Battery state-of-charge and state-of-health joint estimation method and system based on double-branch neural network

Abstract

The invention discloses a battery state-of-charge and state-of-health joint estimation method and system based on a double-branch neural network, and relates to the technical field of battery management. According to the method, data of five dimensions including a voltage value, a current value, a temperature value, a discharge voltage segmentation characteristic and a discharge voltage curve area characteristic are selected for a battery prediction task, and cooperative prediction estimation optimization is carried out on the state of charge and the state of health by means of a double-branch neural network with double-branch estimation and cross sharing characteristics. The key information in the discharge process is effectively extracted, and the coupling relation between the charge state and the health state is fully considered, so that the prediction precision and robustness are effectively improved. According to the invention, the problem of poor prediction and estimation precision of the existing method is solved.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically, to: 1. a method for jointly estimating the state of charge and state of health of a battery based on a dual-branch neural network; 2. a system for jointly estimating the state of charge and state of health of a battery based on a dual-branch neural network. Background Technology

[0002] With the rapid development of new energy technologies, energy storage technology plays a crucial role in power systems, electric vehicles, and renewable energy integration. As the core component of energy storage systems, the accurate assessment of battery performance directly affects the system's safety, economy, and reliability. In battery management systems, the battery's state of charge (SOC) and state of health (SOH) are two key parameters: SOC reflects the battery's remaining usable capacity, while SOH characterizes the battery's aging level. Accurate estimation of these two parameters not only affects the battery's charge and discharge control strategies but also has significant implications for battery life prediction, maintenance decisions, and overall system performance optimization.

[0003] In recent years, data-driven methods have attracted widespread attention due to their powerful nonlinear modeling capabilities. Existing research uses neural networks to establish a mapping relationship between inputs (such as voltage, current, and temperature) and outputs (SOC or SOH) by learning from large amounts of experimental data. While this avoids complex mechanistic modeling, the prediction accuracy is unsatisfactory in practical applications. Summary of the Invention

[0004] Therefore, it is necessary to address the problem of poor prediction and estimation accuracy of existing methods by providing a joint estimation method and system for battery state of charge and health based on a dual-branch neural network.

[0005] This invention is achieved using the following technical solution: In a first aspect, the present invention discloses a method for jointly estimating the state of charge (SOC) and state of health (SHealth) of a battery based on a dual-branch neural network, comprising: inputting the five-dimensional feature vector F5[t] of the target battery at the current time t into a trained dual-branch neural network for processing to predict the SOC of the target battery at time t. final [t]、Health Status SOH final [t].

[0006] Among them, the data types of F5[t] include: standardized voltage value, current value, temperature value, discharge voltage segmentation characteristics, and discharge voltage curve area characteristics.

[0007] The dual-branch neural network includes: a preliminary prediction unit for SOC, a preliminary prediction unit for SOH, and a joint estimation unit.

[0008] The initial SOC prediction unit is used to: extract 32-dimensional local features from F5[t], and then perform feature transformation, dimensionality compression, and preliminary estimation on the 32-dimensional local features to obtain the SOC. pre [t]. The initial prediction unit of SOH is used to: map F5[t] into 64-dimensional hidden features, and then refine them through three levels of feature extraction to obtain SOH. pre [t]. The joint estimation part is used to: convert SOC pre [t] is concatenated with F5[t] to form the first type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOH. final [t]; SOH pre [t] is concatenated with F5[t] to form the second type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOC. final [t].

[0009] This method for jointly estimating the state of charge and health of a battery based on a dual-branch neural network implements the method or process according to embodiments of this disclosure.

[0010] Secondly, the present invention discloses a battery state of charge and health state joint estimation system based on a dual-branch neural network, which uses the battery state of charge and health state joint estimation method based on a dual-branch neural network disclosed in the first aspect.

[0011] The battery state of charge and health state joint estimation system based on a dual-branch neural network includes: a state prediction module.

[0012] The state prediction module is used to: input the five-dimensional feature vector F5[t] of the target battery at the current time t into a trained dual-branch neural network for processing to predict the state of charge (SOC) of the target battery at time t. final [t]、Health Status SOH final [t].

