Lithium battery health state and service life prediction method based on HA-BiLSTM

Abstract

The invention discloses a lithium battery health state and service life prediction method based on HA-BiLSTM, and belongs to the technical field of new energy, and the method comprises the following steps: S1, data preprocessing and loading; s2, calculating the SOH (state of health) of the battery and extracting features; s3, constructing a mixed attention bidirectional LSTM model; s4, generating time sequence data; s5, multiple times of training and optimal model selection are carried out; step S6, predicting a future SOH sequence; according to the method, the physical characteristics of the battery and the advantages of a deep learning algorithm can be effectively fused, an intelligent prediction model with the dynamic feature screening capability is constructed, high-precision prediction of SOH and RUL is achieved, and meanwhile technical support is provided for intelligent operation and maintenance of a new energy automobile battery management system BMS and an energy storage system.

Description

Technical Field

[0001] This invention belongs to the field of new energy technology, specifically relating to a method for predicting the health status and lifespan of lithium batteries based on HA-BiLSTM. Background Technology

[0002] The widespread adoption of new energy electric vehicles and renewable energy sources has made battery life prediction extremely important. With the rapid growth in the number of electric vehicles, the demand for battery life prediction is also increasing. By 2024, the global number of new energy vehicles exceeded 60 million, with an annual growth rate exceeding 30% for five consecutive years. As a core energy storage component, the accurate prediction of the State of Health (SOH) and Remaining Useful Life (RUL) of lithium-ion batteries directly affects vehicle range, safety, and lifecycle cost management. At the same time, the expansion of renewable energy grid connection, such as the average annual growth of photovoltaic and wind power installed capacity exceeding 25%, has also made the reliability assessment of energy storage battery systems a key link in the stable operation of the energy internet.

[0003] Currently, lithium-ion batteries inevitably face capacity decay and performance degradation during charge-discharge cycles, and the lag in battery life prediction technology has become a bottleneck restricting the development of the new energy industry. On the one hand, electric vehicle users' anxiety about battery degradation, such as reduced driving range and decreased charging efficiency, leads to insufficient consumer confidence; on the other hand, if energy storage power stations lack accurate battery status assessment, it may cause safety accidents such as thermal runaway, or lead to resource waste due to over-maintenance.

[0004] Traditional battery life prediction methods, such as electrochemical models and empirical formulas, suffer from drawbacks such as complex parameter identification and poor adaptability, making them ill-suited to dynamic factors like temperature fluctuations and charge / discharge rate variations in real-world operating conditions. While machine learning-based methods, such as support vector machines and random forests, can handle nonlinear relationships, they lack effective capture of the temporal characteristics of battery degradation. With the development of deep learning technology, Long Short-Term Memory (LSTM) networks, due to their ability to memorize time-series data, are widely used for battery health prediction. However, they still have limitations in handling multi-scale degradation features and extracting key signals, especially in terms of prediction accuracy under complex operating conditions, which urgently needs improvement. Against this backdrop, the key to overcoming existing technological bottlenecks lies in integrating battery physical characteristics with the advantages of deep learning algorithms to construct an intelligent prediction model with dynamic feature selection capabilities.

[0005] Therefore, there is a need for a lithium battery health status and lifespan prediction method based on HA-BiLSTM that can accurately predict SOH and RUL and provide technical support for the intelligent operation and maintenance of battery management systems (BMS) and energy storage systems in new energy vehicles. Here, "HA-BiLSTM" stands for Hybrid Attention Bidirectional Long Short-Term Memory Network. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting the state of health and lifespan of lithium batteries based on HA-BiLSTM. This method can effectively integrate the advantages of battery physical characteristics and deep learning algorithms to build an intelligent prediction model with dynamic feature selection capabilities. It can achieve high-precision prediction of SOH and RUL, while providing technical support for the intelligent operation and maintenance of battery management systems (BMS) and energy storage systems in new energy vehicles.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for predicting the state of health and lifespan of lithium batteries based on HA-BiLSTM includes the following steps:

[0009] Step S1: Data preprocessing and loading;

[0010] Step S2: Calculation and feature extraction of battery state of health (SOH);

[0011] Step S3: Construct a hybrid attention bidirectional LSTM model;

[0012] Step S4: Generate time series data;

[0013] Step S5: Multiple training sessions and selection of the optimal model;

[0014] Step S6: Predicting future SOH sequences;

[0015] Step S7: Real-time battery status assessment and lifespan prediction.

