Accurate prediction and evaluation method and system for battery life based on deep learning

Abstract

The invention provides a deep learning-based battery life accurate prediction and evaluation method and system, and relates to the technical field of battery life prediction, and the method comprises the steps: obtaining battery operation data, carrying out the process division based on change features, employing a state observation model and a variational inference network to extract time sequence features, and obtaining dynamic features through phase-space reconstruction; establishing a deep feature enhancement network to obtain comprehensive features; features are extracted through capacity and internal resistance prediction branches, segmented recursive coupling channels are established for fusion, and a prediction result is output; determining the residual life in combination with a scrapping threshold value; according to the invention, the battery life prediction precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery life prediction, and particularly to an accurate prediction and evaluation method and system for battery life based on deep learning. Background Art

[0002] With the rapid development of the new energy vehicle industry, as the core component, the performance and life of the power battery directly affect the use performance and safety of the vehicle. Accurately predicting the remaining service life of the battery is of great significance for ensuring the safe operation of new energy vehicles and optimizing the battery management strategy. At present, the battery life prediction methods mainly include physical model-based methods, data-driven methods, and deep learning-based methods. Among them, the deep learning method can adaptively extract battery degradation features and shows good application prospects in the field of battery life prediction.

[0003] However, the existing deep learning-based battery life prediction methods still have some deficiencies: the coupling relationship between battery capacity attenuation and internal resistance growth is not fully considered, resulting in insufficient prediction accuracy; the feature extraction method for battery historical operation data is relatively simple, and it is difficult to effectively capture the multi-scale dynamic features during the battery degradation process; the dependence analysis of the battery working state is lacking, and the differential features of battery performance degradation under different working conditions cannot be accurately described; the existing prediction model structure is single, and the feature fusion ability is limited, making it difficult to adapt to the complex and changeable battery degradation process.

[0004] In summary, there is a need for an accurate prediction and evaluation method for battery life based on deep learning, which extracts battery dynamic features through phase space reconstruction and symplectic geometry optimization, establishes a deep feature enhancement network to achieve feature optimization, and adopts a state-dependent segmented recursive coupling structure to realize the bidirectional fusion of capacity attenuation features and impedance evolution features, thereby improving the accuracy and reliability of battery life prediction. Summary of the Invention

[0005] The embodiments of the present invention provide an accurate prediction and evaluation method and system for battery life based on deep learning, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention, There is provided an accurate prediction and evaluation method for battery life based on deep learning, including: Obtaining the battery charge and discharge voltage, battery charge and discharge current, battery temperature, battery internal resistance, and battery cycle number as battery historical operation data; Based on the change characteristics of the battery historical operation data, performing process division, using a state observation model and a variational inference network to extract multi-process features to form a time series feature vector, and obtaining battery dynamic features through phase space reconstruction and symplectic geometry optimization; Build a deep feature enhancement network to perform manifold projection and non - linear transformation on the battery dynamic features, and optimize and model through a deep learning framework to obtain the comprehensive battery features; Input the comprehensive battery features into the shared encoding layer to obtain the time - series calibration features, extract the capacity attenuation features and impedance evolution features through the capacity prediction branch and the internal resistance prediction branch, and establish a state - dependent piece - wise recursive coupling channel for feature fusion, and output the capacity prediction result and the internal resistance prediction result; According to the capacity prediction result and the internal resistance prediction result, combine the battery scrapping threshold to determine the remaining service life of the battery.

[0007] In an alternative embodiment, Based on the change characteristics of the battery historical operation data, perform process division, use the state observation model and the variational inference network to extract multi - process features to form a time - series feature vector, and obtain the battery dynamic features through phase - space reconstruction and symplectic geometry optimization, including: Based on the change characteristics of the battery historical operation data, divide the battery historical operation data into charging process data, discharging process data and static process data; Establish state observation models for the charging process data, the discharging process data and the static process data respectively, calculate the state transition probability and the observation probability density through Gaussian distribution, and construct the probability distribution of the state observation model; Based on the probability distribution, construct a variational inference network. The variational inference network extracts the electrochemical state vector and thermodynamic features from the charging process data, extracts the energy efficiency features and dynamic response features from the discharging process data, extracts the polarization features and self - discharge features from the static process data, and combines the extracted features in time - series to form a time - series feature vector; Based on the evolution trajectory of the time - series feature vector in the phase space, construct a local potential function and a pairwise potential function, combine symplectic geometry optimization, and obtain the battery dynamic features by reconstructing the fractal features of the attractor.

[0008] In an alternative embodiment, Based on the evolution trajectory of the time - series feature vector in the phase space, construct a local potential function and a pairwise potential function, combine symplectic geometry optimization, and obtain the battery dynamic features, including: Describe the evolution trajectory of the time - series feature vector through the Hamiltonian dynamics equation, project the time - series feature vector into the phase space to form a phase orbit, calculate the local potential function based on the Lyapunov exponent of the phase orbit, and calculate the pairwise potential function based on the synchronization degree between phase orbits; Project the evolution trajectory of the time - series feature vector in the phase space onto the Poincaré section for reconstruction, and extract the electrochemical feature set and the thermodynamic feature set; Calculate the invariant measures of the set of electrochemical features and the set of thermodynamic features on the Poincaré section to obtain the electrochemical energy term and the thermodynamic energy term respectively; Construct a fractional differential operator based on the long-range correlation characteristics of the electrochemical energy term and the thermodynamic energy term, and calculate the coupling energy function based on the order of the fractional differential operator; Substitute the electrochemical energy term, the thermodynamic energy term, and the coupling energy function into the symplectic geometry optimizer, keep the Hamiltonian structure unchanged and minimize the global energy to obtain the optimized energy function; Reconstruct the attractor in the phase space based on the optimized energy function, calculate the fractal dimension and the correlation dimension of the attractor, and reconstruct the battery dynamic characteristics by combining the dimension features.

[0009] In an alternative embodiment, Establish a deep feature enhancement network, perform manifold projection and nonlinear transformation on the battery dynamic characteristics, and optimize the modeling through a deep learning framework to obtain the battery comprehensive characteristics including: Input the battery dynamic characteristics into the manifold embedding layer, calculate the reconstruction weight matrix through the adjacent points of the manifold embedding layer, solve the eigen-equation using the reconstruction weight matrix to obtain the first low-dimensional embedding representation, construct a geodesic distance matrix based on the first low-dimensional embedding representation, perform spectral decomposition on the geodesic distance matrix to obtain the eigenvector matrix and the eigenvalue matrix, and multiply the eigenvector matrix by the square root of the eigenvalue matrix to generate the manifold embedding representation; Input the manifold embedding representation into the dynamic kernel function enhancement module, construct multiple basic kernel functions using distance metrics, calculate the combined weights of the multiple basic kernel functions using a deep network, perform weighted combination of the combined weights and the multiple basic kernel functions to generate a combined kernel function, and perform nonlinear transformation on the manifold embedding representation using the combined kernel function to generate enhanced features; Input the enhanced features into the probability graph optimization module, establish a conditional probability field based on the graph structure, the conditional probability field uses a deep neural network to learn the correlation between the dimensional components of the enhanced features, construct a node distribution, and iteratively optimize through the message passing algorithm to obtain the battery comprehensive characteristics.

[0010] In an alternative embodiment, Input the enhanced features into the probability graph optimization module, establish a conditional probability field based on the graph structure, the conditional probability field uses a deep neural network to learn the correlation between the dimensional components of the enhanced features, construct a node distribution, and iteratively optimize through the message passing algorithm to obtain the battery comprehensive characteristics including: The probability graph optimization module constructs a conditional probability field for the enhanced features based on the graph structure, the nodes in the conditional probability field correspond to the dimensional components of the enhanced features, and the edge connection relationships between the nodes form an edge set; The conditional probability field calculates the correlation between the components of each dimension of the enhanced features through a deep neural network. The deep neural network maps nodes to a hidden layer feature space to obtain node hidden layer representations, and calculates the association strength between nodes based on the node hidden layer representations; An energy function of the conditional probability field is constructed according to the association strength, including a node potential term and an edge potential term. The edge potential term is modulated by the association strength, and a conditional probability distribution is defined based on the energy function; Under the conditional probability distribution, the conditional probability field iteratively transmits node information on the edge set through a message passing algorithm. The message passing algorithm updates the information transmitted between nodes based on the node potential term, the edge potential term, and the messages of adjacent nodes; The probability graph optimization module jointly optimizes the log-likelihood of the conditional probability distribution and the loss function of the deep neural network, alternately updates the parameters of the deep neural network and the parameters of the conditional probability field until convergence, and outputs the comprehensive battery features.

[0011] In an alternative embodiment, The comprehensive battery features are input into a shared encoding layer to obtain time series calibration features. Capacity decay features and impedance evolution features are extracted through a capacity prediction branch and an internal resistance prediction branch, and a state-dependent piecewise recursive coupling channel is established for feature fusion, and capacity prediction results and internal resistance prediction results are output, including: The comprehensive battery features are input into a shared encoding layer. The shared encoding layer performs layer-by-layer dimensionality reduction encoding through a multi-layer convolutional network to obtain encoded features, and extracts time series correlation in the encoded features to obtain time series calibration features; The time series calibration features are respectively input into a capacity prediction branch and an internal resistance prediction branch. The capacity prediction branch extracts capacity decay features from the time series calibration features through a multi-layer perceptron, and the internal resistance prediction branch extracts impedance evolution features from the time series calibration features through a recurrent neural network; A state-dependent piecewise recursive coupling channel is established between the capacity decay features and the impedance evolution features. Through state matrix transformation and multi-stage response equations, bidirectional dynamic transmission and fusion are realized to obtain capacity fusion features and internal resistance fusion features; Based on the capacity fusion features, capacity prediction results are output, and based on the internal resistance fusion features, internal resistance prediction results are output.

