SOC-SOH joint estimation method based on double-cross physical guidance framework

Abstract

The invention provides an SOC-SOH joint estimation method based on a double-crossing physical guidance framework, and the method comprises the steps: transmitting the charging and discharging data features of a battery into a double-crossing physical guidance model, and extracting a high-dimensional feature tensor in the data features through a spatial-temporal feature encoder of the double-crossing physical guidance model; the high-dimensional feature tensor is converted through two learnable projection heads, a double-flow mutual attention module calculates cross-attention information between output after conversion, the cross-attention information is dynamically regulated and fused through a dynamic interaction mechanism, and a final charging state dynamic interaction result and a final health state dynamic interaction result are obtained; the physical constraint decoder maps the final dynamic interaction result of the charging state and the final dynamic interaction result of the health state to a predicted value of the charging state and a predicted value of the health state; the core problem that short-time response capture and long-term degradation description are difficult to consider in an existing method is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery state estimation, and particularly to a joint SOC-SOH estimation method based on a double-cross physical guidance framework. Background Art

[0002] With the wide application of battery technology in new energy vehicles, energy storage systems, and portable devices, higher requirements are put forward for the accuracy and real-time performance of battery state monitoring. The state of charge (SOC) and state of health (SOH) of the battery, as two key indicators for evaluating battery performance, their accurate estimation is of crucial significance for ensuring device safety, extending battery life, and optimizing energy management. However, in practical applications, battery state estimation faces many challenges. First of all, the battery exhibits obvious dynamic responses and transient characteristics during charge and discharge, such as voltage fluctuations, temperature gradients, and current pulses, which require the model to capture multi-scale and multi-dimensional time-series information. At the same time, the battery aging process is often accompanied by non-linear and complex degradation mechanisms, such as the loss of active substances, the thickening of the SEI film, and lithium deposition, etc., resulting in the attenuation curve of SOH showing the characteristics of a combination of sudden changes and gradual changes. In addition, due to the influence of sensor noise, uneven data sampling frequency, and environmental factors under actual working conditions, traditional single data-driven or physical models have limitations in accurately capturing the internal state of the battery, and are prone to problems such as overfitting or insufficient extrapolation ability. To solve the above problems, it is necessary to introduce prior knowledge of the internal physical mechanism of the battery on the basis of ensuring sufficient data expression, so as to improve the prediction accuracy and physical rationality of the model. Most traditional methods independently model SOC and SOH respectively, ignoring the inherent coupling relationship between the two, and unable to take into account both short-term charge and discharge dynamics and long-term aging evolution at the same time, and lacking physical consistency constraints, resulting in the inability to fully reflect the mutual influence of battery charge and discharge behavior and degradation process in practical applications. Summary of the Invention

[0003] In response to the needs in the prior art, the present invention provides a joint SOC-SOH estimation method based on a double-cross physical guidance framework, aiming to solve the problem that existing methods are difficult to capture both the short-term dynamic response and long-term aging trend of the battery under complex working conditions.

[0004] The joint SOC-SOH estimation method based on a double-cross physical guidance framework includes the following steps:

[0005] Step 1: Obtain the data characteristics of battery charge and discharge;

[0006] Step 2: Send the data characteristics into a double-cross physical guidance model, which includes a spatio-temporal feature encoder, a two-stream mutual attention module, and a physical constraint decoder;

[0007] Step 3: The spatiotemporal feature encoder extracts high-dimensional feature tensors from data features based on dilated convolution, multi-head self-attention, and feedforward networks;

[0008] Step 4: The high-dimensional feature tensor is transformed through two parallel learnable projection heads in the feature-guided decoupling mechanism to generate two task-oriented feature streams, which are recorded as the charging state feature stream and the health state feature stream respectively;

[0009] Step 5: The dual-stream mutual attention module calculates the cross-attention information between the charging state feature stream and the health state feature stream, and dynamically regulates and fuses the cross-attention information through the dynamic interaction mechanism to obtain the final charging state dynamic interaction results and health state dynamic interaction results;

[0010] Step 6: The physical constraint decoder maps the final charging state dynamic interaction results and health state dynamic interaction results to the charging state prediction value and the health state prediction value based on the multi-layer perceptron and the physical correction term.

[0011] Further: in the spatiotemporal feature encoder, the data features are processed by the multi-head self-attention layer and the feedforward network layer in sequence after the dilated convolution, and residual connections are set in the multi-head self-attention layer and the feedforward network layer, and layer normalization is set after the multi-head self-attention layer and the feedforward network layer.