[0016] A further improvement to the technical solution of the present invention is that step S1 includes the following steps:

[0017] Step S101: Data Reading: Use the pandas library to read the battery cycle test data file in Excel format;

[0018] Step S102: Column name adaptation: Automatically identify different column name formats and map them uniformly to the standard English format;

[0019] Step S103: Quality check: Statistically analyze the distribution of missing values ​​and check data integrity;

[0020] Step S104: Data filtering: Prioritize extracting constant current discharge data; if data is insufficient, switch to full discharge data or charging data.

[0021] A further improvement to the technical solution of the present invention is that step S2 includes the following steps:

[0022] Step S201: Calculate the battery state of health (SOH). The calculation formula is as follows:

[0023]

[0024] Where SOH represents the battery health status, C current Indicates the capacity of the current loop, C initial Indicates the initial capacity baseline value;

[0025] Step S202: Signal smoothing: Use the Savitzky-Golay filter as shown below:

[0026]

[0027] Among them, y smooth [i] represents the smoothed value at point i, c j This represents the Savitzky-Golay filter coefficients, where m represents the window radius. The window length is odd, and the polynomial order is 3.

[0028] Step S203: Construct derived features:

[0029] Difference characteristic: SOH_Diff[i] = SOH[i] - SOH[i-1]

[0030] Where SOH_Diff[0] = 0, the first point is filled with 0.

[0031] Rolling average characteristics:

[0032] Rolling standard deviation characteristics:

[0033] Missing data is filled with 0.

[0034] A further improvement to the technical solution of the present invention is that step S3 includes the following steps:

[0035] Step S301: Generate sequence data:

[0036] LSTM sequence construction:

[0037] Forget gate function: f t =σ(Wxf x t +W hf h t-1 +b f )

[0038] Input gate function: i t =σ(W xi x t +W hf h t-1 +b i )

[0039] Cell state function: r t =tanh(W xr x t +W hr h t-1 +b r )

[0040] Cell state update function: c t =f t *c t-1 +i t *r t

[0041] Output gate function: o t =σ(W xo x t +W ho h t-1 +b o )

[0042] Cell state output at time step: m t =o t ⊙tanh(c t )

[0043] Among them, W xf 、W hf 、W xi 、W hi 、W xr 、W hr 、W xo 、W ho b represents the weight. f b i b r b o σ is the bias term, σ is the Sigmoid activation function, ⊙ is the element-wise multiplication, and tanh is the hyperbolic tangent activation function.

[0044] Step S302: Construct BiLSTM sequences:

[0045] Forward LSTM layer output:

[0046] Backward output to the LSTM layer:

[0047] BiLSTM output:

[0048] Step S303: Establish an attention mechanism:

[0049] The output vector Y of the BiLSTM layer, i.e., [y1, y2, y3, ..., y...], is... n As input to the attention layer, the values ​​of Q, K, and V are calculated, i.e.:

[0050] Q = Y * W q

[0051] K = Y * W y

[0052] V = Y * W v

[0053] Among them, D k The dimension of the key vector K; softmax() is the activation function; W q ,W k ,W v This is the weight matrix.

[0054] A further improvement to the technical solution of the present invention is that step S4 includes the following steps:

[0055] Step S401: Construct the time series: Determine the time window size w, for the battery's entire life cycle data D = {d1, d2, ..., d...} n Construct the input time series sequence X = {x1, x2, ..., x}. n-w}, its x j =[d j ,d j+1 ,...,d j+w-1 ] T The corresponding output is y j =d j+w Label values ​​related to the State of Health (SOH) of the battery;

[0056] Step S402: Data normalization:

[0057] Min-Max standardization:

[0058] Where X represents the original feature value, X min X represents the minimum value of the characteristic. max X represents the maximum value of the characteristic. normalized ∈(0,1) represents the normalized value;

[0059] Standardization is performed on each feature dimension of the time series X to make model training more stable;

[0060] Step S403: Data partitioning: Divide the constructed and standardized time series data proportionally.