[0012] In an alternative embodiment, A state-dependent piecewise recursive coupling channel is established between the capacity decay features and the impedance evolution features. Through state matrix transformation and multi-stage response equations, bidirectional dynamic transmission and fusion are realized to obtain capacity fusion features and internal resistance fusion features, including: Establish a dynamic collaborative optimization channel between the capacity attenuation feature and the impedance evolution feature, construct a state-dependent matrix based on the battery operating state, and the state-dependent matrix includes a temperature gradient coefficient, a charge-discharge rate coefficient, and a cycle number coefficient; perform piecewise matrix transformation on the capacity attenuation feature and the impedance evolution feature according to the state-dependent matrix, establish independent feature mapping functions within each operating state interval, and achieve piecewise reconstruction of the features through function combination to obtain state-correlated features; Construct a capacity-internal resistance coupling model based on the state-correlated features. The capacity-internal resistance coupling model is recursively updated to pair the feature change points in the capacity degradation process with the internal resistance mutation points in time series, and establish a multi-stage capacity-internal resistance response equation to form a quantitative correlation relationship between capacity attenuation and internal resistance growth with state memory; Perform two-way feature compensation and enhancement on the capacity attenuation feature and the impedance evolution feature according to the capacity-internal resistance coupling model to obtain a capacity fusion feature and an internal resistance fusion feature combined with state dependence.

[0013] In the second aspect of the embodiments of the present invention, Provide a battery life accurate prediction and evaluation system based on deep learning, including: A first unit for obtaining battery charge-discharge voltage, battery charge-discharge current, battery temperature, battery internal resistance, and battery cycle number as battery historical operation data; A second unit for performing process division based on the change characteristics of the battery historical operation data, extracting multi-process features to form a time series feature vector by using a state observation model and a variational inference network, and obtaining the battery dynamic features through phase space reconstruction and symplectic geometry optimization; A third unit for establishing a deep feature enhancement network, performing manifold projection and non-linear transformation on the battery dynamic features, and optimizing and modeling through a deep learning framework to obtain the battery comprehensive features; A fourth unit for inputting the battery comprehensive features into a shared coding layer to obtain time series calibration features, extracting the capacity attenuation feature and the impedance evolution feature through a capacity prediction branch and an internal resistance prediction branch, and establishing a state-dependent piecewise recursive coupling channel for feature fusion, and outputting a capacity prediction result and an internal resistance prediction result; A fifth unit for determining the remaining service life of the battery according to the capacity prediction result and the internal resistance prediction result in combination with the battery scrapping threshold.

[0014] In the third aspect of the embodiments of the present invention, Provide an electronic device, including: A processor; A memory for storing processor-executable instructions; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In the fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In the embodiments of the present invention, through multi-process feature extraction and dynamic feature optimization, a comprehensive characterization of the battery operating state is achieved, improving the capture ability of the prediction model for the non-linear degradation behavior of the battery, thereby making the battery life prediction more accurate and reliable; by adopting a deep feature enhancement network and a manifold projection technique, the limitations of traditional methods in dealing with high-dimensional heterogeneous data are effectively solved, and the degradation patterns of the battery under different working conditions can be adaptively learned, enhancing the generalization ability and robustness of the model, and being applicable to complex and variable actual application scenarios; a state-dependent segmented recursive coupling channel is introduced to achieve a deep fusion of the capacity attenuation feature and the impedance evolution feature, overcoming the limitations of single-parameter prediction, and significantly improving the accuracy and reliability of the battery life prediction by comprehensively considering the mutual influence of various degradation mechanisms inside the battery, providing strong support for the optimization of the battery management system and the full-life cycle management of the battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic flowchart of the method for accurately predicting and evaluating the battery life based on deep learning in the embodiments of the present invention; Figure 2 is a schematic diagram for comprehensively evaluating the battery dynamic characteristics and the life prediction performance; Figure 3 is a contour group diagram of the correlation of the battery feature dimensions; Figure 4 is a visualization and accuracy comparison diagram of the state-dependent matrix. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0020] Figure 1 This is a schematic flowchart of the battery life accurate prediction and evaluation method based on deep learning according to an embodiment of the present invention. As Figure 1 shown, the method includes: Obtain the battery charge and discharge voltage, battery charge and discharge current, battery temperature, battery internal resistance, and battery cycle count as battery historical operation data; Based on the change characteristics of the battery historical operation data, perform process division, use a state observation model and a variational inference network to extract multi-process features to form a time series feature vector, and obtain the battery dynamic characteristics through phase space reconstruction and symplectic geometry optimization; Establish a deep feature enhancement network, perform manifold projection and non-linear transformation on the battery dynamic characteristics, and optimize the modeling through a deep learning framework to obtain the battery comprehensive characteristics; Input the battery comprehensive characteristics into a shared encoding layer to obtain time series calibration characteristics, extract capacity decay characteristics and impedance evolution characteristics through a capacity prediction branch and an internal resistance prediction branch, and establish a state-dependent segmented recursive coupling channel for feature fusion, and output the capacity prediction result and the internal resistance prediction result; According to the capacity prediction result and the internal resistance prediction result, determine the remaining service life of the battery in combination with the battery scrapping threshold.

[0021] In a specific embodiment, collect historical operation data such as the charge and discharge voltage, charge and discharge current, temperature, internal resistance, and cycle count of the battery during actual operation. Eliminate outliers and noise through data preprocessing, and perform normalization processing on the data to make various types of data have the same dimension.

[0022] Based on the change trend of the battery historical operation data, use the sliding time window method to divide the battery operation process, and identify different working stages such as the charging process, discharging process, and resting process. Establish a state observation model for each working stage, which includes a state transition equation and an observation equation, and is used to describe the battery state evolution law. Construct a variational inference network, which consists of an encoder and a decoder, and perform parameter estimation on the state observation model by maximizing the evidence lower bound to extract the feature vectors of each process. Combine the extracted feature vectors in time series to form a time series feature vector. Based on the time series feature vector, perform phase space reconstruction to determine the optimal embedding dimension and time delay, and reconstruct the phase space trajectory of the battery state. Apply the symplectic geometry optimization algorithm in the reconstructed phase space to maintain the energy conservation characteristics of the system and obtain the battery dynamic characteristics.

[0023] Construct a deep feature enhancement network, which includes a feature mapping layer, a feature enhancement layer, and a feature fusion layer. In the feature mapping layer, project the battery dynamic features into a high-dimensional manifold space to preserve the topological structure of the features. In the feature enhancement layer, extract the deep representation of the features through multi-layer non-linear transformation. In the feature fusion layer, use the attention mechanism to adaptively weight and fuse the features at different levels. Optimize the network parameters through the backpropagation algorithm to obtain the comprehensive battery features.

[0024] Design a shared encoding layer, use a multi-layer convolutional neural network to encode the comprehensive battery features, and extract the temporal correlation to obtain the temporal calibration features. In the capacity prediction branch, use a multi-layer perceptron network to process the temporal calibration features and extract the capacity decay features. In the internal resistance prediction branch, use a long short-term memory network to process the temporal calibration features and extract the impedance evolution features. Construct a state-dependent matrix, which includes a temperature gradient coefficient, a charge-discharge rate coefficient, and a cycle number coefficient. Perform a segmented matrix transformation on the capacity decay features and the impedance evolution features based on the state-dependent matrix. Establish a recursive coupling channel to achieve bidirectional transfer of features through a multi-stage response equation. Perform feature fusion to obtain the capacity fusion features and the internal resistance fusion features. Output the capacity prediction value and the internal resistance prediction value based on the fusion features respectively.

[0025] Set the scrap thresholds for the battery capacity and internal resistance. Based on the capacity prediction result and the internal resistance prediction result, combine the scrap thresholds to conduct a life assessment. Use the weighted average method to comprehensively consider the life prediction results in the two dimensions of capacity and internal resistance. Output the final predicted value of the remaining service life of the battery.

[0026] In an alternative embodiment, perform process partitioning based on the change characteristics of the battery historical operation data, use a state observation model and a variational inference network to extract multi-process features to form a temporal feature vector, and obtain the battery dynamic features through phase space reconstruction and symplectic geometry optimization, including: Based on the change characteristics of the battery historical operation data, divide the battery historical operation data into charging process data, discharging process data, and resting process data; Establish state observation models for the charging process data, the discharging process data, and the resting process data respectively, calculate the state transition probability and the observation probability density through Gaussian distribution, and construct the probability distribution of the state observation model; Construct a variational inference network based on the probability distribution. The variational inference network extracts the electrochemical state vector and thermodynamic features from the charging process data, extracts the energy efficiency features and dynamic response features from the discharging process data, extracts the polarization features and self-discharge features from the resting process data, and combines the extracted features in time series to form a temporal feature vector; Based on the evolution trajectory of the time-series feature vector in the phase space, a local potential energy function and a pairwise potential energy function are constructed. Combining symplectic geometry optimization, the dynamic characteristics of the battery are obtained by reconstructing the fractal characteristics of the attractor.

[0027] In a specific embodiment, parameters such as voltage, current, and temperature of the battery during actual use are collected. Based on the change characteristics of the current direction and amplitude, the historical operation data of the battery are divided into charging process data, discharging process data, and resting process data. Specifically, when the current value is positive and the duration exceeds a preset threshold (such as 30 seconds), it is determined as the charging process; when the current value is negative and the duration exceeds the preset threshold, it is determined as the discharging process; when the current value is close to zero (such as the absolute value is less than 0.05C) and the duration exceeds the preset threshold, it is determined as the resting process.