[0013] This battery state of charge and health state joint estimation system based on a dual-branch neural network implements the method or process according to embodiments of this disclosure.

[0014] Thirdly, the present invention discloses a computer program product comprising a computer program. When executed by a processor, the computer program implements the steps of the battery state of charge and state of health joint estimation method based on a dual-branch neural network as disclosed in the second aspect.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention selects five dimensions of data for battery prediction tasks: voltage, current, temperature, discharge voltage segmentation features, and discharge voltage curve area features. It relies on a bi-branch neural network with bi-branch estimation and cross-shared features to perform collaborative prediction and estimation optimization of SOC and SOH. This not only effectively extracts key information in the discharge process, but also fully considers the coupling relationship between SOC and SOH, thereby effectively improving the accuracy and robustness of the prediction. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 The flowchart shows the joint estimation method of battery state of charge and state of health based on a dual-branch neural network provided in Embodiment 1 of the present invention. Figure 2 for Figure 1 Structural diagram of the preliminary SOC prediction unit; Figure 3 for Figure 1 Structural diagram of the preliminary SOH prediction section; Figure 4 for Figure 1 Structure diagram of the Joint Estimation Department. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that when a component is said to be "installed on" another component, it can be directly on the other component or it may be in a component that is centered on it. When a component is said to be "set on" another component, it can be directly set on the other component or it may also be in a component that is centered on it. When a component is said to be "fixed to" another component, it can be directly fixed to the other component or it may also be in a component that is centered on it.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] As mentioned in the background section, existing methods for predicting SOC or SOH using neural networks suffer from poor prediction accuracy. Analysis reveals that this is because existing methods isolate SOC and SOH, performing single estimations (i.e., predicting only SOC or only SOH), which fails to adequately consider the interaction between the two, leading to error accumulation and divergence.

[0022] For the reasons mentioned above, this invention provides a method and system for jointly estimating the state of charge (SOC) and state of health (SOH) of a battery based on a dual-branch neural network. This method can comprehensively consider the coupling relationship between SOC and SOH and the influence of the external environment, and achieve joint estimation, thereby improving the estimation accuracy and providing a more reliable basis for state assessment of the battery management system.

[0023] Example 1 See Figure 1 This document presents a flowchart of the joint estimation method for battery state of charge and state of health based on a dual-branch neural network provided in Embodiment 1. As described above, this joint estimation method is used to predict the SOC and SOH of a target battery during the discharge process.

[0024] See Figure 1 This joint estimation method can be designed to include the following steps: Step 1: Obtain the five-dimensional information of the target battery at the current time t; the five-dimensional information includes: voltage value, current value, temperature value, discharge voltage segmentation characteristics, and discharge voltage curve area characteristics.

[0025] Voltage, current, and temperature are three types of directly acquired data. Generally, the target battery is equipped with appropriate high-precision sensors to simultaneously collect voltage, current, and temperature values ​​during battery operation; it is recommended that the sampling frequency be no less than 10Hz to ensure data continuity.

[0026] The segmented characteristics of the discharge voltage are extracted from voltage changes using a dynamic segmentation method—specifically including the following steps: S101 divides the discharge voltage range of the target battery into several characteristic voltage intervals. Generally, it is recommended to divide the discharge voltage range into three characteristic voltage intervals: high voltage range, plateau range, and low voltage range. The specific span of each interval is dynamically selected based on the electrochemical characteristics of the target battery. For example, for ternary lithium batteries, the high voltage range is 4.2V~4.0V, the plateau range is 4.0V~3.7V, and the low voltage range is 3.7V~3.0V.

[0027] S102, Calculate the duration of the target battery voltage crossing each characteristic voltage range to serve as the discharge voltage segmentation feature. Specifically, a sliding window method can be used to record the duration of the target battery voltage crossing each characteristic voltage range, thereby forming the discharge voltage segmentation feature.