[0061] A further improvement to the technical solution of the present invention is that step S5 includes the following steps:

[0062] Step S501: Multiple training iterations: Using the training set as input, set the number of training epochs = 100 and the batch size = 32, and use the optimizer Adam and minimize the mean squared error of the loss function. The model performance is evaluated on the test set after each round of training;

[0063] The training set is {X} train ,Y train The test set is {X}. test ,Y test}, prediction set {X pred ,Y pred The real set is {X}. true ,Y true};

[0064] Step S502: Optimal Model Selection: After training, based on the validation set performance metrics, select the optimal model M from the model checkpoints obtained from multiple training iterations. best Record its value in the test set {X} test ,Y test Evaluation error on}

[0065] A further improvement to the technical solution of the present invention is that step S6 includes the following steps:

[0066] Step S601: Constructing Future Time Series: Based on the time series patterns of historical data, extend and construct the future time series input sequence X. future If the feature data {d1, d2, ..., d} of the last w iterations are known... n}, then X future,1 ={d n-w+1 ,...,d n} T In subsequent predictions, the results of the previous prediction are successively added to the sequence as new features to construct X. future,t Where t is the number of prediction steps;

[0067] Step S602: SOH Sequence Prediction: Input the constructed future time series sequence X future Input optimal model Mbest By using forward propagation of the model, a sequence of predicted SOH values ​​for multiple future time steps is obtained. This is manifested in the model's nonlinear mapping of the input sequence.

[0068] A further improvement to the technical solution of the present invention is that step S7 includes the following steps:

[0069] Step S701: Collect real-time data: Obtain the battery's current cycle operation data in real time through the Battery Management System (BMS), including charge / discharge capacity C. current Information such as voltage (V), current (I), and duration (time) serves as the input for real-time condition assessment and lifetime prediction.

[0070] Step S702: Real-time SOH calculation and state assessment: using formula Among them, C rated Based on the battery's rated capacity and real-time collected capacity data, the current SOH value is calculated, and the battery's real-time health status level is assessed according to the preset health status grading standard.

[0071] Step S703: Based on the future SOH prediction sequence obtained in step S6 Based on the SOH threshold corresponding to the battery's rated lifespan, the number of cycles N required from the current moment to reach the threshold is calculated using linear fitting or time-series prediction algorithms. remaining This serves as the prediction result for the remaining battery life.

[0072] The technological advancements achieved by this invention due to the adoption of the above technical solutions are as follows:

[0073] This invention presents a lithium battery health status and lifespan prediction method based on HA-BiLSTM, which effectively integrates the advantages of battery physical characteristics and deep learning algorithms to construct an intelligent prediction model with dynamic feature selection capabilities. The bidirectional long short-term memory model based on a hybrid attention mechanism proposed in this invention enhances the ability to capture key features of battery degradation, achieving high-precision prediction of SOH and RUL, and providing technical support for the intelligent operation and maintenance of battery management systems (BMS) and energy storage systems in new energy vehicles.

[0074] This invention effectively addresses the problem that traditional physical modeling methods require large datasets for statistical models, while the novel hybrid model is suitable for smaller datasets and learns by recognizing patterns. In terms of accuracy and precision, it reduces the error rate to below 1% compared to the Kalman filter model, exhibiting high adaptability. It also performs well in capturing nonlinear degradation and has a strong ability to capture dynamic feature values. Furthermore, it can provide prediction results with small errors in real time.

[0075] This invention constructs a hybrid attention bidirectional LSTM model, which improves the prediction accuracy of battery state of health (SOH) and remaining useful life (RUL) through bidirectional LSTM structure, hybrid attention mechanism, and multi-scale feature extraction. This method realizes a solution for predicting battery state of health and life, which can improve prediction accuracy, reduce system cost, improve operational reliability and charging efficiency, and has significant environmental benefits. It has broad application prospects in the fields of new energy vehicle charging stations and battery cascade utilization. Attached Figure Description

[0076] Figure 1 This is a schematic diagram of the principle of the present invention;

[0077] Figure 2 This is a simulation flowchart of the present invention;

[0078] Figure 3 This is a schematic diagram of the BiLSTM algorithm process in this invention. Detailed Implementation

[0079] The present invention will be further described in detail below with reference to embodiments:

[0080] like Figure 1 and Figure 2 As shown, this invention provides a method for predicting the health status and lifespan of lithium batteries based on HA-BiLSTM, including the following steps:

[0081] Step S1: Data preprocessing and loading, specifically including the following steps:

[0082] Step S101: Data Reading: Use the pandas library to read the battery cycle test data file in Excel format;

[0083] Step S102: Column name adaptation: Automatically identify different column name formats, such as "charging capacity Ah" and "discharging capacity (Ah)", and uniformly map them to standard English formats, such as Cycle and Discharge Cap.

