A method for predicting the state of health of lithium batteries based on hybrid cyclic stacked networks
The lithium battery health state prediction method using a hybrid recurrent stacked network solves the problems of complexity and insufficient prediction accuracy in lithium-ion battery SoH estimation by training a hybrid hidden layer of RNN and GRU layers using SoH data, and achieves higher accuracy in lithium battery health state prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- POWERCHINA HUADONG ENG CORP LTD
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-02
Smart Images
Figure CN122131184A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of lithium battery state prediction technology, specifically to a method for predicting the state of health of lithium batteries based on a hybrid cyclic stacking network. Background Technology
[0002] Lithium-ion batteries, with their significant advantages of high energy density, long cycle life, and being green and pollution-free, have been widely used in many fields such as electric bicycles, new energy vehicles, military equipment, and aerospace equipment. However, during repeated charge-discharge cycles, the core components inside the battery, such as electrodes and separators, will experience varying degrees of performance degradation, leading to battery capacity decay and reduced power performance. In severe cases, this can even threaten the safety of users. Therefore, accurately predicting the battery's health status is crucial to ensuring the safe and stable operation of lithium-ion batteries. Among related technologies, lithium-ion battery SoH estimation methods can be mainly divided into two categories: model-driven and data-driven. The disadvantage of model-driven methods is that they require acquiring a large number of electrochemical parameters, and the modeling process is cumbersome and complex, making it difficult to achieve engineering-scale application. The disadvantage of data-driven methods is that they require accumulating rich domain experience and undergoing multiple trial-and-error optimizations, and the prediction accuracy largely depends on the reasonable selection of input features. Summary of the Invention
[0003] This application provides a lithium battery health state prediction method based on a hybrid cyclic stacked network. This lithium battery health state prediction method uses SoH data during the lithium battery discharge process and a hybrid cyclic stacked network to enhance the stability of the model prediction and improve the prediction accuracy, thereby effectively overcoming the problems of feature extraction and poor prediction accuracy in lithium-ion battery SoH estimation.
[0004] The lithium battery health state prediction method based on hybrid cyclic stacked networks provided in this application includes the following steps: S1. Determine the rated capacity, discharge cut-off voltage, and full charge voltage of the lithium battery; S2. Conduct a discontinuous charge and discharge experiment on the lithium battery. The charging cutoff condition is that the battery reaches the full charge voltage, the discharge cutoff condition is that the voltage drops to the discharge cutoff voltage, and the experiment termination condition is that the SoH drops to the end-of-life threshold. Record the discharge current and discharge time during the experiment, and calculate the discharge capacity in each cycle based on the integral of the discharge current and discharge time. Calculate the ratio of the discharge capacity of the lithium battery in each cycle to the rated capacity to obtain the battery health state SoH of the lithium battery. S3. Perform data cleaning on the SoH data during the experiment, including identifying and processing errors, missing values and outliers in the SoH data, and establishing a cyclic sequence of the health status of the lithium battery. S4. Construct a SoH descent dataset D based on the cyclic sequence, where D = {SoH}1, SoH 2, ...,SoH i}; S5. Split the SoH descent dataset D into a training set and a test set; S6. Obtain the input variable and output variable, wherein the input variable is {SoH}. 1, SoH 2, ...,SoH n}, n < i, the output variable is {SoH n+1, SoH n+2, ...,SoH i}; where SoH i This indicates the battery health status in the i-th cycle; S7. Train the hybrid recurrent stacked network. The training process includes: taking the first n data points in the SoH descent dataset D as input to calculate the true value of the model prediction error and the model input value; inputting the model input value into the hybrid recurrent stacked network to output n predicted values; iterating the model multiple times using the training set to obtain the hybrid recurrent stacked network. S8. Validate the hybrid cyclic stacked network using a test set.
[0005] In one optional embodiment, during step S2, the lithium battery is subjected to three conditions—low temperature, room temperature, and high temperature—during the discontinuous charge-discharge experiment; the SoH formula for the i-th cycle of the lithium battery is expressed as: ; In the formula, t1 is the initial discharge time, t2 is the discharge end time, I is the discharge current, and Q is the discharge current. rated This refers to the rated capacity of the lithium battery.
