0088] The battery health estimation model of the present invention is constructed based on a deep neural network and is trained twice. In the first training, a pre-trained estimation model is obtained by training battery aging experimental data under a specific working condition (aging path). Since it is for energy storage batteries of the same type and working under the same working condition, a pre-trained estimation model with higher accuracy can be obtained by training with relatively small-scale historical data (but the pre-trained estimation model at this time has poor versatility and cannot accurately estimate the health status of energy storage batteries under other chemical compositions or working conditions); in the second training, the pre-trained estimation model is used as the source model, and its core network layer parameters are frozen and its model parameters are called through the fine-tuning technology of transfer learning (the health status estimation between different working conditions of energy storage batteries has certain commonalities), and then the parameters of the fully connected output layer of the estimation model are adjusted through the battery aging experimental data under different working conditions. At this time, the estimation model can be adapted to a variety of working conditions by using relatively small-scale historical data, that is, the battery health estimation model of the present invention can be applicable to battery health status estimation under a variety of working conditions without the need for large-scale training data, which can reduce the training difficulty and cost of the estimation model, increase the scope of application of the estimation model, and thus improve the effect of energy storage battery health status estimation.

[0089] At the same time, the battery charging voltage data used in the present invention is more applicable and is an end-to-end estimation method. It does not need to manually extract features that are strongly related to the battery health status, and can avoid tedious data preprocessing.

[0090] During the specific implementation process, the operating conditions of the energy storage battery include different charge and discharge rates, different temperatures and different discharge depths.

[0091] During the specific implementation process, the battery health estimation model is built based on a long short-term memory neural network (LSTM), a gated recurrent neural network (GRU) or a convolutional neural network (CNN).

[0092] In this embodiment, the battery health estimation model is constructed based on a gated recurrent neural network, including a reset gate and an update gate;

[0093] The reset gate is represented as:

[0094] r t =σ(W r ·[h t-1 ,x t ]);

[0095]

[0096]

[0097]

[0098] Where: r trepresents the output of the reset gate; σ represents the sigmoid activation function; W r Indicates resetting the gate weight; h t-1 represents the hidden state at time t-1; W represents the weight matrix; x t represents the input feature at time t; x represents the input feature; Represents the candidate hidden state of the input feature;

[0099] The update gate is expressed as:

[0100] z t =σ(W z ·[h t-1 ,x t ]);

[0101]

[0102] Where: z t represents the output of the update gate; W z represents the update gate weight; h t Indicates hidden state; h t-1 represents the hidden state at time t-1; x t Represents the input features at time t; Represents the candidate hidden state of the input feature.

[0103] The loss function of the battery health estimation model adopts the following mean square error loss function;

[0104]

[0105] Where: y i Indicates the target value of the battery health status at time i; represents the estimated value of the battery health status at time i; N represents the number of samples.

[0106] The optimizer of the battery health estimation model uses the following adaptive moment estimation algorithm;

[0107] m t =β1m t-1 +(1-β1)g t

[0108]

[0109]

[0110]

[0111] Where: m t and m t-1represents the sliding mean after smoothing of iterations t and t-1; t represents the number of iterations; β1 and β2 represent smoothing constants; g t represents the objective function gradient; V t represents the sliding mean of the squared gradient; w t and w t+1 Respectively represent the update parameters for iterations t and t+1; α t represents the learning rate for iteration t; ε = 10 -8 Indicates avoiding division by 0; and Represents the smoothing constant raised to the power of t.

[0112] The present invention constructs a battery health estimation model based on a gated recurrent neural network, adopts a mean square error loss function as a loss function, adopts an adaptive moment estimation algorithm as an optimizer, and connects the output of the gated recurrent neural network to a fully connected layer, so that a higher-precision estimation model can be trained through relatively small-scale historical data, and some parameters of the estimation model under different working conditions can be adjusted through battery aging experimental data under different working conditions, that is, it can be applicable to battery health status estimation under a variety of working conditions without the need for large-scale training data, thereby reducing the training difficulty and cost of the estimation model and improving the scope of application of the estimation model.

