Intelligent monitoring system and method for new energy automobile battery

Abstract

The invention relates to the technical field of battery intelligent monitoring, and discloses an intelligent monitoring system and method for a new energy automobile battery. The method comprises the following steps: acquiring a target battery operation data set of a new energy automobile battery, and extracting time domain differential characteristics and frequency domain energy loss characteristics in a battery charging and discharging process to obtain a multi-dimensional state characteristic set; performing electrochemical characteristic analysis on the new energy automobile battery based on the multi-dimensional state feature set to obtain a micro degradation state judgment result; and dynamically adjusting the parameter configuration of a hybrid Kalman filter according to the micro degradation state judgment result, and generating a battery state-of-charge estimation value. According to the method, the problems of insufficient estimation precision and accumulative errors in a traditional method are solved, and the use safety and reliability of the battery are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery intelligent monitoring, and in particular to an intelligent monitoring system and method for new energy vehicle batteries. Background Art

[0002] Existing battery monitoring technologies generally suffer from insufficient data acquisition accuracy and low sampling frequency, making it difficult to capture the transient changes in the electrochemical reactions within the battery, resulting in a significant lag in determining the battery's operating status. Furthermore, traditional battery state estimation methods overly rely on single feature extraction and simplified models, such as the first-order RC equivalent circuit model, which cannot fully characterize the dynamic response characteristics of the battery under complex operating conditions. This is particularly true under extreme conditions such as high and low temperatures, and high current charging and discharging. Estimation accuracy is severely reduced, and even misjudgments may occur.

[0003] Current mainstream battery health monitoring technologies lack in-depth analysis of microscopic degradation mechanisms, focusing solely on the surface phenomena of macroscopic parameter changes. They are unable to distinguish the impact of different degradation mechanisms, such as SEI film growth, lithium precipitation, and active material loss, on battery performance, making it difficult to implement targeted protection strategies. Traditional Kalman filtering algorithms, which employ fixed process noise covariance matrices and measurement noise covariance matrices, are unable to adapt to the dynamic characteristics of battery parameters as they change with operating conditions, resulting in large fluctuations in estimation accuracy. Furthermore, in-vehicle environments, abnormalities such as sensor failure and communication interruptions are frequent. Existing monitoring systems lack effective emergency response mechanisms, making it impossible to guarantee the reliability and safety of battery management systems. Summary of the Invention

[0004] The present invention provides an intelligent monitoring system and method for new energy vehicle batteries, which solves the problems of insufficient estimation accuracy and cumulative error in traditional methods and improves the safety and reliability of battery use.

[0005] In a first aspect, the present invention provides an intelligent monitoring method for a new energy vehicle battery, the intelligent monitoring method for a new energy vehicle battery comprising: Obtain a target battery operation dataset for new energy vehicle batteries, extract the time-domain differential features and frequency-domain energy loss features during the battery charging and discharging process, and obtain a multidimensional state feature set; Performing electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; Dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

[0006] Optionally, in a first implementation of the first aspect of the present invention, obtaining a target battery operation data set of a new energy vehicle battery and extracting time-domain differential features and frequency-domain energy loss features during the battery charging and discharging process to obtain a multidimensional state feature set includes: Collect the single cell voltage, charge and discharge current, surface temperature distribution and ambient temperature of new energy vehicle batteries to obtain the original battery parameter data; Performing denoising processing on the original battery parameter data to obtain battery signal data from which sampling noise has been removed, and performing high-frequency interference processing on the battery signal data from which sampling noise has been removed to obtain battery signal data from which high-frequency interference has been eliminated; Reconstructing missing points on the battery signal data from which high-frequency interference has been eliminated to obtain continuous and complete battery signal data, and performing time synchronization and standardization processing on the continuous and complete battery signal data to obtain a target battery operation data set; The time domain differential features and frequency domain energy loss features of the battery during charging and discharging are extracted according to the target battery operation data set to obtain a multi-dimensional state feature set.

[0007] Optionally, in a second implementation of the first aspect of the present invention, extracting time-domain differential features and frequency-domain energy loss features during the battery charging and discharging process according to the target battery operation data set to obtain a multidimensional state feature set includes: performing differential calculation on the voltage and current data in the target battery operation data set to obtain voltage-capacity differential data; Performing grid discretization processing on the voltage-capacity differential data to obtain a differential characteristic matrix; Using the Shannon entropy calculation function to quantify the characteristic distribution of the differential characteristic matrix to obtain a time domain characteristic vector; Performing Laplace transform on the current signal in the target battery operation data set to obtain current spectrum data, and calculating the impedance value of the battery equivalent internal resistance at a preset characteristic frequency point based on the current spectrum data to obtain a frequency domain eigenvector; The time domain feature vector and the frequency domain feature vector are fused and tensor decomposition is performed to obtain a multi-dimensional state feature set.