[0006] In one alternative approach, step S3, during data cleaning, includes the following steps: For outliers in the original SoH of each charge-discharge cycle, a Savitzky-Golay filter is used to smooth the data. SoH data from multiple adjacent cycles are selected each time, and local trends are simulated using a polynomial. The original SoH value is then replaced with the fitted value of the polynomial at the center of the window, thereby establishing a cyclic sequence of lithium battery health status.
[0007] In one alternative scheme, in step S6, the value of i is 44, the input variable of the hybrid cyclic stacking network is the first 30 SoH data during the charge-discharge cycle, i.e., n=30, and the output variable is the last 14 SoH data during the charge-discharge cycle.
[0008] In an alternative embodiment, in step S7, the hybrid cyclic stacked network consists of multiple hybrid hidden layers and multiple hybrid residuals. Each hybrid hidden layer includes an RNN layer and a GRU layer. Information from the RNN layer in the first hybrid hidden layer is connected to the RNN layer of the next hybrid hidden layer through the hybrid residuals, and information from the GRU layer in the first hybrid hidden layer is connected to the GRU layer of the next hybrid hidden layer through the hybrid residuals, forming a hybrid residual connection. Finally, the fully connected layer outputs the result.
[0009] In one alternative embodiment, the hybrid cyclic stacked network is a dual cyclic stacked fully connected network structure, wherein two RNN layers and two GRU layers are alternately arranged, along with one fully connected layer. The first layer is an RNN layer with 30 neurons, the second layer is a GRU layer with 30 neurons, the third layer is an RNN layer with 30 neurons, the fourth layer is a GRU layer with 30 neurons, and the fifth layer is a fully connected layer with 12 neurons.
[0010] In one alternative approach, the RNN layer with the hybrid hidden layer receives the input and the hidden state of the previous time step at each time step, and generates a new hidden state for the current time step. The GRU layer uses two gating mechanisms, a reset gate and an update gate, to generate the new hidden state. In the first RNN layer with the hybrid hidden layer, the formula for updating the hidden state is: ; In the GRU layer of the first hybrid hidden layer, there are reset gates, update gates, candidate hidden states, and the final hidden state. The update formula is as follows: ; ; ; ; In the formula, For the current time step input, Hide the current state. It is the hidden state of the previous time step. , , , These are the reset door, update door, candidate hidden state, and final hidden state. , It is the weight matrix of the RNN layer. It is the bias of the RNN layer. It is the weight matrix of the reset gate. It updates the weight matrix of the gate. , , These are the offsets for resetting the door, updating the door, and the candidate hidden state.
[0011] In one alternative approach, in the RNN layer with the second hybrid hidden layer, the network input is... + The remaining calculations are the same as in the first hybrid hidden layer of the RNN layer. In the second hybrid hidden layer of the GRU layer, the network input is... + The remaining calculations are consistent with those of the GRU layer in the second hybrid hidden layer.
[0012] In one alternative approach, in step S8, when verifying the hybrid cyclic stacked network using a test set, if the mean absolute percentage error (MAPE) on the test set is less than or equal to 1%, the model is proven to be effective; if the mean absolute percentage error (MAPE) on the test set is greater than 1%, the above steps are repeated until the mean absolute percentage error (MAPE) on the test set is less than or equal to 1%, and the experiment ends.
[0013] In an optional embodiment, during step S8, when validating the hybrid cyclic stacked network, a prediction error evaluation analysis is performed based on the predicted SoH data and the actual SoH data, including calculating the mean absolute percentage error (MAPE). ; In the formula, Indicates the predicted value, 𝑦 𝑖 This represents the actual value, and n represents the number of predicted sequences.
[0014] The beneficial effects of this application are as follows: The lithium battery health state prediction method in this application directly obtains the SoH data of the lithium battery discharge process through experiments and calculations. It only uses the sequential relationship between the SoH data as the input of the neural network, and uses a hybrid recurrent stacked network to establish a long-term prediction method for the lithium battery health state. It also uses certain prior knowledge without adding other parameters to help the network better understand the relationship between the SoH data, thereby improving the stability and prediction accuracy of the long-term prediction of lithium battery SoH. It effectively overcomes the problems of feature extraction and poor prediction accuracy in lithium-ion battery SoH estimation.