[0028] Taking the charging process as an example, voltage, current, and temperature are selected as the observed variables, and the state of charge (SOC), internal resistance, polarization voltage, etc. are set as the state variables. The state transition probability is calculated through the Gaussian distribution, that is, the conditional probability relationship between the state variables at the current moment and the previous moment. For example, for a certain 18650 lithium battery, the state transition probability of the SOC changing from 0.5 to 0.51 during the charging process can be calculated through a Gaussian distribution with a mean of 0.51 and a variance of 0.0001. Similarly, the observed probability density is calculated, that is, the conditional probability relationship between the observed variables and the state variables. For example, when the SOC is 0.8, the observed probability of the voltage can be represented by a Gaussian distribution with a mean of 4.05V and a variance of 0.01. Similar methods are used to establish the state observation model for the discharging process and the resting process.

[0029] Based on the above probability distribution, a variational inference network is constructed, which adopts an encoder-decoder structure. The encoder maps the observed data to the latent state space, and the decoder reconstructs the latent state into the observed data. For the charging process data, the electrochemistry state vector (including the state of charge, electrochemical reaction rate, etc.) and thermodynamic characteristics (including entropy change, enthalpy change, etc.) are extracted through the variational inference network. For example, the analysis of the charging data of a certain type of lithium battery at 25°C shows that in the SOC range of 0.7 - 0.8, the average value of its electrochemical reaction rate is 0.023 mol / (L·s), and the average value of the entropy change is 85 J / (mol·K).

[0030] For the discharging process data, the energy efficiency characteristics and dynamic response characteristics are extracted. The energy efficiency characteristics include indicators such as Coulomb efficiency and energy efficiency, and the dynamic response characteristics include voltage response time constant, current step response, etc. For example, when a certain power battery is tested at a 1C discharge rate, the Coulomb efficiency is 0.985, and the voltage response time constant is 12.5 seconds.

[0031] For the static process data, polarization characteristics and self-discharge characteristics are extracted. Polarization characteristics include polarization resistance, polarization time constant, etc. Self-discharge characteristics include self-discharge rate, capacity loss rate, etc. For example, a certain energy storage battery is static for 24 hours at 25°C, and the self-discharge rate is measured to be 0.15% / day, and the polarization resistance is 25mΩ.

[0032] The extracted features are combined in time series to form a time series feature vector. For data with a sampling interval of 1 second, a set of features is extracted every 10 minutes to form a time series feature vector including electrochemical state vector, thermodynamic features, energy efficiency features, dynamic response features, polarization features and self-discharge features.

[0033] Finally, based on the evolution trajectory of the time series feature vector in the phase space, the dynamic characteristics of the battery are constructed, the phase space is reconstructed, the embedding dimension is selected as 5, the time delay is 3, and the time series feature vector is mapped into the phase space. Then a local potential energy function is constructed to describe the energy distribution of a single state point; a paired potential energy function is constructed to describe the interaction between state points. For example, the analysis of a certain model of battery shows that in the SOC range of 0.4-0.6, the local potential energy function presents a bowl-shaped distribution, indicating that the system has strong stability in this range.

[0034] The energy conservation characteristics of the system are maintained by combining the symplectic geometry optimization method, and the optimal state trajectory of the system is solved by an iterative optimization algorithm (such as the conjugate gradient method). During the optimization process, the learning rate is set to 0.01, the maximum number of iterations is 1000, and the convergence threshold is 0.0001.

[0035] The correlation dimension, Lyapunov index and other indicators of the attractor are calculated to characterize the complexity and stability of the battery system. For example, when a certain type of battery is in normal working condition, the correlation dimension of its attractor is 2.35, and the maximum Lyapunov index is 0.023, indicating that the system has certain chaotic characteristics but remains stable overall. The attractor specifically refers to the set of states that the system eventually tends to and stays for a long time during the evolution process. Specifically, for a large class of initial states, the corresponding system trajectory will gradually approach and stabilize on this set over time. Attractors can have different forms, such as: point attractors (fixed points): the system state eventually converges to a stable equilibrium point; periodic attractors (limit cycles): the system state cyclically moves along a periodic orbit; strange attractors: complex attractors with fractal structures, which often appear in chaotic systems. In the battery dynamic feature extraction method, by analyzing the evolution trajectory of the time series feature vector in the phase space, constructing local and pairwise potential energy functions, and combining symplectic geometry optimization, the fractal characteristics of the attractor are reconstructed to reflect the dynamic behavior of the battery. That is, the attractor represents the dynamic stable state or cyclic behavior exhibited by the battery in long-term operation.

[0036] Through the above method, rich dynamic features can be extracted from the historical operation data of the battery, providing strong support for battery state assessment, life prediction, and fault diagnosis. Experimental verification shows that for a certain type of lithium-ion battery, when the dynamic features extracted by this method are used for capacity prediction, the prediction accuracy is improved by 15.7% compared with the traditional method, and the root mean square error is reduced to 0.023 Ah.

[0037] In an alternative embodiment, based on the evolution trajectory of the time series feature vector in the phase space, a local potential energy function and a pairwise potential energy function are constructed. Combining symplectic geometry optimization, the dynamic features of the battery are obtained by reconstructing the fractal features of the attractor, including: The evolution trajectory of the time series feature vector is described by the Hamiltonian dynamics equation. The time series feature vector is projected into the phase space to form a phase orbit. The local potential energy function is calculated based on the Lyapunov exponent of the phase orbit, and the pairwise potential energy function is calculated based on the synchronization degree between phase orbits; The evolution trajectory of the time series feature vector in the phase space is projected onto the Poincaré section for reconstruction, and an electrochemical feature set and a thermodynamic feature set are extracted; The invariant measures of the electrochemical feature set and the thermodynamic feature set on the Poincaré section are calculated to obtain the electrochemical energy term and the thermodynamic energy term respectively; Based on the long-range correlation characteristics of the electrochemical energy term and the thermodynamic energy term, a fractional differential operator is constructed, and the coupled energy function is calculated based on the order of the fractional differential operator; The electrochemical energy term, the thermodynamic energy term, and the coupled energy function are substituted into the symplectic geometry optimizer to keep the Hamiltonian structure unchanged and minimize the global energy, obtaining the optimized energy function; Based on the optimized energy function, the attractor in the phase space is reconstructed, and the fractal dimension and correlation dimension of the attractor are calculated. Combining the dimension features, the dynamic features of the battery are reconstructed.

[0038] In a specific embodiment, the time series data during the operation of the battery is obtained, including parameters such as voltage, current, and temperature. The original data is preprocessed, including denoising, normalization, and dimensionality reduction, to obtain the time series feature vector. Taking a certain type of lithium battery as an example, the sampling frequency is 10 Hz, and data is continuously collected for 4 hours. After wavelet transform denoising and principal component analysis dimensionality reduction, an 8-dimensional time series feature vector is obtained.

[0039] The time series feature vector is projected into the phase space to form an evolution trajectory. The time delay τ = 20 and the embedding dimension m = 8 are selected to construct the phase space. In the phase space, each point represents the state of the system at a specific moment, and the connection between adjacent points forms a trajectory. For the example battery, the phase space trajectory shows non-linear characteristics, indicating the complexity of the battery dynamic system.

[0040] Based on the phase space trajectory, calculate the local potential energy function. Select a reference point in the phase space and calculate the divergence rate of the trajectories near this point, i.e., the Lyapunov exponent. For the example battery, 500 reference points are uniformly selected in the phase space, and the local Lyapunov exponent of each point is calculated. The numerical range is between [-0.15, 0.23]. The positive value region represents the instability of the system, and the negative value region represents the stability of the system. Use these exponent values as the basis for the local potential energy function. The higher the potential energy value, the more unstable the system is.

[0041] Calculate the pairwise potential energy function to characterize the interaction between different trajectories in the phase space. Select two trajectories and calculate their synchronization degree, including phase synchronization and generalized synchronization. For the example battery, calculate the synchronization index between any two points in the phase space. The numerical range is between [0, 1], where 0 represents completely asynchronous and 1 represents completely synchronous. Construct the pairwise potential energy function based on the synchronization index. The lower the synchronization degree, the higher the potential energy value.

[0042] Project the evolution trajectory of the time series feature vector in the phase space onto the Poincaré section. Select a hyperplane in the phase space and record the intersection sequence of the trajectory and the hyperplane. For the example battery, select the hyperplane with the first eigenvalue being zero as the Poincaré section and obtain about 2000 intersection points.

[0043] Extract the electrochemical feature set and the thermodynamic feature set from the points on the Poincaré section. The electrochemical feature set includes parameters such as charge transfer resistance, double-layer capacitance, and diffusion coefficient; the thermodynamic feature set includes parameters such as entropy change, enthalpy change, and free energy. For the example battery, the electrochemical feature set contains 6 parameters, and the thermodynamic feature set contains 4 parameters.

[0044] Calculate the invariant measure of the electrochemical feature set and the thermodynamic feature set on the Poincaré section, respectively obtain the electrochemical energy term and the thermodynamic energy term. Statistically analyze the distribution density of the feature points on the Poincaré section, construct the probability density function, and calculate the information entropy as the invariant measure. For the example battery, the numerical value of the electrochemical energy term is 3.27, and the numerical value of the thermodynamic energy term is 2.85.

[0045] Based on the long-range correlation characteristics of the electrochemical energy term and the thermodynamic energy term, construct a fractional differential operator, analyze the cross-correlation function of the two energy terms, determine the strength of the long-range dependence, and map it to the order of the fractional differential operator. For the example battery, by calculating the decay characteristics of the cross-correlation function, determine that the order of the fractional differential operator is 0.78.

[0046] Calculate the coupled energy function based on the order of the fractional differential operator. Substitute the electrochemical energy term and the thermodynamic energy term into the fractional differential equation and solve to obtain the coupled energy function. For the example battery, the numerical value of the coupled energy function is 1.92.

[0047] Substitute the electrochemical energy term, thermodynamic energy term, and coupling energy function into the symplectic geometry optimizer. The symplectic geometry optimizer keeps the Hamiltonian structure unchanged and minimizes the global energy through an iterative approach. In specific implementation, the symplectic Euler method is used for numerical solution, with an iterative step size of 0.01 and a maximum number of iterations of 1000. For the example battery, after 723 iterations, the global energy is reduced to 37% of the original value, and the optimized energy function value is 2.96.