[0113] In order to better illustrate the advantages of the technical solution of the present invention, the following experiments are disclosed in this embodiment.

[0114] Figure 3 Comparison between the battery health status estimation results obtained by the battery health estimation model and the experimental values. Figure 3 It can be seen that the battery health estimation model of the present invention can obtain smoother estimation results.

[0115] Figure 4 Comparison between the estimated value of the energy storage battery health status under different working conditions estimated by the battery health estimation model and the experimental value. Figure 4 It can be seen that the battery health estimation model of the present invention can accurately estimate the health status of energy storage batteries under different working conditions with relatively small-scale training data.

[0116] Figure 5 Error analysis of estimated health status of energy storage batteries under different working conditions.

[0117] Figure 5 (a) is the root mean square error statistics of health state estimation under 1C-CC charging, dynamic discharge, and 40°C conditions. The root mean square error of health state estimation of the battery health estimation model is 0.0021, which is lower than other neural network models.

[0118] Figure 5(b) is the mean absolute error statistics of health state estimation under 1C-CC charging, dynamic discharge and 40°C conditions. The mean absolute error of health state estimation of the battery health estimation model is 0.0016, which is lower than that of other neural network models.

[0119] Figure 5 (c) is the root mean square error statistics of health state estimation under 2C-CC-CV charging, 1C-CC discharging and 24°C conditions. The root mean square error of health state estimation of the battery health estimation model is 0.0181, which is smaller than that of the directly trained neural network model.

[0120] Figure 5 (d) is the mean absolute error statistics of health state estimation under 2C-CC-CV charging, 1C-CC discharging and 24°C conditions. The mean absolute error of health state estimation by the battery health estimation model is 0.0148, which is smaller than that of the directly trained neural network model.

[0121] Figure 5 (e) The root mean square error statistics of health state estimation under 2C-CC-CV charging, 1C-CC discharging, and 4°C conditions. The root mean square error of health state estimation of the battery health estimation model is 0.0376, which is smaller than that of the directly trained neural network model.

[0122] Figure 5 (f) is the mean absolute error statistics of health state estimation under 2C-CC-CV charging, 1C-CC discharging and 4°C conditions. The mean absolute error of health state estimation of the battery health estimation model is 0.0261, which is smaller than that of the directly trained neural network model.

[0123] Experimental results show that under different operating cycle experimental conditions, the battery health estimation model of the present invention can accurately estimate the health status of the energy storage battery under different operating conditions.

[0124] Compared with the neural network model trained based on big data, the deep neural network model established based on smaller data can reliably and accurately estimate the health status of lithium-ion energy storage batteries with only smaller data. Compared with the neural network model trained under a single working condition, it cannot accurately estimate the battery status when facing energy storage batteries under different working conditions. However, the source deep neural network model established under a certain working condition by the method proposed in the present invention can accurately estimate the health status of energy storage batteries under different working conditions with only a small amount of data and without changing the core layer parameters of the model.