[0015] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description
[0016] Figure 1 A schematic diagram of the composition of the hybrid cyclic stacked network provided in this application in a specific embodiment; Figure 2A flowchart illustrating the lithium battery health status prediction method provided in this application; Figure 3 A schematic diagram illustrating the prediction results of the lithium battery health status prediction method provided in this application.
[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. Detailed Implementation
[0018] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0019] It should be understood that the described embodiments are merely some embodiments of this application, and not all embodiments. All other technical solutions obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0020] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.
[0021] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0022] It should be noted that the directional terms such as "upper," "lower," "left," and "right" described in the embodiments of this application are used to describe the angles shown in the accompanying drawings and should not be construed as limiting the embodiments of this application. Furthermore, in the context, it should be understood that when it is mentioned that an element is connected "upper" or "lower" to another element, it can be directly connected to the other element "upper" or "lower," or indirectly connected to the other element "upper" or "lower" through an intermediate element.
[0023] like Figure 1-3 As shown in the figure, this application provides a method for predicting the state of health of a lithium battery based on a hybrid cyclic stacked network. The method mainly includes the following steps: S1. Determine the rated capacity, discharge cutoff voltage U1, and full charge voltage U2 of the lithium battery; for example, determine the rated capacity specified by the manufacturer when the lithium battery leaves the factory, determine the voltage at 5% SoC as the discharge cutoff voltage U1 for each experiment, and the voltage at 90% SoC as the full charge voltage U2 for each experiment.
[0024] S2. Conduct discontinuous charge-discharge experiments on the lithium battery under three conditions: low temperature, room temperature, and high temperature. Charge is performed at a constant current. The charging cutoff condition is when the battery reaches the full charge voltage U2, and the discharging cutoff condition is when the voltage drops to the discharge cutoff voltage U1. The experiment terminates when the SoH (State of Health) drops to the end-of-life threshold, for example, when SoH drops to 75% of the end-of-life threshold. Record the discharge current and discharge time during the experiment. Calculate the discharge capacity in each cycle based on the integral of the discharge current and discharge time. Then, the battery health state SoH can be obtained by calculating the ratio of the discharge capacity of the lithium battery in each cycle to its rated capacity.
[0025] In this step, the SoH formula for the i-th cycle of the lithium battery is expressed as: ; In the formula, t1 is the initial discharge time, t2 is the discharge end time, I is the discharge current, and Q is the discharge current. rated This refers to the rated capacity of the lithium battery.
[0026] S3. Data cleaning is performed on the SoH data during the experiment. Data cleaning includes identifying and processing errors, missing values and outliers in the SoH data. The Savitzky-Golay filter is used for data smoothing. 51 adjacent cycles of SoH data are selected each time. Local trends are simulated by polynomials. The fitted value of the polynomial at the center of the window is then used to replace the original SoH value, thereby establishing a cyclic sequence of lithium battery health status.
[0027] S4. Based on the cyclic sequence of lithium battery health states, construct the SoH descent dataset D using a sliding window approach, where D = {SoH} 1, SoH 2, ...,SoH i}, where SoH i This represents the battery health status in the i-th cycle. In one specific embodiment, the value of i can be 44, i.e., D = {SoH} 1, SoH 2, ...,SoH 44}
[0028] S5. Split the SoH descent dataset D into a training set and a test set.
[0029] S6. Obtain the input and output variables. The input variable is {SoH}.1, SoH 2, ...,SoH n}, n < i, the output variable is {SoH n+1, SoH n+2, ...,SoH i In one specific embodiment, the input variable is the first 30 SoH data points during the charge-discharge cycle, i.e., {SoH} 1, SoH 2, ...,SoH 30 The output variable is the last 14 SoH data points during the charge / discharge cycle, i.e., {SoH} 31, SoH 32, ...,SoH 44}
[0030] S7. Training the hybrid recurrent stacked network. The training process includes: using the first n data points from the SoH descent dataset D as input to calculate the true value of the model prediction error and the model input value; inputting the model input value into the hybrid recurrent stacked network to output n predicted values; and iterating the model multiple times using the training set to obtain the hybrid recurrent stacked network. Specifically, since the value of i is 44 in this embodiment, the input variable of the hybrid recurrent stacked network is the first 30 SoH data points during the charge-discharge cycle, i.e., n=30, and the output variable is the last 14 SoH data points during the charge-discharge cycle.