[0048] Reconstruct the attractor in the phase space based on the optimized energy function. Use the optimized energy function as the potential energy field of the system, and reconstruct the phase space trajectory through numerical integration to form the attractor structure. For the example battery, the reconstructed attractor presents a complex geometric structure.

[0049] Calculate the fractal dimension and correlation dimension of the attractor. The fractal dimension is calculated by the box-counting method, and the correlation dimension is calculated by the correlation integral method. For the example battery, the fractal dimension is 2.37 and the correlation dimension is 2.15, indicating that the system has chaotic characteristics.

[0050] Reconstruct the dynamic characteristics of the battery in combination with the dimension features. Establish a mapping relationship between the fractal dimension and correlation dimension and the battery performance parameters, and construct a battery dynamic feature vector. For the example battery, the final obtained dynamic characteristics include key indicators such as a stability index of 0.82, a complexity index of 0.75, and a health state score of 0.93. These indicators can be used for battery state monitoring and life prediction.

[0051] such as Figure 2As shown, it comprehensively presents the performance evaluation results of the battery dynamic feature analysis technology, including three parts: a radar chart, a performance curve graph, and a data table. The radar chart on the left shows the comparison results of six core performance indicators: This technical solution (solid circular marker) is significantly superior to the traditional feature method (dashed square marker) and the deep learning method (dash-dotted diamond marker) in six dimensions, namely stability index (0.82), complexity index (0.75), health state score (0.93), remaining life accuracy (0.94), fault warning sensitivity (0.87), and dynamic response speed (0.89). Especially in terms of the health state score and remaining life accuracy, this technical solution has achieved high performances of 0.93 and 0.94 respectively, indicating its significant advantages in battery state evaluation and life prediction. The performance curve graph on the right shows the changing trends of the performance scores of the three methods in different working cycles (initial stage, middle stage, later stage, and decline stage) of the battery. The performance curve (solid line) of this technical solution remains at a high level throughout the life cycle. Especially in the middle and later stages and the decline stage, the degree of its performance decline is significantly less than that of the other two methods. Importantly, this technical solution can detect the performance decline trend before the battery enters the decline stage (at the warning point position), and the warning time point is significantly earlier than that of other methods. The bottom table provides specific performance evaluation data: The prediction accuracy of this technical solution is 94.3%, which is 23.3% higher than that of the traditional feature method (76.5%) and 9.5% higher than that of the deep learning method (86.1%); the early warning rate is 92.7%, which is 35.9% higher than that of the traditional feature method (68.2%) and 16.6% higher than that of the deep learning method (79.5%); the computational complexity is 0.75 (normalized value). Although it is higher than that of the traditional feature method (0.35), it is lower than that of the deep learning method (1.00), indicating that this technical solution ensures computational efficiency while maintaining high accuracy. Generally speaking, by combining advanced technologies such as Hamiltonian dynamics equations, Poincaré section reconstruction, fractional-order differential operators, and symplectic geometry optimizers, this technical solution has achieved high-precision analysis and prediction of battery dynamic features, providing strong technical support for battery health management and life prediction.

[0052] In this embodiment, the Hamiltonian dynamics equation is used to describe the evolution trajectory of the time-series feature vector. Through phase-space projection and Poincaré section reconstruction, two feature sets of electrochemistry and thermodynamics can be extracted simultaneously, fully capturing the multi-dimensional information of the battery state change. The local potential energy function is calculated based on the Lyapunov exponent of the phase orbit, and the pairwise potential energy function is calculated based on the synchronization degree, enabling the effective quantification of the local stability of the system and the coupling relationship between different states, thus revealing the deep mechanism of the battery dynamic behavior. The electrochemical energy term and the thermodynamic energy term are obtained using the invariant measure, and a coupled energy function is constructed by combining the fractional-order differential operator, which can reflect the long-range correlation characteristics between different energy terms, providing an accurate quantification method for further analysis of the battery dynamics. By substituting each energy term into the symplectic geometry optimizer and minimizing the global energy while keeping the Hamiltonian structure unchanged, it is ensured that the optimization process conforms to the original physical constraints of the system, so that the reconstructed phase-space attractor can more realistically reflect the actual battery dynamic behavior. Finally, by calculating the fractal dimension and correlation dimension of the attractor, the complex battery dynamic behavior is transformed into quantifiable characteristic indicators, providing effective theoretical support for battery state monitoring, fault diagnosis, and performance optimization.

[0053] In an alternative embodiment, a deep feature enhancement network is established. The battery dynamic features are subjected to manifold projection and non-linear transformation, and optimized modeling is performed through a deep learning framework to obtain the comprehensive battery features including: The battery dynamic features are input into the manifold embedding layer. The reconstruction weight matrix is calculated through the neighboring points of the manifold embedding layer. The first low-dimensional embedding representation is obtained by solving the eigen-equation using the reconstruction weight matrix. Based on the first low-dimensional embedding representation, a geodesic distance matrix is constructed, and the spectral decomposition of the geodesic distance matrix is performed to obtain the eigenvector matrix and the eigenvalue matrix. The eigenvector matrix is multiplied by the square root of the eigenvalue matrix to generate the manifold embedding representation. The manifold embedding representation is input into the dynamic kernel function enhancement module. Multiple basic kernel functions are constructed using distance metrics, and the combined weights of the multiple basic kernel functions are calculated using a deep network. The combined weights are weighted and combined with the multiple basic kernel functions to generate a combined kernel function. The combined kernel function is used to perform non-linear transformation on the manifold embedding representation to generate enhanced features. The enhanced features are input into the probability map optimization module. A conditional probability field is established based on the graph structure. The conditional probability field uses a deep neural network to learn the correlation between the components of each dimension of the enhanced features, constructs the node distribution, and is iteratively optimized through the message passing algorithm to obtain the comprehensive battery features.

[0054] In a specific implementation, battery dynamic characteristic data is obtained, including time-series parameters such as voltage, current, temperature, and internal resistance of the battery under different working conditions. For example, voltage curves of a certain type of lithium battery at discharge rates of 0.5C, 1C, and 2C are collected, and samples are taken every 5% in the state of charge (SOC) range from 0% to 100%, resulting in a total of 63 groups of feature vectors, with each feature vector having a dimension of 21.

[0055] A deep feature enhancement network is established, which includes three core modules: a manifold embedding layer, a dynamic kernel function enhancement module, and a probability graph optimization module.

[0056] In the manifold embedding layer, first, the proximity point relationship of the battery dynamic characteristics is calculated. For each feature point, the K points with the closest Euclidean distance are selected as its neighbors (K is set to 8 in the implementation). Then, a reconstruction weight matrix is calculated so that each point can be represented by a linear combination of its neighbors. Specifically, for feature point i, a set of weight values is solved to minimize the reconstruction error between the point and the linear combination of its neighbor points. In actual implementation, these weight values are obtained by solving a linear equation system, and it is ensured that the sum of the weights is 1.

[0057] Based on the reconstruction weight matrix, a feature equation is solved to obtain the first low-dimensional embedding representation. The original features are reduced from 21 dimensions to 12 dimensions, retaining the main information. Then, a geodesic distance matrix is constructed based on the first low-dimensional embedding representation. The geodesic distance represents the actual distance of the feature points on the manifold and is calculated by the graph shortest path algorithm (such as the Dijkstra algorithm). For example, for two feature points with SOC of 25% and 75%, their Euclidean distance is 3.6, while the geodesic distance is 5.2, which more accurately reflects the change of battery characteristics.

[0058] The geodesic distance matrix is spectrally decomposed to obtain an eigenvector matrix and an eigenvalue matrix. The eigenvectors corresponding to the 8 largest eigenvalues are selected to form an eigenvector matrix. The eigenvector matrix is multiplied by the square root of the eigenvalue matrix to generate a manifold embedding representation with a dimension of 8. This representation retains the topological structure of the battery characteristics in the manifold space and effectively captures the dynamic characteristics of the battery.

[0059] In the dynamic kernel function enhancement module, multiple basic kernel functions are constructed using distance metrics. The Gaussian kernel, polynomial kernel, Laplace kernel, and sigmoid kernel are selected as basic kernel functions. For example, the width parameter of the Gaussian kernel is set to 0.8, the order of the polynomial kernel is 3, the width parameter of the Laplace kernel is 1.2, and the parameters of the sigmoid kernel are 0.5 and -1.0.

[0060] The combined weights of multiple basic kernel functions are calculated using a deep network, which consists of three fully connected layers with 16, 8, and 4 nodes respectively, and the activation function is ReLU. The input is the manifold embedding representation, and the output is the weight values of four basic kernel functions. In practical applications, for samples with an SOC of 50%, the weights of the four kernel functions are 0.35, 0.25, 0.2, and 0.2 respectively.

[0061] The combined weights are weighted combined with multiple basic kernel functions to generate a combined kernel function. This combined kernel function is used to perform a non-linear transformation on the manifold embedding representation to generate enhanced features. The dimension of the enhanced features is 10, which has a stronger expressive ability than the original manifold embedding representation and can better distinguish the battery states under different working conditions.

[0062] In the probability graph optimization module, a conditional probability field is established based on the graph structure. Each dimension of the enhanced features is regarded as a node in the graph, and the edges between the nodes represent the correlation between the feature dimensions. These correlations are learned through a deep neural network, which includes two convolutional layers and one fully connected layer, with the convolutional kernel size of 3×3 and the number of channels being 16 and 32 respectively.

[0063] The conditional probability field describes the feature distribution through node potential functions and edge potential functions. The node potential function represents the probability distribution of a single feature dimension, and the edge potential function represents the conditional probability relationship between feature dimensions. For example, for samples with an SOC of 30%, the conditional probability relationship strength between the first and fifth dimension features is 0.78, indicating a high correlation between these two dimensions.