[0084] Step S103: Quality check: Statistically analyze the distribution of missing values ​​and check data integrity, such as the continuity of cycle numbers and the rationality of capacity data;

[0085] Step S104: Data filtering: Prioritize extracting constant current discharge data; if data is insufficient, switch to full discharge data or charging data.

[0086] Step S2: Battery State of Health (SOH) calculation and feature extraction, specifically including the following steps:

[0087] Step S201: Calculate the battery state of health (SOH). The calculation formula is as follows:

[0088]

[0089] Here, SOH represents the battery's state of health, usually expressed as a percentage, and C... current C represents the capacity of the current cycle, i.e., the charging capacity or the discharging capacity. initial This represents the initial capacity baseline value, which is the average capacity of the first 10 cycles;

[0090] Step S202: Signal smoothing: Use the Savitzky-Golay filter as shown below:

[0091]

[0092] Among them, y smooth [i] represents the smoothed value at point i, c j This represents the Savitzky-Golay filter coefficients, where m represents the window radius. The window length must be an odd number, and the polynomial order must be 3.

[0093] Step S203: Construct derived features:

[0094] Difference characteristic: SOH_Diff[i] = SOH[i] - SOH[i-1]

[0095] Where SOH_Diff[0] = 0, the first point is filled with 0.

[0096] Rolling average characteristics:

[0097] Rolling standard deviation characteristics:

[0098] Missing data is filled with 0;

[0099] Step S3: As Figure 3 As shown, constructing a hybrid attention bidirectional LSTM model includes the following steps:

[0100] Step S301: Generate sequence data:

[0101] LSTM sequence construction:

[0102] Forget gate function: f t =σ(W xf x t +W hf h t-1 +b f )

[0103] Input gate function: i t =σ(W xi x t+W hf h t-1 +b i )

[0104] Cell state function: r t =tanh(W xr x t +W hr h t-1 +b r )

[0105] Cell state update function: c t =f t *c t-1 +i t *r t

[0106] Output gate function: o t =σ(W xo x t +W ho h t-1 +b o )

[0107] Cell state output at time step: m t =o t ⊙tanh(c t )

[0108] Among them, W xf W hf W xi W hi W xr W hr W xo W ho b represents the weight. f b i b r b o σ is the bias term, σ is the Sigmoid activation function, ⊙ is the element-wise multiplication, and tanh is the hyperbolic tangent activation function.

[0109] Step S302: Construct BiLSTM sequences:

[0110] Forward LSTM layer output:

[0111] Backward output to the LSTM layer:

[0112] BiLSTM output:

[0113] Step S303: Establish an attention mechanism:

[0114] The output vector Y of the BiLSTM layer, i.e., [y1, y2, y3, ..., y...], is... n As input to the attention layer, the values ​​of Q, K, and V are calculated, i.e.:

[0115] Q = Y * W q

[0116] K = Y * W y

[0117] V = Y * W v

[0118] Among them, D k The dimension of the key vector K; softmax() is the activation function; W q ,W k ,W v This is the weight matrix;

[0119] Step S4: Generate time series data, which specifically includes the following steps:

[0120] Step S401: Construct the time series: Determine the time series window size w (which can be adjusted according to data characteristics and model requirements, such as selecting the optimal value through cross-validation). For the battery's entire life cycle data, D = {d1, d2, ..., d...} n}(d i Let the feature set of the i-th cycle contain information such as capacity and voltage, i∈{1,2,…,n}, and construct the input time series sequence X={x1,x2,...,x n-w}, its x j =[d j ,d j+1 ,...,d j+w-1 ] T The corresponding output is y j =d j+w Label values ​​related to the State of Health (SOH) of the battery;

[0121] Step S402: Data normalization:

[0122] Min-Max standardization:

[0123] Where X represents the original feature value, X min X represents the minimum value of the characteristic. max X represents the maximum value of the characteristic. normalized ∈(0,1) represents the normalized value;

[0124] Standardization is performed on each feature dimension of the time series X to make model training more stable;

[0125] Step S403: Data partitioning: Divide the constructed and standardized time series data according to a certain ratio, such as training set: test set = 8:2;

[0126] Step S5: Multiple training iterations and optimal model selection, specifically including the following steps:

[0127] Step S501: Multiple training iterations: Using the training set as input, set the number of training epochs = 100 and the batch size = 32, and use the optimizer Adam and minimize the mean squared error of the loss function. The model performance is evaluated on the test set after each round of training;

[0128] The training set is {X} train ,Y train The test set is {X}. test ,Y test}, prediction set {X pred ,Y pred The real set is {X}. true ,Y true};