[0031] In step S7, the hybrid recurrent stacked network consists of multiple hybrid hidden layers and multiple hybrid residuals. Each hybrid hidden layer includes an RNN layer and a GRU layer. The information from the RNN layer in the first hybrid hidden layer is connected to the RNN layer of the next hybrid hidden layer through the hybrid residuals, and the information from the GRU layer in the first hybrid hidden layer is connected to the GRU layer of the next hybrid hidden layer through the hybrid residuals, forming hybrid residual connections. Finally, the fully connected layer produces the output. Furthermore, the hybrid recurrent stacked network is a double-recurrent stacked fully connected network structure, in which two RNN layers and two GRU layers are alternately arranged, along with one fully connected layer. The first layer is an RNN layer with 30 neurons, the second layer is a GRU layer with 30 neurons, the third layer is an RNN layer with 30 neurons, the fourth layer is a GRU layer with 30 neurons, and the fifth layer is a fully connected layer with 12 neurons.
[0032] The hybrid hidden layer RNN layer receives input and the hidden state from the previous time step at each time step, and generates a new hidden state for the current time step. The GRU layer uses two gating mechanisms, a reset gate and an update gate, to generate the new hidden state. In the first hybrid hidden layer RNN layer, the formula for updating the hidden state is: ; In the GRU layer of the first hybrid hidden layer, there are reset gates, update gates, candidate hidden states, and the final hidden state. The update formula is as follows: ; ; ; ; In the formula, For the current time step input, Hide the current state. It is the hidden state of the previous time step. , , , These are the reset door, update door, candidate hidden state, and final hidden state. , It is the weight matrix of the RNN layer. It is the bias of the RNN layer. It is the weight matrix of the reset gate. It updates the weight matrix of the gate. , , These are the offsets for resetting the door, updating the door, and the candidate hidden state.
[0033] In the second hybrid hidden layer of the RNN, the network input is... + The remaining calculations are the same as in the first hybrid hidden layer of the RNN layer. In the second hybrid hidden layer of the GRU layer, the network input is... + The remaining calculations are consistent with those of the GRU layer in the second hybrid hidden layer.
[0034] S8. Use the test set to validate the hybrid cyclic stacked network. If the MAPE (mean absolute percentage error) on the test set is less than or equal to 1%, the model is proven to be effective. If the MAPE on the test set is greater than 1%, repeat the above experimental steps until the MAPE on the test set is less than or equal to 1%, and then end the experiment.
[0035] Furthermore, in this step, when validating the hybrid cyclic stacked network, a prediction error evaluation analysis is performed based on the predicted SoH data and the actual SoH data, which also includes calculating the mean absolute percentage error (MAPE). ; In the formula, Indicates the predicted value, 𝑦 𝑖 This represents the actual value, and n represents the number of predicted sequences.
[0036] like Figure 2 As shown, in order to more intuitively illustrate the lithium battery health status prediction method in the embodiments of this application, the above scheme will be described in detail below with reference to specific implementation cases.
[0037] Step S1: Select 5 18650 lithium batteries for the experiment, and obtain the rated capacity of 2Ah and OCV-SoC curves of the 5 lithium batteries.
[0038] Step S2: Label the five experimental batteries as No. 1 to No. 5. Refer to the OCV-SoC curves provided by the battery manufacturers to determine the cutoff voltage for the charging and discharging process. Use a battery capacity drop to 1.4 Ah as the critical condition for stopping the experiment. The grouping scheme is designed as follows: batteries 1–4 are included in the model training dataset, and battery No. 5 is used as the test dataset. The experimental temperature of the test set is different from all groups in the training set to verify the model's cross-temperature adaptability. The temperature conditions of the training set are distributed as follows: battery No. 1 is tested at a low temperature of 10℃, battery No. 2 at a room temperature of 25℃, battery No. 3 at a high temperature of 30℃, battery No. 4 at a 45℃ environment, and battery No. 5 at a 20℃ environment.