[0028] The area under the discharge voltage curve is extracted from voltage changes using a numerical integration method—specifically including the following steps: S103, numerically integrate the discharge voltage curve composed of the target battery voltage and time, and use the integration result as the area feature of the discharge voltage curve.

[0029] It should be noted that, in order to improve the accuracy of numerical integration, an adaptive Simpson integration algorithm can also be used, and the integration step size can be dynamically adjusted within the range of 0.1 to 1 s according to the voltage change rate.

[0030] Of course, compensation can also be introduced to improve feature accuracy—the discharge voltage curve composed of the target battery voltage and time is numerically integrated; the integration result is compensated based on temperature correction, and the compensation result is used as the area feature of the discharge voltage curve.

[0031] Specifically, the compensation formula is as follows: ; ; In the formula, V_area' represents the compensation result; V_area represents the integral result; Ea represents the activation energy; R represents the gas constant; ΔTem represents the temperature term; Tem represents the real-time temperature; and Tref represents the reference temperature.

[0032] Step 2: Standardize the five-dimensional information at the current time t and form the five-dimensional feature vector F5[t] of the target battery at the current time t.

[0033] Step two, in order to eliminate inconsistencies in data units, specifically includes: First, outliers are removed from the five-dimensional information using the 3σ criterion, and then a moving median filter is used to smooth out data fluctuations. Then, Min-Max standardization is applied to the voltage value, temperature value, discharge voltage segmentation characteristics, and discharge voltage curve area characteristics, and Z-score standardization is applied to the current.

[0034] Therefore, the data types of F5[t] include: standardized voltage value, current value, temperature value, discharge voltage segmentation characteristics, and discharge voltage curve area characteristics.

[0035] Step 3: Input F5[t] into the trained dual-branch neural network for processing to predict the target battery's state of charge (SOC) at time t. final [t]、Health Status SOH final [t].

[0036] In summary, the dual-branch neural network is used to: extract 32-dimensional local features from F5[t], and then perform feature transformation, dimensionality compression, and preliminary estimation on the 32-dimensional local features to obtain the SOC. pre [t]; will SOC pre [t] is concatenated with F5[t] to form the first type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOH. final [t]; Map F5[t] into 64-dimensional hidden features, and then refine them through three levels of feature extraction to obtain SOH. pre [t]; SOH pre [t] is concatenated with F5[t] to form the second type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOC. final [t].

[0037] See Figure 1 A dual-branch neural network can be divided into: a preliminary prediction unit for SOC, a preliminary prediction unit for SOH, and a joint estimation unit.

[0038] ① For the initial SOC prediction unit, it is used to: extract 32-dimensional local features from F5[t], and then perform feature transformation, dimensionality compression, and preliminary estimation on the 32-dimensional local features to obtain the SOC. pre [t].

[0039] See Figure 2 The initial prediction unit of SOC consists of: one one-dimensional convolutional layer, one activation function layer, and three fully connected layers.

[0040] In the SOC preliminary forecasting section: One-dimensional convolutional layers are used to perform one-dimensional convolution on F5[t]. The activation function layer 1 is used to process the output of the one-dimensional convolutional layer through the ReLU activation function to obtain 32-dimensional local features; The first fully connected layer is used to perform feature transformation on 32-dimensional local features; The second fully connected layer is used to compress the dimension of the output of the first fully connected layer. The third fully connected layer is used to perform a preliminary estimation of the output of the second fully connected layer and output the SOC. pre [t]; It should be noted that the hidden layers of the three fully connected layers all use the ReLU activation function.

[0041] ② For the preliminary prediction unit of SOH, it is used to: map F5[t] into 64-dimensional hidden features and refine them through three levels of feature extraction to obtain SOH. pre [t].

[0042] See Figure 3 The initial prediction unit of SOH consists of: one LSTM layer, one activation function layer, and three fully connected layers.