[0129] Step S502: Optimal Model Selection: After training, based on the validation set performance metrics, i.e., minimum validation set loss and highest SOH prediction accuracy, the optimal model M is selected from the model checkpoints trained multiple times. best Record its value in the test set {X} test ,Y test Evaluation error on}

[0130] Step S6: Predicting future SOH sequences, specifically including the following steps:

[0131] Step S601: Constructing Future Time Series: Based on the time series patterns of historical data, extend and construct the future time series input sequence X. future If the feature data {d1, d2, ..., d} of the last w iterations are known... n}, then X future,1 ={d n-w+1 ,...,d n} T In subsequent predictions, the results of the previous prediction are successively added to the sequence as new features to construct X. future,t Where t is the number of prediction steps;

[0132] Step S602: SOH Sequence Prediction: Input the constructed future time series sequence X future Input optimal model M best By using forward propagation of the model, a sequence of predicted SOH values ​​for multiple future time steps is obtained. This is manifested in the model's nonlinear mapping of the input sequence.

[0133] Step S7: Real-time battery status assessment and lifespan prediction, specifically including the following steps:

[0134] Step S701: Collect real-time data: Obtain the battery's current cycle operation data in real time through the Battery Management System (BMS), including charge / discharge capacity C. current Information such as voltage (V), current (I), and duration (time) serves as the input for real-time condition assessment and lifetime prediction.

[0135] Step S702: Real-time SOH calculation and state assessment: using formula Among them, C rated Based on the battery's rated capacity and real-time collected capacity data, the current SOH value is calculated. According to the preset health status grading standard, SOH ≥ 90% is healthy, 80% ≤ SOH < 90% is good, 60% ≤ SOH < 80% is sub-healthy, and SOH < 60% requires replacement / repair, thus assessing the battery's real-time health status level.

[0136] Step S703: Based on the future SOH prediction sequence obtained in step S6 Based on the SOH threshold corresponding to the battery's rated lifespan, the number of cycles N required from the current moment to reach the threshold is calculated using linear fitting or time-series prediction algorithms. remaining This serves as the prediction result for the remaining battery life.

[0137] The timing data of electrical parameters during the charge-discharge cycle of a lithium battery are shown in the table below:

[0138]

[0139] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for predicting the state of health and lifespan of lithium batteries based on HA-BiLSTM, characterized in that... Includes the following steps: Step S1: Data preprocessing and loading; Step S2: Calculation and feature extraction of battery state of health (SOH); Step S3: Construct a hybrid attention bidirectional LSTM model; Step S4: Generate time series data; Step S5: Multiple training sessions and selection of the optimal model; Step S6: Predicting future SOH sequences; Step S7: Real-time battery status assessment and lifespan prediction.

2. The method for predicting the state of health and lifespan of lithium batteries based on HA-BiLSTM according to claim 1, characterized in that: Step S1 includes the following steps: Step S101: Data Reading: Use the pandas library to read the battery cycle test data file in Excel format; Step S102: Column name adaptation: Automatically identify different column name formats and map them uniformly to the standard English format; Step S103: Quality check: Statistically analyze the distribution of missing values ​​and check data integrity; Step S104: Data filtering: Prioritize extracting constant current discharge data; if data is insufficient, switch to full discharge data or charging data.

3. The method for predicting the health status and lifespan of lithium batteries based on HA-BiLSTM according to claim 2, characterized in that: Step S2 includes the following steps: Step S201: Calculate the battery state of health (SOH). The calculation formula is as follows: Where SOH represents the battery health status, C current Indicates the capacity of the current loop, C initial Indicates the initial capacity baseline value; Step S202: Signal smoothing: Use the Savitzky-Golay filter as shown below: Among them, y smooth [i] represents the smoothed value at point i, c j This represents the Savitzky-Golay filter coefficients, where m represents the window radius. The window length is odd, and the polynomial order is 3. Step S203: Construct derived features: Difference characteristic: SOH_Diff[i] = SOH[i] - SOH[i-1] Where SOH_Diff[0] = 0, the first point is filled with 0. Rolling average characteristics: Rolling standard deviation characteristics: Missing data is filled with 0.