[0012] Further: Cross-attention information includes attention from charging status to health status And the attention from health status to charging status ;

[0013] in, is the feature vector mapping in the charging state feature stream, and They are all feature vector mappings in the health status feature stream; is the scaling factor; is the feature vector mapping in the health status feature stream, and They are all feature vector mappings in the charging state feature stream.

[0014] Further: The dynamic interaction mechanism is adaptively regulated by the gating function and the interaction scoring function, and the gating function value and the interaction scoring function value are respectively , Perform weighted fusion.

[0015] Further: Step 6 is specifically:

[0016] Step 6.1: Input the charging state dynamic interaction result and the health state dynamic interaction result into two multi-layer perceptrons respectively to obtain the preliminary and ;

[0017] Step 6.2: In each time segment, calculate the Coulomb integral constraint term and the Arrhenius degradation term respectively by integrating or recursively calculating variables such as the cumulative power and temperature in the current period;

[0018] Step 6.3: Fuse the preliminary and with the Coulomb integral constraint term and the Arrhenius degradation term at the end of the decoder to form the final and .

[0019] Further, it is: ;

[0020] Among them, is the rated capacity, When the Coulomb efficiency is discretely implemented, the integral can be replaced by step-by-step accumulation, and is limited (clamped) at each time step to prevent numerical drift; in the decoder, the Coulomb constraint often appears in the form of a CoulombConstraint layer. First, is estimated according to the current period's current, temperature and other characteristics of the network, and then it is coupled with the output of the MLP; the simplified coupling method is: ;

[0021] Among them, is the Sigmoid function to ensure that the output is in the range of [0,1]; is an adjustable weight or gating factor used to balance the data-driven prediction and the physical integration result.

[0022] Further, it is: ;

[0023] Among them, is the cumulative discharge amount or the number of cycles, represents the attenuation nonlinear coefficient; the core of this equation is to dynamically calculate the attenuation amount according to the current temperature and the cumulative capacity, and output a correction term that is multiplied or added to the predicted by the MLP; the coupling method is: ;

[0024] Among them, is the Sigmoid function.

[0025] Furthermore, it is: from the prediction error and physical consistency and cross-consistency constrain and optimize the double-cross physical-guided Transformer model from three dimensions, and the overall loss is: ;

[0026] where and are hyperparameters used to balance the contribution degrees of different sub-losses; ; ; ;

[0027] Among them, and respectively represent the predicted values of the model for the state of charge and the state of health, and respectively represent the true values of the model for the state of charge and the state of health; is a regulation parameter used to balance the predicted capacity and the deviation between the actual charge transfer calculated for the integral value of the actually observed current ; is a regulation parameter used to control and the degree of consistency of the Arrhenius equation ; is the rated capacity of the new battery; represents a physical quantity inherent in the battery material, A is the pre-exponential factor in the Arrhenius model; T is the current battery temperature, and I is the current.

[0028] Furthermore, it is: Step 1 specifically includes the following steps:

[0029] Step 1.1: Establish an experimental platform for the battery cycle life test, conduct a cycle life test on the lithium-ion battery, and obtain the battery charge and discharge data;

[0030] Step 1.2: Perform multi-scale processing and derivative feature extraction on the original signal from the battery charge and discharge data, and the obtained features include the voltage change rate, temperature gradient, cumulative throughput power, equivalent cycle number, sliding window statistic, and relaxation period voltage recovery rate;

[0031] Step 1.3: Process all the extracted features with a dynamic normalization strategy to obtain the data features of battery charge and discharge.

[0032] Specifically, the dynamic normalization strategy is as follows in mathematical expression: ;

[0033] where is the output of the dynamic normalization strategy, is the original feature value at time step t; is a small constant to prevent division by zero; is the mean value, is the standard deviation, which are respectively expressed as: ; ;

[0034] is the length of the dynamic sliding window and is: ;

[0035] where is the initial length of the window; is the growth rate coefficient; is the health state predicted by the model at the current time step, then represents the degree of health degradation.