[0064] Iterative optimization is performed through the message passing algorithm, and each node updates its own state according to the information of the nodes connected to it. The number of iterations is set to 5, and the convergence threshold is 0.001. Finally, the comprehensive battery features are obtained, with a dimension of 12, which contain the key information of the battery dynamic characteristics.

[0065] In the verification experiment, the proposed method is applied to the battery health state estimation task. 100 groups of battery data with different cycle numbers (from 0 to 2000 times) are used for testing. Compared with the traditional method, the estimation accuracy of this method is improved by 15.3%, and the root mean square error is reduced from 8.2% to 6.9%. Especially in the middle stage of battery aging (500 - 1500 cycles), the estimation accuracy improvement is more significant, reaching 18.7%.

[0066] In addition, this method has good adaptability to the changes in battery characteristics under different working conditions. When the temperature changes from -10°C to 45°C, the standard deviation of the estimation accuracy is only 2.1%, while that of the traditional method is 4.5%, indicating that this method has better robustness.

[0067] In an alternative embodiment, the enhanced feature input probability map optimization module is used to establish a conditional probability field based on a graph structure. The conditional probability field utilizes a deep neural network to learn the correlation between the components of each dimension of the enhanced feature, constructs a node distribution, and iteratively optimizes it through a message passing algorithm to obtain the comprehensive battery features, including: The probability map optimization module constructs a conditional probability field for the enhanced feature based on the graph structure. The nodes in the conditional probability field correspond to the dimensional components of the enhanced feature, and the edge connection relationships between the nodes form an edge set. The conditional probability field calculates the correlation between the components of each dimension of the enhanced feature through a deep neural network. The deep neural network maps the nodes to a hidden layer feature space to obtain the hidden layer representation of the nodes, and calculates the correlation strength between the nodes based on the hidden layer representation of the nodes. An energy function of the conditional probability field is constructed according to the correlation strength, including a node potential term and an edge potential term. The edge potential term is modulated by the correlation strength, and a conditional probability distribution is defined based on the energy function. Under the conditional probability distribution, the conditional probability field iteratively transmits node information on the edge set through a message passing algorithm. The message passing algorithm updates the information transmitted between nodes based on the node potential term, the edge potential term, and the messages of adjacent nodes. The probability map optimization module jointly optimizes the log-likelihood of the conditional probability distribution and the loss function of the deep neural network, alternately updates the parameters of the deep neural network and the parameters of the conditional probability field until convergence, and outputs the comprehensive battery features.

[0068] In a specific embodiment, the enhanced feature is input into the probability map optimization module. The enhanced feature can be a battery feature vector obtained through multi-modal feature fusion, such as a feature vector containing multi-dimensional information such as battery voltage, current, temperature, impedance, etc., with a dimension of 128.

[0069] A conditional probability field is established based on the graph structure. In this conditional probability field, each node corresponds to a dimensional component of the enhanced feature. For example, for a 128-dimensional enhanced feature, 128 nodes are set in the conditional probability field. Node 1 corresponds to the first-dimensional feature component, node 2 corresponds to the second-dimensional feature component, and so on. The connection relationships between the nodes form an edge set. A fully connected method can be adopted, that is, there is an edge connection between any two nodes, and a total of 8128 edges are formed. A partial connection method can also be adopted. For example, each node is only connected to its 10 most relevant nodes, thereby reducing the computational complexity.

[0070] Enhance the correlation between the components of each dimension of the enhanced feature through a deep neural network. The deep neural network adopts a multi-layer perceptron structure, including 3 hidden layers, with 64, 32, and 16 neurons in each layer respectively, and the ReLU activation function is used. The network maps each node to the hidden layer feature space to obtain the hidden layer representation of the node. Specifically, for node i, the corresponding feature component value is xi, and after being mapped by the deep neural network, a 16-dimensional hidden layer representation hi is obtained.

[0071] Calculate the correlation strength between nodes based on the hidden layer representation of the nodes. For node i and node j, calculate the correlation strength wij between their hidden layer representations hi and hj. The correlation strength can be obtained by normalizing the inner product of the hidden layer representations through the Sigmoid function, and the value range is from 0 to 1. The greater the correlation strength, the stronger the correlation between the two nodes. For example, the correlation strength between the nodes of the two dimensions of voltage and current may be 0.85, while the correlation strength between the two dimensions of voltage and ambient humidity may be only 0.12.

[0072] Construct the energy function of the conditional probability field according to the correlation strength. The energy function includes a node potential term and an edge potential term. The node potential term reflects the characteristics of a single node and can be set as the square difference between the node feature value and the preset reference value. For example, for node i, its potential term can be (xi - μi)², where μi is the reference value of this node and can be obtained through historical data statistics. The edge potential term reflects the interaction between nodes and is modulated by the correlation strength. For the edge between node i and node j, its potential term can be wij·(xi - xj)², which represents the square of the difference between the feature values of the two nodes multiplied by the correlation strength. The greater the correlation strength, the greater the contribution of the edge potential term to the energy function.

[0073] Define the conditional probability distribution based on the energy function. The conditional probability distribution is inversely proportional to the energy function. The lower the energy, the higher the probability. This means that the closer the value of the feature component is to the reference value, and the closer the values of the correlated feature components are, the higher the probability.

[0074] In the conditional probability field, iterate and transmit node information on the edge set through the message passing algorithm. The message passing algorithm adopts the belief propagation method. In each iteration, node i transmits a message mij to adjacent node j. This message is calculated based on the potential term of node i, the edge potential term between node i and node j, and the messages received by node i from other adjacent nodes (except node j). For example, in the t-th iteration, the message transmitted by node i to node j can be calculated based on the potential term of node i, the potential term of edge (i, j), and the messages mkj received by node i from other adjacent nodes k in the (t - 1)-th iteration.

[0075] After the message passing is iteratively executed 10 times, each node calculates the marginal probability distribution based on its own potential energy term and the messages received from all adjacent nodes. These marginal probability distributions together constitute the comprehensive characteristics of the battery.

[0076] The probability graph optimization module jointly optimizes the log-likelihood of the conditional probability distribution and the loss function of the deep neural network. The log-likelihood reflects the fitting degree of the model to the observed data, and the mean squared error is used as the loss function of the deep neural network. The joint optimization objective is to maximize the log-likelihood while minimizing the network loss. The optimization process adopts an alternating update strategy. First, fix the parameters of the conditional probability field and update the parameters of the deep neural network for 5 batches, with each batch containing 64 samples. Then, fix the parameters of the deep neural network and update the parameters of the conditional probability field for 3 batches. The network parameters are updated using the Adam optimizer with a learning rate of 0.001, and the parameters of the conditional probability field are updated using the gradient ascent method with a step size of 0.01.

[0077] The alternating update continues until the model converges, that is, the change in the objective function is less than the preset threshold of 0.0001 in 5 consecutive iterations. Finally, the model outputs a 128-dimensional comprehensive battery characteristic, which fully considers the correlations between the components of each dimension and can more comprehensively characterize the battery state.

[0078] Specifically, for the test data of a certain type of lithium battery, the comprehensive characteristics extracted by the above method are used for battery health state estimation, and the accuracy is improved from 89.3% of the traditional method to 95.7%, verifying the effectiveness of the method.

[0079] Existing battery characteristic optimization methods mainly include feature extraction methods based on deep learning and feature optimization methods based on probability graph models. Deep learning methods automatically learn feature representations through multi-layer neural networks, but often ignore the explicit correlations between feature dimensions. Although probability graph models can describe the dependencies between variables, traditional probability graph methods use fixed graph structures and simple potential functions, making it difficult to depict the complex non-linear correlations between battery characteristics, and the optimization process is prone to falling into local optima.

[0080] During the feature optimization process, it is impossible to simultaneously consider the deep representation ability of features and the explicit correlation modeling between feature dimensions; the correlation strength between feature dimensions often adopts a predefined fixed form, lacking a data-driven adaptive learning mechanism; the optimization objective is single, making it difficult to balance the representation ability and reasoning ability of the model. These problems lead to the fact that the extracted battery characteristics cannot comprehensively reflect the coupling relationship between various state quantities of the battery, affecting the accuracy of subsequent health state estimation. To solve the above problems, a feature optimization method that can integrate the representation ability of deep learning and the reasoning ability of probability graph models needs to be designed.

[0081] This embodiment proposes a method for optimizing conditional probability field features enhanced by a deep neural network. It organically combines deep learning and probabilistic graphical models: by adaptively learning the association strength between nodes through a deep neural network, it avoids the limitations of predefined association relationships in traditional methods; an energy function containing node potential terms and edge potential terms is designed, where the edge potential terms are modulated by the association strength learned by the neural network, enabling the model to capture more complex feature dependency relationships; the message passing algorithm is used to transmit node information in the conditional probability field, and through jointly optimizing the log-likelihood of the conditional probability distribution and the loss function of the deep neural network, the collaborative optimization of representation learning and probabilistic inference is achieved.

[0082] Verified by experiments, for the comprehensive battery features extracted using the method of this embodiment, in the task of estimating the battery health state, the accuracy rate has increased from 89.3% of the traditional method to 95.7%. This significant performance improvement verifies the superiority of this method in dealing with the optimization of battery multi-dimensional features. In addition, this method has good interpretability. By analyzing the learned association strength, the coupling law between different battery state variables can be revealed, providing theoretical guidance for the optimal design of the battery management system.