4. The method for predicting the health status and lifespan of lithium batteries based on HA-BiLSTM according to claim 3, characterized in that: Step S3 includes the following steps: Step S301: Generate sequence data: LSTM sequence construction: Forget gate function: f t =σ(W xf x t +W hf h t-1 +b f ) Input gate function: i t =σ(W xi x t +W hf h t-1 +b i ) Cell state function: r t =tanh(W xr x t +W hr h t-1 +b r ) Cell state update function: c t =f t *c t-1 +i t *r t Output gate function: o t =σ(W xo x t +W ho h t-1 +b o ) Cell state output at time step: m t =o t ⊙tanh(c t ) Among them, W xf 、W hf 、W xi 、W hi 、W xr 、W hr 、W xo 、W ho b represents the weight. f 、b i 、b r 、b o σ is the bias term, σ is the Sigmoid activation function, ⊙ is the element-wise multiplication, and tanh is the hyperbolic tangent activation function. Step S302: Construct BiLSTM sequences: Forward LSTM layer output: Backward output to the LSTM layer: BiLSTM output: Step S303: Establish an attention mechanism: The output vector Y of the BiLSTM layer, i.e., [y1, y2, y3, ..., y...], is... n As input to the attention layer, the values ​​of Q, K, and V are calculated, i.e.: Q=Y*W q K=Y*W y V=Y*W v Among them, D k The dimension of the key vector K; softmax() is the activation function; W q W k W v This is the weight matrix.

5. The method for predicting the state of health and lifespan of lithium batteries based on HA-BiLSTM according to claim 4, characterized in that: Step S4 includes the following steps: Step S401: Construct the time series: Determine the time window size w, for the battery's entire life cycle data D = {d1, d2, ..., d...} n Construct the input time series sequence X = {x1, x2, ..., x}. n-w }, its x j =[d j ,d j+1 ,...,d j+w-1 ] T The corresponding output is y j =d j+w Label values ​​related to the State of Health (SOH) of the battery; Step S402: Data normalization: Min-Max standardization: Where X represents the original feature value, X min X represents the minimum value of the characteristic. max X represents the maximum value of the characteristic. normalized ∈(0,1) represents the normalized value; Standardization is performed on each feature dimension of the time series X to make model training more stable; Step S403: Data partitioning: Divide the constructed and standardized time series data proportionally.

6. The method for predicting the state of health and lifespan of a lithium battery based on HA-BiLSTM according to claim 5, characterized in that: Step S5 includes the following steps: Step S501: Multiple training iterations: Using the training set as input, set the number of training epochs = 100 and the batch size = 32, and use the optimizer Adam and minimize the mean squared error of the loss function. The model performance is evaluated on the test set after each round of training; The training set is {X} train ,Y train The test set is {X}. test ,Y test }, prediction set {X pred ,Y pred The real set is {X}. true ,Y true }; Step S502: Optimal Model Selection: After training, based on the validation set performance metrics, select the optimal model M from the model checkpoints obtained from multiple training iterations. best Record its value in the test set {X} test ,Y test Evaluation error on} 7. The method for predicting the state of health and lifespan of a lithium battery based on HA-BiLSTM according to claim 6, characterized in that: Step S6 includes the following steps: Step S601: Constructing Future Time Series: Based on the time series patterns of historical data, extend and construct the future time series input sequence X. future If the feature data {d1, d2, ..., d} of the last w iterations are known... n }, then X future,1 ={d n-w+1 ,...,d n } T In subsequent predictions, the results of the previous prediction are successively added to the sequence as new features to construct X. future,t Where t is the number of prediction steps; Step S602: SOH Sequence Prediction: Input the constructed future time series sequence X future Input optimal model M best By using forward propagation of the model, a sequence of predicted SOH values ​​for multiple future time steps is obtained. This is manifested in the model's nonlinear mapping of the input sequence.

8. The method for predicting the state of health and lifespan of a lithium battery based on HA-BiLSTM according to claim 7, characterized in that: Step S7 includes the following steps: Step S701: Collect real-time data: Obtain the battery's current cycle operation data in real time through the Battery Management System (BMS), including charge / discharge capacity C. current Information such as voltage (V), current (I), and duration (time) serves as the input for real-time condition assessment and lifetime prediction. Step S702: Real-time SOH calculation and state assessment: using formula Among them, C rated Based on the battery's rated capacity and real-time collected capacity data, the current SOH value is calculated, and the battery's real-time health status level is assessed according to the preset health status grading standard. Step S703: Based on the future SOH prediction sequence obtained in step S6 Based on the SOH threshold corresponding to the battery's rated lifespan, the number of cycles N required from the current moment to reach the threshold is calculated using linear fitting or time-series prediction algorithms. remaining This serves as the prediction result for the remaining battery life.

Images