[0036] Advantages of the present invention: By means of the spatio-temporal feature encoder and the task decoupling mechanism, it can effectively distinguish short-term dynamic behaviors and long-term aging evolution features, and achieve heterogeneous modeling at the feature level; through the bidirectional cross-task attention mechanism and the gated interaction structure, it explicitly enhances the coupling perception ability between SOC and SOH, and improves the multi-task collaborative estimation effect; it effectively solves the core problem that existing methods are difficult to balance short-time response capture and long-term degradation characterization, and provides a solid support for constructing a highly reliable intelligent battery management system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the flow chart of the present invention;

[0038] Figure 2 is the overall framework diagram of the double-cross physical-guided model in the present invention;

[0039] Figure 3 is the SOC estimation result of the B201 battery for multiple charge and discharge cycles;

[0040] Figure 4 is the SOH estimation result of the B201 battery;

[0041] Figure 5SOH estimation results for B203 battery; Detailed implementation mode

[0042] The present invention will be described in detail below with reference to the accompanying drawings. The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as limiting the present invention. The orientation terms such as left, middle, right, up, and down in the embodiments of the present invention are only relative concepts to each other or are referenced based on the normal use state of the product, and should not be considered restrictive.

[0043] The SOC-SOH joint estimation method based on the double-cross physical guidance framework includes the following steps:

[0044] Step 1: Obtain the data characteristics of battery charge and discharge; specifically, it includes the following steps:

[0045] Step 1.1: Establish an experimental platform for battery cycle life tests, conduct cycle life tests on lithium-ion batteries, and obtain battery charge and discharge data;

[0046] The INR 18650 battery adopts a stepped charging method. First, charge the battery at a constant current of 6C to a voltage of 4.2V; after a short rest, charge the battery at a constant current of 3C to a voltage of 4.2V. After another one-minute rest, charge the battery completely with constant current and constant voltage (CC-CV); the current in its CC stage is 0.5C, the cut-off current is 0.05C, and the cut-off voltage is 4.2V; after standing for one hour, discharge the battery with six times of CLTC-P (China Light Vehicle Test Cycle - Passenger Vehicle) repeatedly until the battery voltage drops to 2.7V; CLTC-P is a standard working condition based on traffic big data, including three intervals: low speed, medium speed, and high speed, with a total duration of 1800s; CLTC-P has high dynamics and can reflect the aging situation of the battery in actual applications. Capacity measurement is carried out every 50 cycles;

[0047] Step 1.2: Perform multi-scale processing and derivative feature extraction on the original signal from the battery charge and discharge data. The obtained features include voltage change rate, temperature gradient, cumulative throughput power, equivalent cycle number, sliding window statistic, and relaxation period voltage recovery rate;

[0048] Two differential features (dynamic response characterization):

[0049] (1) Voltage change rate It reflects the change rate of the polarization voltage and is directly related to the lithium-ion diffusion kinetics. SOC correlation: The sudden change of dV / dt at the end of charging indicates that the SOC is approaching 100%; SOH correlation: The slope of the dV / dt curve of the aged battery decreases;

[0050] (2)Temperature gradient It represents the temperature rise rate per unit power consumption; SOH correlation: The increases (the increase in internal resistance leads to an increase in Joule heat), where represents the temperature of the battery (cell temperature); represents the ambient temperature; represents the instantaneous input power of the battery, which is represented by here;

[0051] Two cumulative features (long-term degradation metrics):

[0052] (3)Cumulative throughput It represents a direct measure of the total charge transfer; SOC correlation: It provides a baseline estimate through Coulomb counting; SOH correlation: There is a non-linear mapping with capacity decay;

[0053] (4)Equivalent cycle number It represents the standardized aged cycle count, which has an exponential relationship with the capacity decay rate;

[0054] Cycle statistical features (local behavior patterns):

[0055] (5)Sliding window statistic It represents the short-term statistical characteristics of voltage fluctuations; SOC correlation: is the average voltage within the sliding window and has a quasi-linear relationship with the SOC (discharge plateau region); SOH correlation: is the standard deviation of the voltage within the sliding window, and an increase reflects the non-uniformity of the phase change of the electrode material; among them, is the size of the sliding window; is the voltage data within the window;

[0056] (6)Relaxation period voltage recovery rate It represents the voltage relaxation rate after the current interruption. The physical mechanism involves the competition between the dissipation of concentration polarization and the interfacial impedance. The relaxation rate of the aged battery decreases and has an exponential correlation with the thickening of the SEI film impedance (R²>0.89), where and respectively represent the values of the battery voltage recovery at times and ;