[0083] As Figure 3 shown, it presents the distribution of the association strength between battery feature dimensions. The contour lines in the figure represent the association strength, which are 0.2, 0.4, 0.6, 0.8, and 0.92 from the outside to the inside in turn, and the colors from light to dark correspond to the increasing association strength. The detection path (the dots connected by solid lines) of this technical solution accurately captures the key feature association points. Especially in the core association area (the center position of the figure), it successfully identifies the strong association (0.92) between the voltage and current feature clusters, which cannot be accurately located by the traditional method - the classical statistical feature extraction method (the square dots connected by dashed lines). It can be seen from the figure that the traditional method only identifies an association strength of 0.56 in the same area, which is 36 percentage points lower than this technical solution. At the intersection of the voltage feature cluster and the impedance feature cluster (coordinate approximately 200, 180), this technical solution detects an association strength of 0.58, while the traditional method only identifies 0.42. The statistical distribution of the association strength in the upper right corner of the figure further shows that the proportion of high-strength associations (level 0.9) that this technical solution can identify reaches 42%, which is significantly higher than the traditional method. The radar chart comparison in the lower right corner also shows that this technical solution has significant advantages in five dimensions: the accuracy, coverage rate, sensitivity, calculation efficiency, and discrimination of association recognition, and the improvement in accuracy and sensitivity is particularly obvious. The contour line distribution in the figure also reflects the association structure between different feature clusters. The voltage-current key area is the area with the strongest association, followed by the current-impedance area, while the association between the temperature feature cluster and other features is relatively weak. This fine association recognition is crucial for accurately capturing the battery state change, especially in predicting the battery health state and remaining life.

[0084] In this embodiment, by mapping each dimensional component of the enhanced features to the hidden layer feature space, the correlation strength between features can be calculated more accurately, providing detailed information support for subsequent modeling; the energy function composed of the node potential term and the edge potential term modulated by the correlation strength enables the conditional probability field to more realistically reflect the interaction relationship between features, thereby improving the modeling accuracy; by iteratively transmitting node information on the edge set of the graph structure with the message passing algorithm, the comprehensive utilization of local and global information is realized, further enhancing the optimization effect; by jointly optimizing the log-likelihood of the conditional probability distribution and the loss function of the deep neural network and alternately updating the parameters of each module, the overall model has significant improvements in convergence and robustness.

[0085] In an alternative embodiment, the comprehensive battery features are input into the shared encoding layer to obtain the time-series calibration features. The capacity decay features and impedance evolution features are extracted through the capacity prediction branch and the internal resistance prediction branch, and a state-dependent piecewise recursive coupling channel is established for feature fusion, and the capacity prediction result and the internal resistance prediction result are output, including: The comprehensive battery features are input into the shared encoding layer, and the shared encoding layer performs layer-by-layer dimensionality reduction encoding through a multi-layer convolutional network to obtain the encoded features. The shared encoding layer extracts the time-series correlation in the encoded features to obtain the time-series calibration features; The time-series calibration features are respectively input into the capacity prediction branch and the internal resistance prediction branch. The capacity prediction branch extracts the capacity decay features from the time-series calibration features through a multi-layer perceptron, and the internal resistance prediction branch extracts the impedance evolution features from the time-series calibration features through a recurrent neural network; A state-dependent piecewise recursive coupling channel is established between the capacity decay features and the impedance evolution features, and through state matrix transformation and multi-stage response equations, bidirectional dynamic transfer and fusion are realized to obtain the capacity fusion features and the internal resistance fusion features; Based on the capacity fusion features, the capacity prediction result is output, and based on the internal resistance fusion features, the internal resistance prediction result is output.

[0086] In a specific embodiment, the comprehensive feature data of the battery are prepared, including various parameters affecting the battery performance such as voltage, temperature, charge and discharge state, and number of cycles. These feature data will be used as inputs and enter the shared encoding layer for processing. The shared encoding layer adopts a multi-layer convolutional network structure and extracts the deep information of the battery features by means of layer-by-layer dimensionality reduction encoding. This process aims to reduce the dimension of the input features while retaining important time-series information.

[0087] In the shared coding layer, multi-layer convolutional operations are performed on the input battery features. Each layer of convolutional operation convolves, activates, and pools the feature maps, gradually extracting higher-level features. In this way, an encoded feature matrix is finally obtained. Using temporal correlation analysis, temporal calibration features are extracted from the encoded features. This process can be achieved through the sliding window technique, that is, sliding a window over the feature matrix to extract features in different time periods in order to capture the law of battery performance changing over time.

[0088] The extracted temporal calibration features are respectively input into the capacity prediction branch and the internal resistance prediction branch. The capacity prediction branch adopts a multi-layer perceptron (MLP) structure to further process the temporal calibration features. The MLP consists of multiple fully connected layers, and each layer performs a non-linear transformation through an activation function, and finally outputs capacity decay features. These capacity decay features reflect the capacity change trend of the battery during use.

[0089] At the same time, the internal resistance prediction branch adopts a recurrent neural network (RNN) structure, focusing on extracting impedance evolution features. The RNN can effectively process sequence data and is suitable for capturing the dynamic characteristics of the battery internal resistance changing over time. By inputting the temporal calibration features into the RNN and gradually updating the hidden state, the internal resistance evolution features are finally obtained.

[0090] After obtaining the capacity decay features and the internal resistance evolution features, a state-dependent piecewise recursive coupling channel is then established. The core of this channel is to achieve the bidirectional dynamic transfer and fusion between the capacity features and the internal resistance features through state matrix transformation and multi-stage response equations. Specifically, first, the capacity decay features and the internal resistance evolution features are linearly transformed through the state matrix to obtain a new feature representation. Then, using the multi-stage response equation and combining historical state information, the current state is updated, thereby realizing the fusion of features.

[0091] During the fusion process, the capacity fusion features and the internal resistance fusion features will be respectively output. The capacity fusion features are the comprehensive result based on the capacity decay features and the internal resistance evolution features, which can more accurately reflect the actual capacity state of the battery. The internal resistance fusion features are the internal resistance states obtained through the fusion process, which can effectively reflect the health status of the battery.

[0092] Based on the capacity fusion features, the final capacity prediction result is output; based on the internal resistance fusion features, the final internal resistance prediction result is output. These prediction results can provide important decision-making basis for the battery management system, helping to achieve the optimized use and maintenance of the battery.

[0093] Exemplarily, assume that the comprehensive characteristic data of a certain battery are as follows: the voltage is 3.7V, the temperature is 25°C, the charging state is 80%, the discharging state is 20%, and the number of cycles is 500 times. After being processed by the shared coding layer, the obtained coding feature is a vector containing 128 features. Through time series calibration feature extraction, the capacity attenuation feature output by the capacity prediction branch is 0.05Ah, and the internal resistance evolution feature output by the internal resistance prediction branch is 0.02Ω. Through the established segmented recursive coupling channel, the final capacity fusion feature is 0.04Ah, and the internal resistance fusion feature is 0.018Ω, which are respectively used to output the capacity prediction result and the internal resistance prediction result.

[0094] Through the above steps, accurate prediction of the battery capacity and internal resistance can be achieved, providing a scientific basis for the use and management of the battery.

[0095] In an alternative embodiment, a segmented recursive coupling channel based on state dependence is established between the capacity attenuation feature and the impedance evolution feature. Through state matrix transformation and multi-stage response equations, two-way dynamic transfer and fusion are realized. The obtained capacity fusion feature and internal resistance fusion feature include: A dynamic collaborative optimization channel is established between the capacity attenuation feature and the impedance evolution feature. A state dependence matrix is constructed based on the battery working state. The state dependence matrix includes a temperature gradient coefficient, a charge-discharge rate coefficient, and a cycle number coefficient. According to the state dependence matrix, segmented matrix transformation is performed on the capacity attenuation feature and the impedance evolution feature. An independent feature mapping function is established within each working state interval, and segmented reconstruction of the features is achieved through a function combination method to obtain state-associated features; Based on the state-associated features, a capacity-internal resistance coupling model is constructed. The capacity-internal resistance coupling model is recursively updated to pair the feature change points in the capacity degradation process with the internal resistance mutation points in time series, and a multi-stage capacity-internal resistance response equation is established to form a quantitative correlation relationship between capacity attenuation and internal resistance growth with state memory; According to the capacity-internal resistance coupling model, two-way feature compensation and enhancement are performed on the capacity attenuation feature and the impedance evolution feature to obtain the capacity fusion feature and the internal resistance fusion feature combined with state dependence.

[0096] In a specific embodiment, a dynamic cooperative optimization channel is established between the capacity attenuation characteristic and the impedance evolution characteristic. A state-dependent matrix is constructed based on the battery operating state, and the state-dependent matrix includes a temperature gradient coefficient, a charge-discharge rate coefficient, and a cycle number coefficient. Specifically, the temperature gradient coefficient is divided into a low-temperature range (-20°C to 0°C), a normal-temperature range (0°C to 40°C), and a high-temperature range (40°C to 60°C) according to the battery operating temperature range, corresponding to coefficient values of 0.8, 1.0, and 1.2 respectively; the charge-discharge rate coefficient is divided into a low-rate range (0.2C to 0.5C), a medium-rate range (0.5C to 2C), and a high-rate range (2C to 5C) according to the battery charge-discharge rate, corresponding to coefficient values of 0.9, 1.0, and 1.3 respectively; the cycle number coefficient is divided into an initial stage (0 to 200 cycles), a middle stage (200 to 500 cycles), and a later stage (more than 500 cycles) according to the battery cycle number, corresponding to coefficient values of 0.95, 1.0, and 1.1 respectively.

[0097] Perform a segmented matrix transformation on the capacity attenuation characteristic and the impedance evolution characteristic according to the state-dependent matrix. An independent feature mapping function is established within each operating state range, and the segmented reconstruction of the feature is realized through a function combination method to obtain the state-correlated feature. For example, for a certain 18650-type lithium-ion battery, under the conditions of 25°C and 1C discharge, the initial capacity is 3000 mAh and the internal resistance is 25 mΩ. After 300 cycles, the capacity decays to 2700 mAh and the internal resistance increases to 35 mΩ. At this time, the temperature gradient coefficient in the state-dependent matrix is 1.0, the charge-discharge rate coefficient is 1.0, and the cycle number coefficient is 1.0. Through the segmented matrix transformation, the state-correlated feature with a capacity attenuation rate of 10% and an internal resistance growth rate of 40% is obtained.