[0043] In the preliminary SOH prediction section: LSTM layers are used to capture temporal dependencies in F5[t] through a long short-term memory network; The second activation function layer is used to process the output of the LSTM layer using the ReLU activation function to obtain 64-dimensional hidden features; The first fully connected layer is used to extract features from the 64-dimensional hidden features; The second fully connected layer is used to extract features from the output of the first fully connected layer. The third fully connected layer 2 is used to extract features from the output of the second fully connected layer 2 and output SOH. pre [t]; It should be noted that the hidden layers of the three fully connected layers all use the ReLU activation function.

[0044] ③ Regarding the joint estimation department, it is used to: adjust the SOC pre [t] is concatenated with F5[t] to form the first type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOH. final [t]; SOH pre [t] is concatenated with F5[t] to form the second type of six-dimensional feature input, and then subjected to two levels of feature extraction to obtain SOC. final [t].

[0045] See Figure 4 The joint estimation department consists of three layers: two splicing layers and four fully connected layers.

[0046] In the joint estimation department: The first splicing layer is used to integrate the SOC. pre [t] is concatenated with F5[t] to form the first type of six-dimensional feature input; The first fully connected layer 3 is used to extract features from the first type of six-dimensional feature input; The second fully connected layer 3 is used to extract features from the output of the first fully connected layer 3 and output SOH. final [t]; The second splicing layer is used to integrate SOH. pre [t] is concatenated with F5[t] to form a second type of six-dimensional feature input; The third fully connected layer is used to extract features from the second type of six-dimensional feature input; The fourth fully connected layer 3 is used to extract features from the output of the third fully connected layer 3 and output the SOC. final [t]; It should be noted that the hidden layers of the four fully connected third layers all use the ReLU activation function followed by a Dropout layer; the output layer of the second fully connected third layer uses the linear activation function; and the output layer of the fourth fully connected third layer uses the Sigmoid activation function.

[0047] The aforementioned dual-branch neural network first extracts local features through one branch to capture the high-frequency dynamic changes of SOC, and performs temporal pattern recognition through the other branch to capture the slow degradation trend of SOH. It then achieves deep information fusion through a bidirectional information cross-fusion mechanism, effectively capturing the coupling relationship between SOC and SOH, thereby improving the joint estimation effect of SOC and SOH.

[0048] As mentioned above, this joint estimation method uses a pre-trained two-branch neural network—the model is trained on a sample dataset with real labels (e.g., NASA public dataset), and a jointly optimized loss function Loss is constructed. The model is trained with the Adam optimizer (learning rate recommended to be set to 0.001), and L2 regularization is introduced to prevent overfitting until training is complete.

[0049] The expression for Loss is: Loss = α × MSE_SOC + β × MSE_SOH; In the formula, α and β represent adjustable weighting coefficients; α + β = 1; MSE_SOC represents the mean square error of SOC prediction; MSE_SOH represents the mean square error of SOH prediction. In this Example 1, α is 0.7 and β is 0.3.

[0050] Furthermore, it should be noted that the aforementioned dual-branch neural network is a lightweight design, and its parameters are fixed after training—it can be directly deployed on edge computing devices and can complete local real-time inference without relying on cloud services, which has significant engineering application value.

[0051] This embodiment 1 also introduces three existing methods (including: 1. using a CNN model to predict SOC or SOH; 2. using an LSTM model to predict SOC or SOH; 3. using a BP neural network to predict SOC or SOH) for simulation verification and comparison to illustrate the effectiveness and superiority of this joint estimation method.

[0052] The results show that the proposed joint estimation method outperforms both SOC and SOH predictions. In SOC prediction, this joint estimation method not only has lower error, but also far surpasses other models in error fluctuation and extreme value control, indicating that it is suitable for battery management systems with extremely high prediction accuracy requirements.

[0053] In SOH prediction, this joint estimation method also exhibits extremely low average error and volatility. At the same time, it comprehensively outperforms other models in terms of error distribution range and extreme value control, indicating that it is also the best choice for SOH estimation.

[0054] Example 2 This embodiment 2 provides a battery state of charge and health state joint estimation system based on a dual-branch neural network, which uses the battery state of charge and health state joint estimation method based on a dual-branch neural network provided in embodiment 1.