[0057] Step 1.3: Process all the extracted features with a dynamic normalization strategy to obtain the data features of battery charge and discharge. Among them, the multi-source features after extraction (including voltage, current, temperature, and their differential / cumulative features) may exhibit significant scale differences and non-stationarity at different aging stages and under different working conditions. To improve the generalization ability of the model during long-term use, this paper introduces a dynamic normalization strategy (Dynamic Normalization) to perform time-varying standardization on all input features, and at the same time combines a window adaptive mechanism guided by the state of health (SOH-guided Adaptive Window Strategy) to achieve dynamic leveling of feature scales and smooth changes. Traditional normalization methods (such as Z-score, Min-Max) often use fixed statistics to perform global normalization on the entire data segment. However, for aging batteries, their feature distributions themselves change over time, which makes it difficult for static normalization methods to adapt to the non-stationarity of feature distributions and may lead to systematic biases in estimation errors. Therefore, this paper adopts a dynamic normalization strategy updated within a sliding window, and its mathematical expression is as follows: ;

[0058] where, is the output of the dynamic normalization strategy, is the original feature value at time step t; is a small constant to prevent division by zero; is the mean value, is the standard deviation, which are respectively expressed as: ; ;

[0059] is the dynamic sliding window length. Since the state of health (SOH) of the battery shows stage changes during the degradation process, if a fixed window length is used, it may lead to that in the initial stage of battery aging, the SOH is stable and the short window can quickly respond to instantaneous changes; but in the later stage of aging, the SOH declines rapidly, and the short window cannot effectively smooth mutations, resulting in violent fluctuations in normalization. Therefore, this paper proposes a strategy for dynamically adjusting the window length based on the SOH value, which is: ;

[0060] where, is the initial window length (tunable hyperparameter); is the growth rate coefficient; is the state of health predicted by the model at the current time step, then represents the degree of health decline;

[0061] Step 2: Feed the data features into a double-cross physical guidance model, which includes a spatio-temporal feature encoder, a two-stream mutual attention module, and a physical constraint decoder;

[0062] Step 3: The spatio-temporal feature encoder extracts high-dimensional feature tensors from the data features based on dilated convolution, multi-head self-attention, and a feed-forward network; in the spatio-temporal feature encoder, the data features are processed by a multi-head self-attention layer and a feed-forward network layer in sequence after dilated convolution, and residual connections are set in both the multi-head self-attention layer and the feed-forward network layer, and layer normalization is set after both the multi-head self-attention layer and the feed-forward network layer;

[0063] Among them, dilated convolution performs local perception and feature extraction on the original input; specifically, let the input sequence be , where is the length of the time series, is the feature dimension, represents the set of real numbers, then the output of the dilated convolution can be expressed as: ;

[0064] Among them, represents the convolution kernel size, is the dilation rate, is the learnable convolution kernel parameter, is usually a non-linear activation function; on this basis, by stacking or paralleling multiple layers of dilated convolution, multi-scale sampling of the sequence can be performed at different dilation rates, so as to capture the feature changes of battery data in the short-term, medium-term, and long-term time ranges;

[0065] If the output of the dilated convolution is denoted as , is the length of the time series, represents the dimension after the new feature mapping, then the multi-head self-attention can be expressed as: ;

[0066] Among them, are all linearly transformed by . Through the multi-head self-attention mechanism, multiple attention patterns can be learned in parallel in different subspaces, enabling the model to not only focus on the global trend but also capture local dependencies or special events;

[0067] After the output of the multi-head self-attention, a two-layer fully connected network is used for non-linear mapping, which can usually be written as:

[0068] in, is the output of multi-head self-attention, , , , All are learnable parameters, GELU is a commonly used activation function;

[0069] Step 4: The high-dimensional feature tensor is transformed through two parallel learnable projection heads in the feature-guided decoupling mechanism to generate two task-oriented feature streams, which are recorded as the charging state feature stream and the health state feature stream respectively. The two learnable projection heads are both fully connected networks with independent parameter update paths to learn the semantic space representation that best suits the corresponding tasks. The core idea of ​​this design is to complete the exclusive customization of the upper-level representation through the task diversion module on the basis of sharing the underlying encoder to capture the unified temporal features, thereby enhancing the collaborative modeling capability among multiple tasks.