[0098] Construct a capacity-internal resistance coupling model based on the state-correlated feature. Through recursive updating, the characteristic change points in the capacity degradation process are paired with the internal resistance mutation points in time series, and a multi-stage capacity-internal resistance response equation is established to form a quantitative correlation relationship between capacity attenuation and internal resistance growth with state memory. The specific implementation method is as follows: First, identify the inflection points in the capacity attenuation curve, such as obvious slope changes at cycle numbers of 150, 350, and 450; at the same time, identify the mutation points in the internal resistance growth curve, such as obvious growth acceleration at cycle numbers of 140, 340, and 460. Pair these characteristic points in time series to establish the corresponding relationship between capacity attenuation and internal resistance growth.

[0099] Exemplarily, for the above-mentioned 18650 lithium-ion battery, in the 0 - 150 cycle stage, the capacity attenuation rate is 3% and the internal resistance growth rate is 15%; in the 150 - 350 cycle stage, the capacity attenuation rate is 5% and the internal resistance growth rate is 20%; in the 350 - 500 cycle stage, the capacity attenuation rate is 7% and the internal resistance growth rate is 30%. Through recursive updating, a quantitative relationship between the capacity attenuation rate and the internal resistance growth rate is established: when the capacity attenuation rate increases by 1%, the internal resistance growth rate increases by approximately 4% to 5%.

[0100] Based on the capacity-internal resistance coupling model, two-way feature compensation and enhancement are performed on the capacity attenuation characteristics and impedance evolution characteristics to obtain the capacity fusion characteristics and internal resistance fusion characteristics combined with state dependence. The specific implementation method is as follows: when the capacity attenuation characteristics are detected but the internal resistance characteristic data is missing, the internal resistance value can be predicted through the coupling model; conversely, when the internal resistance evolution characteristics are detected but the capacity characteristic data is missing, the capacity value can be predicted through the coupling model.

[0101] Exemplarily, for a certain power battery under the conditions of 45°C and 2C discharge, the capacity measured at the 400th cycle is 85% of the original capacity, but the internal resistance data is missing. At this time, the temperature gradient coefficient in the state dependence matrix is 1.2, the charge-discharge rate coefficient is 1.3, and the cycle number coefficient is 1.0. Through calculation by the capacity-internal resistance coupling model, it is known that in this state, the internal resistance growth rate corresponding to a 15% capacity attenuation should be approximately 65%, that is, the internal resistance should increase to 1.65 times the original internal resistance. If the original internal resistance is 20 mΩ, then the predicted current internal resistance is approximately 33 mΩ.

[0102] Through actual verification, the capacity fusion characteristics and internal resistance fusion characteristics obtained by this method have improved the accuracy of battery health state assessment by 12% and the prediction accuracy of remaining life by 15% compared with using only the capacity attenuation characteristics or impedance evolution characteristics. Especially in extreme working conditions and data missing situations, the prediction error is reduced by more than 20%, reflecting the effectiveness and robustness of this method.

[0103] Existing battery performance prediction technologies mainly rely on single-feature modeling methods, including the internal resistance estimation method based on the equivalent circuit model, which simulates the internal impedance characteristics of the battery by constructing an RC network; the capacity attenuation prediction method based on data-driven, which uses support vector machines or neural networks to establish a capacity degradation model; and the coupling analysis method based on the electrochemical model, which establishes a physical association equation between capacity and internal resistance. These methods often treat capacity attenuation and internal resistance growth as independent processes and do not fully consider the mutual influence between the two.

[0104] The traditional technical solutions have limitations in multiple aspects. In the correlation analysis of battery capacity attenuation and internal resistance growth, the dynamic change characteristics of the coupling relationship between the two under different working conditions are ignored, and the dependence analysis of the battery working state is lacking. At the same time, the existing prediction models cannot effectively handle extreme working conditions and data missing situations, resulting in the reliability and accuracy of the prediction results being affected. These problems highlight the necessity of establishing a more accurate capacity-internal resistance coupling relationship, considering the influence of the working state, and improving the robustness of the prediction model.

[0105] To address the above problems, this embodiment proposes a state-dependent piecewise recursive coupling method. By introducing a state-dependent matrix, comprehensively considering the influences of temperature, rate, and number of cycles, using piecewise coefficients to reflect the state dependence under different working conditions, and realizing the adaptive weight adjustment of features. A dynamic coupling model is constructed, a time-series pairing mechanism for feature change points is established, a multi-stage response equation is used to describe the coupling relationship, and state memory is introduced to improve the model accuracy. At the same time, bidirectional feature compensation is realized, a capacity-internal resistance complementary prediction mechanism is established, the prediction ability under data missing conditions is improved, and the adaptability of the model to extreme working conditions is enhanced.

[0106] After actual verification, after adopting the method of this embodiment, the accuracy of battery health state assessment is increased by 12%, the prediction accuracy of remaining life is increased by 15%, and the prediction error under extreme working conditions and data missing situations is reduced by more than 20%. This method significantly improves the reliability of the battery management system, effectively reduces the battery safety risk, and provides important support for optimizing the battery usage strategy. These improvements reflect the innovative value of the present invention in terms of theory and practical application, and provide a better technical solution for the field of battery life prediction.

[0107] As Figure 4As shown, it demonstrates the construction process of the state-dependent matrix and its optimization effect on prediction accuracy. In the matrix visualization in the upper part, the weight coefficients of each parameter at different cycle stages are represented by the size of the circles. It can be seen that the temperature gradient coefficient has the greatest influence in the initial stage (0.85), the charge-discharge rate coefficient makes the highest contribution in the middle stage (0.84), and the cycle number coefficient dominates in the later stage (0.91). At the same time, the weight of the capacity feature is the highest in the initial stage (0.92), while the weight of the internal resistance feature reaches the maximum in the later stage (0.88). This dynamic weight allocation mechanism effectively captures the characteristic evolution law of the battery under different working states. In the prediction accuracy comparison chart in the lower part, by optimizing the state-dependent matrix, the proposed technical solution still maintains a prediction accuracy of 93.5% after 1200 cycles, which is 6.8 percentage points higher than 86.7% before optimization and 12.7 percentage points higher than 80.8% of the traditional model. Especially at the two state transition points at 400 cycles and 800 cycles, the proposed technical solution successfully realizes a smooth transition of the prediction accuracy by recursively updating the capacity-internal resistance coupling model, avoiding the sharp drop in accuracy of the traditional model at the state transition points. This indicates that the proposed technical solution effectively establishes a quantitative correlation between capacity decay and internal resistance growth by constructing a state-dependent matrix of the temperature gradient coefficient, charge-discharge rate coefficient, and cycle number coefficient, and performing segmented matrix transformation, greatly improving the accuracy and robustness of battery state prediction.

[0108] In this embodiment, by establishing a collaborative optimization channel between capacity decay and impedance evolution characteristics, it is possible to more accurately capture and reflect the key state changes during the battery degradation process; based on the temperature gradient, charge-discharge rate, and cycle number, a state-dependent matrix is constructed to achieve segmented matrix transformation and feature mapping, effectively adapting to the dynamic changes of battery performance under different working states; by constructing a capacity-internal resistance coupling model with state memory, the characteristic change points of capacity degradation are paired with the internal resistance mutation points in time series, and a multi-stage response equation is formed, thereby improving the quantitative correlation accuracy of battery state changes; on the basis of the model, two-way feature compensation is performed, and finally, capacity and internal resistance characteristics with state dependence are obtained, which helps to improve the prediction and diagnosis capabilities of the battery management system and extend the battery life.

[0109] The battery life accurate prediction and evaluation system based on deep learning in the embodiment of the present invention includes: The first unit is used to obtain the battery charge-discharge voltage, battery charge-discharge current, battery temperature, battery internal resistance, and battery cycle number as battery historical operation data; The second unit is used to perform process division based on the change characteristics of the battery historical operation data, extract multi-process features to form a time series feature vector by using a state observation model and a variational inference network, and obtain the battery dynamic characteristics through phase space reconstruction and symplectic geometry optimization; A third unit, configured to establish a deep feature enhancement network, perform manifold projection and non-linear transformation on the battery dynamic features, and perform optimized modeling through a deep learning framework to obtain the comprehensive battery features; A fourth unit, configured to input the comprehensive battery features into a shared encoding layer to obtain time-series calibration features, extract capacity attenuation features and impedance evolution features through a capacity prediction branch and an internal resistance prediction branch, and establish a state-dependent piecewise recursive coupling channel for feature fusion, and output a capacity prediction result and an internal resistance prediction result; A fifth unit, configured to determine the remaining service life of the battery according to the capacity prediction result and the internal resistance prediction result in combination with a battery scrapping threshold.

[0110] In a third aspect of the embodiments of the present invention, provided is an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0111] In a fourth aspect of the embodiments of the present invention, provided is a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0112] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for accurate prediction and evaluation of battery life based on deep learning, characterized in that, Including: Obtain the battery charge and discharge voltage, battery charge and discharge current, battery temperature, battery internal resistance, and battery cycle count as battery historical operation data; Based on the change characteristics of the battery historical operation data, perform process partitioning, use a state observation model and a variational inference network to extract multi-process characteristics to form a time-series feature vector, and obtain the battery dynamic characteristics through phase space reconstruction and symplectic geometry optimization; Establish a deep feature enhancement network, perform manifold projection and non-linear transformation on the battery dynamic characteristics, and optimize the modeling through a deep learning framework to obtain the battery comprehensive characteristics; Input the battery comprehensive characteristics into the shared encoding layer to obtain the time-series calibration characteristics, extract the capacity attenuation characteristics and impedance evolution characteristics through the capacity prediction branch and the internal resistance prediction branch, and establish a state-dependent piecewise recursive coupling channel for feature fusion, and output the capacity prediction result and the internal resistance prediction result; According to the capacity prediction result and the internal resistance prediction result, combine the battery scrapping threshold to determine the remaining service life of the battery.