[0125] The method proposed in the present invention overcomes the problem of inaccurate estimation in the absence of prior knowledge when facing energy storage batteries under different usage climates, charging modes and other conditions. The method of estimating the health status of energy storage batteries under different working conditions based on deep learning is adopted, and the battery characteristics extracted by the neural network layer are used to estimate the battery status, which can realize fast and accurate estimation of the health status of lithium-ion energy storage batteries under different usage environments.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A method for estimating the health status of energy storage batteries with different aging paths based on deep learning, characterized in that: include: S1: Obtain battery charging voltage data under different working conditions of the energy storage battery; S2: Input the battery charging voltage data of the energy storage battery under the corresponding working condition into the trained battery health estimation model, and output the estimated value of the battery health state under the corresponding working condition; The battery health estimation model is built based on a deep neural network and trained twice; First training: Train the battery health estimation model through battery aging experimental data under a certain working condition to obtain a pre-trained estimation model; Second training: Freeze the core network layer parameters of the pre-trained estimation model and call its model parameters to obtain the optimized estimation model; then train the optimized estimation model with battery aging experimental data under different working conditions to adjust the parameters of its fully connected output layer to obtain the trained battery health estimation model; In step S2, a battery health estimation model is constructed based on a gated recurrent neural network, including a reset gate and an update gate; The reset gate is represented as: r t =σ(W r ·[h t-1 ,x t ]); Where: r t represents the output of the reset gate; σ represents the sigmoid activation function; W r Indicates resetting the gate weight; h t-1 represents the hidden state at time t-1; W represents the weight matrix; x t represents the input feature at time t; x represents the input feature; Represents the candidate hidden state of the input feature; The update gate is expressed as: z t =σ(W z ·[h t-1 ,x t ]); Where: z t represents the output of the update gate; W z represents the update gate weight; h t Indicates hidden state; h t-1 represents the hidden state at time t-1; x t Represents the input features at time t; Represents the candidate hidden state of the input feature; After the first training, the kernel network layer parameters of the pre-trained estimation model are frozen and its model parameters are called through the fine-tuning technology of transfer learning. Then, the output of the pre-trained estimation model is connected to the fully connected layer to form a fully connected output layer, and the activation of the hidden layer neurons is linearly changed through the following formula to obtain the optimized estimation model. Finally, the optimized estimation model is trained for the second time. y=aA T +b; ReLU(y)=max(0,y); Where: y represents the output of the fully connected output layer, that is, the estimated value of the battery health state; a represents the number of neurons; A represents the weight; T represents the transposed sign; b represents the bias; ReLU represents the linear rectified unit; max represents the maximum value; S3: The estimated value of the battery health state under each working condition of the energy storage battery is used as its health state estimation result.

2. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 1, characterized in that: In step S1, the operating conditions of the energy storage battery include different charge and discharge rates, different temperatures and different discharge depths.

3. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 1, characterized in that: In step S2, the battery health estimation model is constructed based on a long short-term memory neural network, a gated recurrent neural network or a convolutional neural network.

4. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 1, characterized in that: The loss function of the battery health estimation model adopts the following mean square error loss function; Where: y i Indicates the target value of the battery health status at time i; represents the estimated value of the battery health status at time i; N represents the number of samples.

5. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 1, characterized in that: The optimizer of the battery health estimation model uses the following adaptive moment estimation algorithm; m t =β1m t-1 +(1-β1)g t Where: m t and m t-1 represents the sliding mean after smoothing of iterations t and t-1; t represents the number of iterations; β1 and β2 represent smoothing constants; g t represents the objective function gradient; V t represents the sliding mean of the squared gradient; w t and w t+1 Respectively represent the update parameters for iterations t and t+1; α t represents the learning rate for iteration t; ε = 10 -8 Indicates avoiding division by 0; and Represents the smoothing constant raised to the power of t.

6. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 1, characterized in that: In step S2, the first training is performed through the following steps: S201: Carry out battery cycle aging experiments under different working conditions and construct battery aging experimental data sets under various working conditions; S202: Acquire battery charging voltage data from a battery aging experiment data set under a certain working condition as training data; S203: training a battery health estimation model using primary training data; S204: Calculate the training loss of the battery health state estimation value based on the battery health estimation model output: If the training loss reaches a preset value, save the model parameters to obtain a pre-trained estimation model; otherwise, return to step S202.

7. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 6, characterized in that: In step S202, the battery charging voltage data of the first 20% capacity of the energy storage battery is used as training data.

8. The method for estimating the health status of energy storage batteries with different aging paths based on deep learning as claimed in claim 6, characterized in that: In step S2, the second training is performed through the following steps: S211: Through the fine-tuning technology of transfer learning, freeze the core network layer parameters of the pre-trained estimation model and call its model parameters, then connect the output of the pre-trained estimation model to the fully connected layer to form a fully connected output layer and make linear changes to the hidden layer neuron activations to obtain an optimized estimation model; S212: Acquire battery charging voltage data from a battery aging experiment data set under different working conditions as secondary training data; S213: training the optimization estimation model through secondary training data to adjust the parameters of its fully connected output layer; S214: Calculate the training loss of the battery health status estimation value based on the output of the optimized estimation model: If the training loss reaches a preset value, save the model parameters and obtain the trained battery health estimation model; otherwise, return to step S212.

Images