[0070] Among them, the high-dimensional feature tensor can be recorded as: ;

[0071] Here, T represents the number of time steps, and d represents the feature dimension after encoding. In order to further serve the dual-task modeling objectives (SOC and SOH), the charge state feature flow is: , the health status characteristic flow is: ;

[0072] Step 5: The dual-stream mutual attention module calculates the cross-attention information between the charging state feature stream and the health state feature stream, and dynamically regulates and fuses the cross-attention information through the dynamic interaction mechanism to obtain the final charging state dynamic interaction results and health state dynamic interaction results; the cross-attention information includes the attention from the charging state to the health state. And the attention from health status to charging status ;in, is the characteristic flow of the charging state ( ), and All are health status characteristic flows ( ) in the feature vector mapping; is the scaling factor; is the health status feature flow ( ), and All are characteristic flows of the charging state ( Feature vector mapping in (); the dynamic interaction mechanism is adaptively regulated through the gating function and the interaction scoring function, and the gating function value and the interaction scoring function value are respectively weighted and fused with and to enable the model to dynamically determine "how much to fuse" and "how to fuse" the information from the feature streams of each other at different aging stages and under different working conditions;

[0073] Gating Function

[0074] First, define an average gating function to characterize the interaction willingness or interaction intensity of the two feature streams at the current stage; use the sigmoid function to map the learnable parameter W to the interval of 0-1: ;

[0075] Let W receive the predicted values of SOC and SOH as inputs at the same time, so that the gating can better reflect the actual battery state; if the G value is large, it means that the model relies more on cross-attention fusion; if the G value is small, it means that the independent features of this stream are more important;

[0076] Interaction Scoring Function (score)

[0077] To make the gating have higher decision-making flexibility in different scenarios, further define the interaction scoring function , where and respectively represent the SOC and SOH predicted by the model currently; measure the consistency or complementarity of SOC and SOH:

[0078] Among them, represents the non-linear mapping of the difference or correlation degree of SOC and SOH, is an adjustable hyperparameter; if the difference between the two is large or the complementarity is strong, the score value may be higher; if the correlation degree between the two is weak or the difference is too small, the score value is low;

[0079] Dynamic Fusion Strategy

[0080] After obtaining G and score, the weighted fusion of the cross-attention output can be carried out to form the final dynamic interaction result. For the SOC stream, it can be coupled in the following way:

[0081] Among them, Denotes element-wise multiplication, which is used to further perform fine-grained adjustment on the cross-attention output according to the score. Similarly, a similar gating and weighting strategy can also be adopted for the SOH stream; through such a dynamic mechanism, the model can reduce the impact of mutual attention on the final output in the early aging stage (when the relationship between SOC and SOH is relatively loose); while in the later aging stage (as the capacity attenuation intensifies and the correlation between the two increases), the mutual attention weight is increased to capture more significant coupling effects;

[0082] Step 6: The physical constraint decoder maps the final dynamic interaction results of the state of charge and the dynamic interaction results of the state of health to the predicted values of the state of charge and the state of health based on a multi-layer perceptron and a physical correction term; specifically:

[0083] Step 6.1: Input the dynamic interaction results of the state of charge and the dynamic interaction results of the state of health into two multi-layer perceptrons respectively to obtain preliminary and ;

[0084] Step 6.2: In each time segment, by integrating or recursively calculating variables such as the cumulative charge and temperature in the current period, the Coulomb integral constraint term and the Arrhenius degradation term are calculated respectively;

[0085] Step 6.3: Fuse the preliminary and with the Coulomb integral constraint term and the Arrhenius degradation term at the end of the decoder to form the final and ;

[0086] Among them,

[0087] Based on Coulomb's law constraint, its continuous form is: ;

[0088] Among them, is the rated capacity, When the Coulomb efficiency is discretely implemented, the integral can be replaced by step-by-step accumulation, and is clamped at each time step to prevent numerical drift; in the decoder, the Coulomb constraint often appears in the form of a CoulombConstraint layer, first estimating based on the current period's current, temperature and other characteristics of the network, and then coupling it with the output of the MLP; the simplified coupling method is: ;

[0089] Among them, is the Sigmoid function to ensure that the output is in the range of [0, 1]; is an adjustable weight or gating factor used to balance data-driven prediction and physical integration results; in this way, the network can automatically correct the SOC deviation under most working conditions and maintain the physical consistency of charge conservation;

[0090] Based on the Arrhenius model, the capacity decay and life decay rates are related to temperature and cycle depth: ;

[0091] Among them, is the cumulative discharge amount or the number of cycles, represents the decay nonlinear coefficient; the core of this equation is to dynamically calculate the decay amount according to the current temperature and cumulative capacity (or cycle depth) and output a correction term that can be multiplied or added to the predicted by the MLP; the coupling method is: ;

[0092] Among them, is the Sigmoid function. Through such a joint output layer, the model can not only maintain high prediction accuracy but also strictly follow electrochemistry priors such as charge conservation and temperature-reaction rate, providing more physically credible results for subsequent applications;