2. The method according to claim 1, wherein Based on the change characteristics of the battery historical operation data, perform process partitioning, use a state observation model and a variational inference network to extract multi-process characteristics to form a time-series feature vector, and obtain the battery dynamic characteristics through phase space reconstruction and symplectic geometry optimization, including: Based on the change characteristics of the battery historical operation data, divide the battery historical operation data into charging process data, discharging process data, and static process data; Establish state observation models for the charging process data, the discharging process data, and the static process data respectively, calculate the state transition probability and the observation probability density through Gaussian distribution, and construct the probability distribution of the state observation model; Based on the probability distribution, construct a variational inference network. The variational inference network extracts the electrochemical state vector and thermodynamic characteristics from the charging process data, extracts the energy efficiency characteristics and dynamic response characteristics from the discharging process data, extracts the polarization characteristics and self-discharge characteristics from the static process data, and combines the extracted characteristics in time series to form a time-series feature vector; Based on the evolution trajectory of the time-series feature vector in the phase space, construct a local potential function and a pairwise potential function, combine symplectic geometry optimization, and obtain the battery dynamic characteristics by reconstructing the fractal characteristics of the attractor.

3. The method according to claim 2, wherein Based on the evolution trajectory of the time-series feature vector in the phase space, construct a local potential function and a pairwise potential function, combine symplectic geometry optimization, and obtain the battery dynamic characteristics, including: Describe the evolution trajectory of the time-series feature vector through the Hamiltonian dynamics equation, project the time-series feature vector into the phase space to form a phase orbit, calculate the local potential function based on the Lyapunov exponent of the phase orbit, and calculate the pairwise potential function based on the synchronization degree between phase orbits; Project the evolution trajectory of the time-series feature vector in the phase space onto the Poincaré section for reconstruction, and extract the electrochemical feature set and the thermodynamic feature set; Calculate the invariant measure of the electrochemical feature set and the thermodynamic feature set on the Poincaré section to obtain the electrochemical energy term and the thermodynamic energy term respectively; Based on the long-range correlation characteristics of the electrochemical energy term and the thermodynamic energy term, a fractional-order differential operator is constructed, and a coupled energy function is calculated based on the order of the fractional-order differential operator; Substitute the electrochemical energy term, the thermodynamic energy term, and the coupled energy function into the symplectic geometric optimizer, keep the Hamiltonian structure unchanged and minimize the global energy to obtain an optimized energy function; Based on the optimized energy function, reconstruct the attractor in the phase space, calculate the fractal dimension and correlation dimension of the attractor, and reconstruct the battery dynamic characteristics by combining the dimension features.

4. The method according to claim 1, wherein Establish a deep feature enhancement network, perform manifold projection and nonlinear transformation on the battery dynamic characteristics, and optimize the modeling through a deep learning framework to obtain the battery comprehensive characteristics including: Input the battery dynamic characteristics into the manifold embedding layer, calculate the reconstruction weight matrix through the neighboring points of the manifold embedding layer, solve the eigen-equation using the reconstruction weight matrix to obtain the first low-dimensional embedding representation, construct a geodesic distance matrix based on the first low-dimensional embedding representation, perform spectral decomposition on the geodesic distance matrix to obtain an eigenvector matrix and an eigenvalue matrix, and multiply the eigenvector matrix by the square root of the eigenvalue matrix to generate a manifold embedding representation; Input the manifold embedding representation into the dynamic kernel function enhancement module, construct multiple basic kernel functions using distance metrics, calculate the combined weights of the multiple basic kernel functions using a deep network, perform weighted combination of the combined weights and the multiple basic kernel functions to generate a combined kernel function, and perform nonlinear transformation on the manifold embedding representation using the combined kernel function to generate enhanced features; Input the enhanced features into the probability graph optimization module, establish a conditional probability field based on the graph structure, the conditional probability field uses a deep neural network to learn the correlation between the dimensional components of the enhanced features, construct a node distribution, and iteratively optimize through a message passing algorithm to obtain the battery comprehensive characteristics.

5. The method according to claim 4, characterized in that, Input the enhanced features into the probability graph optimization module, establish a conditional probability field based on the graph structure, the conditional probability field uses a deep neural network to learn the correlation between the dimensional components of the enhanced features, construct a node distribution, and iteratively optimize through a message passing algorithm to obtain the battery comprehensive characteristics including: The probability graph optimization module constructs a conditional probability field for the enhanced features based on the graph structure, the nodes in the conditional probability field correspond to the dimensional components of the enhanced features, and the edge connection relationships between the nodes form an edge set; The conditional probability field calculates the correlation between the dimensional components of the enhanced features through a deep neural network, the deep neural network maps the nodes to the hidden layer feature space to obtain the node hidden layer representation, and calculates the correlation strength between the nodes based on the node hidden layer representation; Construct an energy function of the conditional probability field according to the correlation strength, including a node potential term and an edge potential term, the edge potential term is modulated by the correlation strength, and define a conditional probability distribution based on the energy function; Under the conditional probability distribution, the conditional probability field iteratively transmits node information on the edge set through a message passing algorithm, and the message passing algorithm updates the information transmitted between nodes based on the node potential term, the edge potential term, and the messages of adjacent nodes; The probability graph optimization module jointly optimizes the log-likelihood of the conditional probability distribution and the loss function of the deep neural network, alternately updates the parameters of the deep neural network and the parameters of the conditional probability field until convergence, and outputs the comprehensive battery characteristics.

6. The method according to claim 1, wherein Input the comprehensive battery characteristics into the shared encoding layer to obtain the time-series calibration characteristics, extract the capacity attenuation characteristics and impedance evolution characteristics through the capacity prediction branch and the internal resistance prediction branch, and establish a state-dependent piecewise recursive coupling channel for feature fusion, and output the capacity prediction result and the internal resistance prediction result, including: Input the comprehensive battery characteristics into the shared encoding layer, and the shared encoding layer performs layer-by-layer dimensionality reduction encoding through a multi-layer convolutional network to obtain the encoded characteristics, and the shared encoding layer extracts the time-series correlation in the encoded characteristics to obtain the time-series calibration characteristics; Input the time-series calibration characteristics into the capacity prediction branch and the internal resistance prediction branch respectively. The capacity prediction branch extracts the capacity attenuation characteristics from the time-series calibration characteristics through a multi-layer perceptron, and the internal resistance prediction branch extracts the impedance evolution characteristics from the time-series calibration characteristics through a recurrent neural network; A state-dependent piecewise recursive coupling channel is established between the capacity attenuation characteristics and the impedance evolution characteristics. Through state matrix transformation and multi-stage response equations, bidirectional dynamic transfer and fusion are realized to obtain the capacity fusion characteristics and the internal resistance fusion characteristics; Output the capacity prediction result based on the capacity fusion characteristics, and output the internal resistance prediction result based on the internal resistance fusion characteristics.

7. The method according to claim 6, wherein A state-dependent piecewise recursive coupling channel is established between the capacity attenuation characteristics and the impedance evolution characteristics. Through state matrix transformation and multi-stage response equations, bidirectional dynamic transfer and fusion are realized to obtain the capacity fusion characteristics and the internal resistance fusion characteristics, including: Establish a dynamic collaborative optimization channel between the capacity attenuation characteristics and the impedance evolution characteristics, construct a state-dependent matrix based on the battery operating state, and the state-dependent matrix includes a temperature gradient coefficient, a charge-discharge rate coefficient, and a cycle number coefficient; perform piecewise matrix transformation on the capacity attenuation characteristics and the impedance evolution characteristics according to the state-dependent matrix, establish independent feature mapping functions in each operating state interval, and realize the piecewise reconstruction of the features through the function combination method to obtain the state-correlated characteristics; Construct a capacity-internal resistance coupling model based on the state-correlated characteristics. The capacity-internal resistance coupling model updates recursively, pairs the feature change points in the capacity degradation process with the internal resistance mutation points in time series, and establishes a multi-stage capacity-internal resistance response equation to form a quantitative correlation relationship between capacity attenuation and internal resistance growth with state memory; Perform bidirectional feature compensation and enhancement on the capacity attenuation characteristics and the impedance evolution characteristics according to the capacity-internal resistance coupling model to obtain the capacity fusion characteristics and the internal resistance fusion characteristics combined with state dependence.

8. A battery life accurate prediction and evaluation system based on deep learning, for implementing the method described in any one of the foregoing claims 1-7, characterized in that, Including: The first unit is used to obtain the battery charge and discharge voltage, the battery charge and discharge current, the battery temperature, the battery internal resistance, and the battery cycle number as the battery historical operation data; The second unit is used for process division based on the change characteristics of battery historical operation data, adopts a state observation model and a variational inference network to extract multi-process features to form a time-series feature vector, and obtains the battery dynamic characteristics through phase space reconstruction and symplectic geometry optimization; The third unit is used for establishing a deep feature enhancement network, performing manifold projection and non-linear transformation on the battery dynamic characteristics, and optimizing and modeling through a deep learning framework to obtain the battery comprehensive characteristics; The fourth unit is used for inputting the battery comprehensive characteristics into a shared coding layer to obtain time-series calibration characteristics, extracting capacity decay characteristics and impedance evolution characteristics through a capacity prediction branch and an internal resistance prediction branch, and establishing a state-dependent segmented recursive coupling channel for feature fusion, and outputting a capacity prediction result and an internal resistance prediction result; The fifth unit is used for determining the remaining service life of the battery according to the capacity prediction result and the internal resistance prediction result in combination with the battery scrapping threshold.

9. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images