[0055] See Figure 2 The battery state of charge and health state joint estimation system based on a dual-branch neural network includes: a state prediction module; of course, it may also include: an information acquisition module and a standardization processing module.

[0056] The information acquisition module is used to acquire the five-dimensional information of the target battery at the current time t. The state prediction module is used to standardize the five-dimensional information at the current time t and form the five-dimensional feature vector F5[t] of the target battery at the current time t. The state prediction module is used to input the five-dimensional feature vector F5[t] of the target battery at the current time t into a trained dual-branch neural network for processing to predict the state of charge (SOC) of the target battery at time t. final [t]、Health Status SOH final [t].

[0057] Since this system uses the battery state of charge and health state joint estimation method based on dual-branch neural network in Example 1, it has the same effect, so it will not be repeated here.

[0058] Example 3 This embodiment 3 discloses a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the battery state of charge and health state joint estimation method based on a dual-branch neural network disclosed in embodiment 1.

[0059] The computer equipment can be either a mobile terminal or a fixed terminal. Examples of the former include mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (such as in-vehicle navigation terminals); examples of the latter include digital TVs and desktop computers.

[0060] This embodiment 3 also discloses a readable storage medium that stores computer program instructions. When the computer program instructions are read and run by a processor, the steps of the battery state of charge and health state jointly estimated method based on dual-branch neural network disclosed in embodiment 1 are executed.

[0061] The readable storage medium may include, but is not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

[0062] This embodiment 3 also discloses a computer program product, including a computer program. When executed by a processor, this computer program implements the steps of the battery state of charge and health state joint estimation method based on a dual-branch neural network disclosed in embodiment 1.

[0063] It should be noted that the computer program used to execute the above can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, C++, and Python—as well as conventional procedural programming languages—such as C or similar languages. The computer program can be executed entirely on the user's computer, partially on the user's computer, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer through any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN).

[0064] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A battery state of charge and state of health joint estimation method based on a two-branch neural network, characterized in that, It comprises: inputting the five-dimensional feature vector F5[t] of the target battery at the current time t into the trained double-branch neural network for processing to predict the state of charge SOC of the target battery at t final [t], the state of health SOH final [t]; The data type of F5[t] includes: a voltage value after standardization processing, a current value, a temperature value, a discharge voltage segmentation feature, and a discharge voltage curve area feature. The double-branch neural network includes: a SOC preliminary prediction unit configured to extract 32-dimensional local features from F5[t], and sequentially perform feature transformation, dimension compression, and preliminary estimation on the 32-dimensional local features to obtain SOC pre [t]; a SOH preliminary prediction unit configured to map F5[t] into 64-dimensional hidden features and to perform three levels of feature refinement to obtain SOH pre [t]; and a joint estimation unit configured to: concatenate SOC pre [t] with F5[t] to form a first type of six-dimensional feature input, and perform two-stage feature refinement to obtain SOH final [t]; and concatenate SOH pre [t] with F5[t] to form a second type of six-dimensional feature input, and perform two-stage feature refinement to obtain SOC final [t].

2. The battery state of charge and state of health joint estimation method based on a two-branch neural network according to claim 1, characterized in that, The method for obtaining F5[t] includes: Step one: obtaining five-dimensional information of the target battery at the current time t, wherein the five-dimensional information includes: a voltage value, a current value, a temperature value, a discharge voltage segmentation feature, and a discharge voltage curve area feature. Step two: standardizing the five-dimensional information at the current time t, and forming F5[t].

3. The battery state of charge and state of health joint estimation method based on a two-branch neural network according to claim 2, characterized in that, The method for obtaining the discharge voltage segmentation feature includes: dividing the discharge voltage range of the target battery into a plurality of characteristic voltage intervals; counting the duration of the target battery voltage passing through each characteristic voltage interval as the discharge voltage segmentation feature.