[0093] In addition, the dual-cross physically guided Transformer model is constrained and optimized from three dimensions: prediction error , physical consistency and cross-consistency . The overall loss is: ;

[0094] Among them and are hyperparameters used to balance the contribution degrees of different sub-losses; ; ; ;

[0095] Among them, and respectively represent the predicted values of the model for the state of charge and the state of health, and respectively represent the true values of the model for the state of charge and the state of health; is a tuning parameter used to balance the prediction capacity and the deviation between the calculated actual charge transfer and the integral value of the actually observed current This term ensures that the model output conforms to the principle of charge conservation, avoiding serious deviation of SOC estimation due to pure data-driven; is a tuning parameter used to control the degree of consistency with the Arrhenius equation ; is the rated capacity of a new battery; represents a physical quantity inherent in the battery material. A is the pre-exponential factor in the Arrhenius model, which is set as a constant here; T here is the current battery temperature, and I is the current.

[0096] A joint verification experiment was carried out on the inventive method. The results are shown in Table 1, listing the SOC-SOH joint estimation error indexes for multiple experimental samples such as batteries B201, B202, B203, etc., including MAE, RMSE, and MAXE.

[0097] Figure 3 The SOC prediction results of battery B201 during multiple charge and discharge cycles are shown. The red curve is the true SOC, and the blue curve is the model prediction value; it can be observed that the model can accurately track the true curve within a wide charge and discharge range. Especially in the discharge platform and fast charging stage, the error between the two always remains at a low level, with only small deviations at individual peaks and valleys; this kind of deviation is usually related to instantaneous rate change or polarization internal resistance fluctuation, indicating that the model can still maintain good robustness when facing relatively complex dynamic working conditions; overall, the SOC prediction curve shows a highly consistent trend with the true value over time, proving the effectiveness of the multi-task fusion and physical constraint strategy in capturing charge conservation and charge and discharge behavior;

[0098] and combined with Figure 4 and Figure 5 as shown, the overall error level of each battery in SOC prediction is relatively low, and the maximum error is usually controlled within an acceptable range, indicating that the model can maintain good fitting under different cycle working conditions and health levels; for SOH, although the attenuation mechanism is more complex, the error can still be maintained in a relatively small interval, especially in the middle and early stages of attenuation, the prediction curve basically coincides with the true value; in summary, it can be seen that after the joint estimation method integrates multi-task learning and physical constraints, it can not only effectively capture the capacity attenuation law of the battery at different aging stages, but also accurately depict the current charging state, achieving stable and reliable joint prediction under most working conditions;

[0099] In summary, after integrating multi-task learning and physical constraints, the joint estimation method can not only effectively capture the capacity attenuation law of the battery at different aging stages, but also accurately characterize the current state of charge, achieving stable and reliable joint prediction under most working conditions.

[0100] Table 1: Results of SOC-SOH Joint Estimation Evaluation Index

[0101] In the experiment of comparative estimation of the SOH model, we used the "Dual Cross Physical Guided Model" (DCPGM) and compared its performance with three classical comparative models: Support Vector Machine (SVM), Extreme Learning Machine (ELM), and Bidirectional Long Short-Term Memory Network (BiLSTM); through the performance comparison of these models, we verified the advantages of DCPGM in SOH estimation; Table 2: Results of SOH Comparative Experiment

[0102] As shown in Table 2, the experimental results show that DCPGM performs excellently in SOH estimation; specifically, the MAE of DCPGM is 0.003, the RMSE is 0.004, and the MAXE is 0.024. These results are significantly lower than those of the other three comparative models. Especially in the MAXE (maximum error) index, DCPGM successfully controls the maximum error at a low level; this model can accurately capture the health state of the battery at different degradation stages, especially in the early and middle stages of degradation, showing strong stability and accuracy; this indicates that DCPGM can effectively cope with the complex changes in the battery health state, especially in the initial and middle stages of battery degradation, and can stably provide accurate SOH predictions;

[0103] In contrast, the SVM and ELM models perform poorly in SOH estimation. Especially when the battery degradation is more significant, the maximum error (MAXE) of SVM is 0.052, much higher than that of DCPGM. This shows that SVM cannot accurately track the health state of the battery and has a large prediction deviation when dealing with the relatively fast changes in battery health; the MAXE of ELM is 0.061. Although ELM has a high training efficiency, its accuracy in SOH estimation is still insufficient, especially in the early stage of battery aging, and its performance cannot compare with that of DCPGM;

[0104] In summary, DCPGM demonstrates obvious advantages in SOH estimation. Through the dual-cross physical guidance framework, DCPGM can better capture the non-linear characteristics during the battery degradation process and maintain high prediction accuracy at different degradation stages of the battery. Compared with traditional SVM and ELM models, DCPGM can more accurately reflect the health state of the battery, especially during the battery degradation process. Although BiLSTM can also handle time series data, it fails to fully consider the coupling relationship between SOC and SOH, so its accuracy is slightly inferior to DCPGM. Therefore, DCPGM shows excellent performance in battery health state estimation, especially in complex battery degradation processes, showing higher accuracy and stability.