[0039] Step S3: Record the current data and time information of training batteries 1–4 and test battery 5 during the experimental discharge process; calculate the SoH data, perform data preprocessing, and obtain the lithium battery SoH cyclic decrease sequence; Step S4: Construct the SoH descent dataset D, D={SoH 1, SoH 2, ...,SoH 44}; Step S5: Split the SoH descent dataset D into a training set and a test set; Step S6: In this case, the input variable is the battery health state SoH after 30 discharge cycles, and the output variable is the battery health state SoH after 14 discharge cycles. Step S7: Train a hybrid recurrent stacked network using the training set; Step S8: Validate the hybrid recurrent stacked network using a test set; if the MAPE (mean absolute percentage error) on the test set is less than or equal to 1%, the model is proven to be effective; if the MAPE on the test set is greater than 1%, repeat the above experimental steps until the MAPE on the test set is less than or equal to 1%, and end the experiment.
[0040] This case study uses the aforementioned AA battery to validate the SoH model and outputs the prediction results; as follows: Figure 3 As shown in the figure, it is clear that the predicted values in the 14 steps are close to the actual values, and the MAPE on the test set is 0.18%, indicating that the overall prediction effect of the model is good.
[0041] As can be seen from the above embodiments, the lithium battery health state prediction method in this application directly obtains the SoH data of the lithium battery discharge process through experiments and calculations. It only uses the sequential relationship between the SoH data as the input of the neural network, and establishes a long-term prediction method for the lithium battery health state by using a hybrid recurrent stacked network RNN network and GRU network. It also uses certain prior knowledge without adding other parameters to help the network better understand the relationship between the SoH data, thereby improving the stability and prediction accuracy of the long-term prediction of lithium battery SoH. It effectively overcomes the problems of feature extraction and poor prediction accuracy in lithium-ion battery SoH estimation.
[0042] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for predicting the state of health of a lithium battery based on a hybrid cyclic stacked network, characterized in that, Includes the following steps: S1. Determine the rated capacity, discharge cut-off voltage, and full charge voltage of the lithium battery; S2. Conduct a discontinuous charge and discharge experiment on the lithium battery. The charging cutoff condition is when the battery reaches the full charge voltage, the discharge cutoff condition is when the voltage drops to the discharge cutoff voltage, and the experiment termination condition is when SoH drops to the end of life threshold. Record the discharge current and discharge time during the experiment, and calculate the discharge capacity in each cycle based on the integral of the discharge current and discharge time. Calculate the ratio of the discharge capacity of the lithium battery in each cycle to the rated capacity to obtain the state of health (SoH) of the lithium battery. S3. Perform data cleaning on the SoH data during the experiment, including identifying and processing errors, missing values and outliers in the SoH data, and establishing a cyclic sequence of the health status of the lithium battery. S4. Construct a SoH descent dataset D based on the cyclic sequence, where D = {SoH} 1, SoH 2, ...,SoH i }; S5. Split the SoH descent dataset D into a training set and a test set; S6. Obtain the input and output variables, where the input variable is {SoH}. 1, SoH 2, ...,SoH n }, n < i, the output variable is {SoH n+1, SoH n+2, ...,SoH i }; where SoH i This indicates the battery health status in the i-th cycle; S7. Train the hybrid recurrent stacked network. The training process includes: taking the first n data points in the SoH descent dataset D as input to calculate the true value of the model prediction error and the model input value; inputting the model input value into the hybrid recurrent stacked network to output n predicted values; iterating the model multiple times using the training set to obtain the hybrid recurrent stacked network. S8. Validate the hybrid cyclic stacked network using a test set.
2. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 1, characterized in that, In step S2, the lithium battery is subjected to three conditions—low temperature, room temperature, and high temperature—during the discontinuous charge-discharge experiment. The SoH formula for the i-th cycle of the lithium battery is expressed as: ； In the formula, t1 is the initial discharge time, t2 is the discharge end time, I is the discharge current, and Q is the discharge current. rated This refers to the rated capacity of the lithium battery.
3. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 2, characterized in that, Step S3, when cleaning the data, includes the following steps: For outliers in the original SoH of each charge-discharge cycle, a Savitzky-Golay filter is used to smooth the data. SoH data from multiple adjacent cycles are selected each time, and local trends are simulated using a polynomial. The original SoH value is then replaced with the fitted value of the polynomial at the center of the window, thereby establishing a cyclic sequence of lithium battery health status.
4. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 2, characterized in that, In step S6, the value of i is 44. The input variable of the hybrid cyclic stacking network is the first 30 SoH data during the charge-discharge cycle, i.e., n=30. The output variable is the last 14 SoH data during the charge-discharge cycle.
5. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to any one of claims 1-4, characterized in that, In step S7, the hybrid cyclic stacked network consists of multiple hybrid hidden layers and multiple hybrid residuals. Each hybrid hidden layer includes an RNN layer and a GRU layer. The information of the RNN layer in the first hybrid hidden layer is connected to the RNN layer of the next hybrid hidden layer through the hybrid residual. The information of the GRU layer in the first hybrid hidden layer is connected to the GRU layer of the next hybrid hidden layer through the hybrid residual, forming a hybrid residual connection. Finally, the fully connected layer obtains the output.
6. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 5, characterized in that, The hybrid cyclic stacked network is a dual-cyclic stacked fully connected network structure, in which two RNN layers and two GRU layers are arranged alternately, as well as one fully connected layer. The first layer is an RNN layer with 30 neurons, the second layer is a GRU layer with 30 neurons, the third layer is an RNN layer with 30 neurons, the fourth layer is a GRU layer with 30 neurons, and the fifth layer is a fully connected layer with 12 neurons.
7. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 5, characterized in that, The RNN layer with hybrid hidden layers receives input and the hidden state from the previous time step at each time step, and generates a new hidden state for the current time step. The GRU layer uses two gating mechanisms, a reset gate and an update gate, to generate the new hidden state. In the first RNN layer with hybrid hidden layers, the formula for updating the hidden state is: ； In the GRU layer of the first hybrid hidden layer, there are reset gates, update gates, candidate hidden states, and the final hidden state. The update formula is as follows: ； ； ； ； In the formula, Input for the current time step. Hide the current state. It is the hidden state of the previous time step. , , , These are the reset door, update door, candidate hidden state, and final hidden state. , It is the weight matrix of the RNN layer. It is the bias of the RNN layer. It is the weight matrix of the reset gate. It updates the weight matrix of the gate. , , These are the offsets for resetting the door, updating the door, and the candidate hidden state.
8. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 7, characterized in that, In the second hybrid hidden layer of the RNN, the network input is... + The remaining calculations are the same as in the first hybrid hidden layer of the RNN layer. In the second hybrid hidden layer of the GRU layer, the network input is... + The remaining calculations are consistent with those of the GRU layer in the second hybrid hidden layer.
9. The method for predicting the state of health of a lithium battery based on a hybrid cyclic stacked network according to any one of claims 1-4 or 6-8, characterized in that, In step S8, when verifying the hybrid cyclic stacked network using a test set, if the mean absolute percentage error (MAPE) on the test set is less than or equal to 1%, the model is proven to be effective; if the mean absolute percentage error (MAPE) on the test set is greater than 1%, the above steps are repeated until the mean absolute percentage error (MAPE) on the test set is less than or equal to 1%, and the experiment ends.
10. The lithium battery health state prediction method based on a hybrid cyclic stacked network according to claim 9, characterized in that, In step S8, when validating the hybrid cyclic stacked network, a prediction error evaluation analysis is performed based on the predicted SoH data and the actual SoH data, including calculating the mean absolute percentage error (MAPE). ； In the formula, Indicates the predicted value, 𝑦 𝑖 This represents the actual value, and n represents the number of predicted sequences.