4. The battery state-of-charge and state-of-health joint estimation method based on a two-branch neural network according to claim 2, characterized in that, The method for obtaining the discharge voltage curve area feature includes: performing numerical integration on the discharge voltage curve composed of the target battery voltage and time, and taking the integration result as the discharge voltage curve area feature; or, performing numerical integration on the discharge voltage curve composed of the target battery voltage and time; compensating the integration result based on temperature correction, and taking the compensation result as the discharge voltage curve area feature.

5. The battery state-of-charge and state-of-health joint estimation method based on a two-branch neural network according to claim 2, characterized in that, The SOC preliminary prediction part includes: one one-dimensional convolution layer, one activation function layer one, and three fully connected layers one. In the SOC preliminary prediction unit: a one-dimensional convolution layer is used to perform one-dimensional convolution on F5[t]; an activation function layer one is used to process the output of the one-dimensional convolution layer by a Relu activation function to obtain 32-dimensional local features; a first full connection layer one is used to perform feature transformation on the 32-dimensional local features; a second full connection layer one is used to perform dimension compression on the output of the first full connection layer one; and a third full connection layer one is used to perform preliminary estimation on the output of the second full connection layer one and output the SOC pre [t]; and the hidden layers of the three full connection layers one all adopt a Relu activation function.

6. The battery state-of-charge and state-of-health joint estimation method based on a two-branch neural network according to claim 1, characterized in that, The SOH preliminary prediction part includes: one LSTM layer, one activation function layer two, and three fully connected layers two. In the SOH preliminary prediction unit: the LSTM layer is used to capture the time series dependence in F5[t] through a long short-term memory network; the activation function layer two is used to process the output of the LSTM layer through a Relu activation function to obtain 64-dimensional hidden features; the first fully connected layer two is used to extract features of the 64-dimensional hidden features; the second fully connected layer two is used to extract features of the output of the first fully connected layer two; the third fully connected layer two is used to extract features of the output of the second fully connected layer two and output SOH pre [t]; the hidden layers of the three fully connected layers two all adopt a Relu activation function.

7. The battery state-of-charge and state-of-health joint estimation method based on a two-branch neural network according to claim 1, characterized in that, The joint estimation part includes: two splicing layers and four fully connected layers three. In the joint estimation unit: the first splicing layer is used for splicing SOC pre [t] and F5[t] to form the first type of six-dimensional feature input; the first fully connected layer three is used for feature extraction on the first type of six-dimensional feature input; the second fully connected layer three is used for feature extraction on the output of the first fully connected layer three, and outputs SOH final [t]; the second splicing layer is used for splicing SOH pre [t] and F5[t] to form the second type of six-dimensional feature input; the third fully connected layer three is used for feature extraction on the second type of six-dimensional feature input; the fourth fully connected layer three is used for feature extraction on the output of the third fully connected layer three, and outputs SOC final [t]; wherein the hidden layers of the four fully connected layer threes all adopt a Relu activation function followed by a Dropout layer; the output layer of the second fully connected layer three adopts a linear activation function; and the output layer of the fourth fully connected layer three adopts a Sigmoid activation function.

8. The battery state-of-charge and state-of-health joint estimation method based on a two-branch neural network according to claim 1, characterized in that, The loss function Loss used by the double-branch neural network during training is: Loss=α×MSE_SOC+β×MSE_SOH; wherein, α and β represent adjustable weight coefficients; α+β=1; MSE_SOC represents the mean square error of SOC prediction; and MSE_SOH represents the mean square error of SOH prediction. 9.A battery state of charge and state of health joint estimation system based on a two-branch neural network, characterized in that, The battery state of charge and health state joint estimation method based on the double-branch neural network includes: The computer program, when executed by a processor, implements the steps of the battery state of charge and health state joint estimation method based on the double-branch neural network. a state prediction module, configured to: input a five-dimensional feature vector F5[t] of the target battery at a current time t into the trained double-branch neural network for processing to predict a state of charge SOC of the target battery at t final [t], a health state SOH final [t].

10. A computer program product comprising a computer program, characterised in that, The computer program, when executed by a processor, implements the steps of the battery state of charge and health state joint estimation method based on the double-branch neural network.

Images