[0105] In the experiment of SOC model comparative estimation, we used the dual-cross physical guidance model (DCPGM) and compared its performance with three other comparative models: Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), and Particle Swarm Optimization Combined with Long Short-Term Memory Network (PSO-LSTM). Through the performance comparison of these models, we verified the advantages of DCPGM in SOC estimation. Table 3: Results of SOC Comparative Experiment

[0106] As shown in Table 3, the experimental results show that DCPGM performs the best in SOC estimation, with an MAE of 0.003, an RMSE of 0.004, and a MAXE of 0.019. All error metrics are significantly lower than those of other comparative models. Through the dual-cross physical guidance framework, DCPGM can accurately capture the short-term dynamic response of the battery, especially during the rapid charging and discharging process of the battery, showing high accuracy and stability. This indicates that DCPGM can not only handle the time series data of the battery well but also maintain high precision under complex battery operating conditions.

[0107] In contrast, CNN also performs well in SOC estimation, but it is still inferior to DCPGM. The MAE of CNN is 0.004, the RMSE is 0.005, and the MAXE is 0.050. Although CNN can effectively extract features during the battery charge and discharge process, its prediction accuracy is relatively low under complex battery states, especially during battery aging or rapid charge and discharge, where the error is relatively large; The LSTM model can better handle the time-series data of the battery, with a MAE of 0.003, an RMSE of 0.004, and a MAXE of 0.048; The performance of the LSTM model is close to that of DCPGM, but under some complex dynamic working conditions, the prediction accuracy of LSTM is slightly insufficient, especially during the rapid change process of the battery, where the error is large; PSO-LSTM optimizes the hyperparameters of LSTM through particle swarm optimization. Its MAE and RMSE are both close to DCPGM, but it performs slightly worse in terms of MAXE, with a maximum error of 0.045; Although PSO-LSTM can improve the accuracy of LSTM, there are still certain errors when dealing with the complex dynamic characteristics of the battery;

[0108] In summary, DCPGM shows obvious advantages in SOC estimation. By combining physical-guided and data-driven learning capabilities, DCPGM can accurately capture the short-term dynamic response and long-term degradation characteristics of the battery. Especially during rapid charge and discharge, it maintains high accuracy and stability; In contrast, other models have larger errors when dealing with complex battery states, especially in the case of battery aging or rapid changes, and DCPGM performs more excellently.

[0109] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A SOC-SOH joint estimation method based on a dual crossover physical guidance framework, characterized by: The following steps are involved: Step 1: Obtain data characteristics of battery charging and discharging; Step 2: Feed the data features into a dual cross-physics guidance model, which includes a spatiotemporal feature encoder, a two-stream mutual attention module, and a physical constraint decoder; Step 3: The spatiotemporal feature encoder extracts high-dimensional feature tensors from data features based on dilated convolution, multi-head self-attention, and feedforward networks; Step 4: The high-dimensional feature tensor is transformed through two parallel learnable projection heads in the feature-guided decoupling mechanism to generate two task-oriented feature streams, which are recorded as the charging state feature stream and the health state feature stream respectively; Step 5: The dual-stream mutual attention module calculates the cross-attention information between the charging state feature stream and the health state feature stream, and dynamically regulates and fuses the cross-attention information through the dynamic interaction mechanism to obtain the final charging state dynamic interaction results and health state dynamic interaction results; Step 6: The physical constraint decoder maps the final charging state dynamic interaction results and health state dynamic interaction results to the charging state prediction value and the health state prediction value based on the multi-layer perceptron and the physical correction term.

2. The SOC-SOH joint estimation method based on the dual crossover physical guidance framework according to claim 1 is characterized by: In the spatiotemporal feature encoder, the data features are processed by the multi-head self-attention layer and the feedforward network layer in sequence after the dilated convolution, and residual connections are set in both the multi-head self-attention layer and the feedforward network layer, and layer normalization is set after both the multi-head self-attention layer and the feedforward network layer.

3. The SOC-SOH joint estimation method based on the dual crossover physical guidance framework according to claim 1 is characterized by: Cross-attention information includes attention from charging state to health state And the attention from health status to charging status ; in, is the feature vector mapping in the charging state feature stream, and They are all feature vector mappings in the health status feature stream; is the scaling factor; is the feature vector mapping in the health status feature stream, and They are all feature vector mappings in the charging state feature stream.

4. The SOC-SOH joint estimation method based on the dual crossover physical guidance framework according to claim 3 is characterized by: The dynamic interaction mechanism is adaptively regulated by the gating function and the interaction scoring function, and the gating function value and the interaction scoring function value are respectively , Perform weighted fusion.

5. The SOC-SOH joint estimation method based on the dual crossover physical guidance framework according to claim 1 is characterized by: Step 6 is as follows: Step 6.1: Input the dynamic interaction results of the charging state and the health state into two multi-layer perceptrons respectively to obtain preliminary and ; Step 6.2: In each time segment, the Coulomb integral constraint terms are calculated by integrating or recursively calculating the accumulated electricity and temperature variables in the current period. and the Arrhenius degenerate term ; Step 6.3: Initial and Respectively with the Coulomb integral constraint term and the Arrhenius degenerate term Fusion is performed at the end of the decoder to form the final and .

6. The SOC-SOH joint estimation method based on the dual crossover physical guidance framework according to claim 5 is characterized by: ； in, is the rated capacity, When the Coulomb efficiency is discretized, the integral can be replaced by a step-by-step accumulation, and Clamping is done to prevent numerical drift; in the decoder, Coulomb constraints often appear in the form of CoulombConstraint layers, which first estimate the current, temperature and other characteristics of the network during the current period. , and then with the output of MLP To couple; the simplified coupling method is: ； in, Sigmoid function is used to ensure that the output is in the interval [0,1]; is an adjustable weight or gating factor used to balance data-driven predictions with physical integration results.

7. The SOC-SOH joint estimation method based on a dual crossover physical guidance framework according to claim 5 is characterized by: ； in, is the cumulative discharge amount or number of cycles, Represents the attenuation nonlinear coefficient; the core of this equation is to dynamically calculate the attenuation amount based on the current temperature and cumulative capacity, and output a Multiplicative or additive correction terms; the coupling method is: ； in, is the Sigmoid function.

8. The SOC-SOH joint estimation method based on a dual crossover physical guidance framework according to claim 1 is characterized in that: From the prediction error , physical consistency and cross-consistency The dual cross-physics guided Transformer model is constrained and optimized in three dimensions, with an overall loss for: ； in and is a hyperparameter used to balance the contribution of different sub-losses; ； ； ； in, and Respectively represent the model's predicted values ​​for the state of charge and state of health, and Respectively represent the true values ​​of the model for the state of charge and the state of health; is a tuning parameter used to balance the predicted capacity The actual observed current integral value Deviation between calculated and actual charge transfer; To adjust the parameters, used to control Arrhenius equation degree of consistency; is the rated capacity of a new battery; It represents the physical quantity inherent to the battery material, A is the pre-exponential factor in the Arrhenius model, T is the current battery temperature, and I is the current.

9. The SOC-SOH joint estimation method based on a dual crossover physical guidance framework according to claim 1, characterized in that: Step 1 specifically includes the following steps: Step 1.1: Establish an experimental platform for battery cycle life test, conduct cycle life test on lithium-ion batteries and obtain battery charge and discharge data; Step 1.2: Perform multi-scale processing and derivative feature extraction on the original signal from the battery charge and discharge data. The features obtained include voltage change rate, temperature gradient, cumulative throughput, equivalent cycle number, sliding window statistics, and relaxation period voltage recovery rate; Step 1.3: All extracted features are processed using a dynamic normalization strategy to obtain the data features of battery charging and discharging.

10. The SOC-SOH joint estimation method based on a dual crossover physical guidance framework according to claim 9, characterized in that: The dynamic normalization strategy is mathematically expressed as follows: ； in, is the output of the dynamic normalization strategy, is the original eigenvalue at time step t; To prevent division by a small constant of 0; is the mean, is the standard deviation, respectively expressed as: ； ； is the dynamic sliding window length and is: ； in, is the initial length of the window; is the growth rate coefficient; is the health status predicted by the model at the current time step, then Indicates the degree of health decline.

Images