[0008] Optionally, in a third implementation of the first aspect of the present invention, performing electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result includes: Constructing initial values ​​of battery polarization characteristic parameters including open circuit voltage, ohmic internal resistance, and two RC parallel networks based on the multidimensional state feature set; Setting a prediction error function for the initial value of the battery polarization characteristic parameter and performing iterative calculations using a recursive least squares algorithm to obtain a parameter vector iterative update formula; Generate a parameter correction coefficient matrix based on the parameter vector iterative update formula combined with the current battery state of charge and temperature value to obtain battery equivalent circuit parameters that are adjusted in real time according to the operating conditions; Based on the battery equivalent circuit parameters adjusted in real time according to the operating conditions, a state vector equation including the state of charge and polarization voltage is established to obtain a battery state space model structure; Setting calculation rules of the state prediction equation and the system output equation according to the battery state space model structure to obtain a battery dynamic equivalent model; The battery dynamic equivalent model is used to analyze the electrochemical characteristics of the new energy vehicle battery to obtain a microscopic degradation state determination result.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, performing electrochemical characteristic analysis on the new energy vehicle battery using the battery dynamic equivalent model to obtain a microscopic degradation state determination result includes: Applying a small AC excitation signal at a preset logarithmically divided frequency point based on the battery dynamic equivalent model to obtain battery voltage response time domain data; Performing a complex domain convolution operation on the battery voltage response time domain data to obtain frequency domain electrochemical impedance spectroscopy data; constructing an impedance characteristic vector based on the frequency-domain electrochemical impedance spectroscopy data and calculating the difference from a standard impedance spectrum of a healthy battery to obtain an impedance difference characteristic vector; Constructing a battery degradation feature dataset based on the impedance difference feature vector and a circuit parameter time series in the battery dynamic equivalent model; Input the battery degradation feature dataset into the bidirectional LSTM layer and the multi-head self-attention mechanism to perform probability distribution calculation to obtain the degradation mechanism probability distribution; According to the probability distribution of the degradation mechanism, the microscopic degradation state judgment results of six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion and electrode structure collapse are determined.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, determining the microscopic degradation state determination results of six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse according to the degradation mechanism probability distribution includes: Performing threshold screening based on the degradation mechanism probability distribution to obtain a set of candidate dominant degradation mechanisms, wherein the candidate dominant degradation mechanism set includes six degradation mechanisms: SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse; Calculating the relative contribution ratio of each dominant degradation mechanism based on the set of candidate dominant degradation mechanisms and sorting them to obtain a degradation mechanism sequence sorted by importance; assigning a life decay model in the form of an exponential-logarithmic composite function, a quadratic function, or an exponential function to the degradation mechanism ranked first in the sequence of degradation mechanisms sorted by importance; The lifetime decay model is used to perform parameter fitting calculations to obtain degradation trend prediction parameters, and the remaining number of cycles and service life are calculated based on the degradation trend prediction parameters and the current battery state to obtain a microscopic degradation state determination result.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, dynamically adjusting the parameter configuration of the hybrid Kalman filter and generating the battery state of charge estimate according to the microscopic degradation state determination result includes: Constructing a prediction and update equation group of a hybrid Kalman filter based on the state space equations in the battery dynamic equivalent model; Creating a variable process noise covariance matrix update rule according to the probability values ​​of different degradation mechanisms in the microscopic degradation state determination result; Perform sliding window analysis on the measurement residual sequence generated during the filtering process and calculate the residual covariance to obtain the measurement noise adaptive adjustment matrix; Executing prediction and update iterative calculations of a hybrid Kalman filter based on the variable process noise covariance matrix update rule and the measurement noise adaptive adjustment matrix to obtain an initial estimate of the battery state of charge; By monitoring the inflection point characteristics of the battery voltage-current curve, the static stage and the fully charged stage are identified, and the state of charge recalibration time point is obtained; The initial estimated value of the battery state of charge is corrected at the state of charge recalibration time point to obtain an estimated value of the battery state of charge.

[0012] Optionally, in a seventh implementation of the first aspect of the present invention, creating a variable process noise covariance matrix update rule according to the probability values ​​of different degradation mechanisms in the microscopic degradation state determination result includes: Normalizing the probability values ​​of the six degradation mechanisms of SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse in the microscopic degradation state determination results to obtain a standardized probability distribution of each degradation mechanism; Based on the standardized probability distribution, a mapping relationship of weight coefficients corresponding to different degradation mechanisms is designed to obtain a mechanism weight table; Calculating the adaptive coefficient according to the mechanism weight table to obtain the adaptive coefficient reflecting the comprehensive influence of multiple degradation mechanisms, and setting the diagonal elements of the initial process noise covariance matrix to obtain the basic noise covariance matrix; Creating a noise matrix generating function that is dynamically adjusted according to the degradation state based on the adaptive coefficient and the basic noise covariance matrix; An update trigger condition is set for the noise matrix generation function, and an update calculation is performed when the variation range of the degradation mechanism probability distribution exceeds a preset target value to obtain an update rule for the variable process noise covariance matrix.

[0013] Optionally, in an eighth implementation of the first aspect of the present invention, the intelligent monitoring method for new energy vehicle batteries further includes: Constructing a multi-dimensional evaluation system for battery abnormality based on the battery state of charge estimation value and the microscopic degradation state determination result, and setting an abnormality level classification standard based on the multi-dimensional evaluation system for battery abnormality; Performing real-time monitoring on the battery state of charge estimation value and the microscopic degradation state determination result according to the abnormality level classification standard to obtain a current abnormality level determination result of the battery; Under the condition that communication is normal and sensor data is reliable, a state feedback gain matrix is ​​constructed based on the battery dynamic equivalent model to obtain a control law in a model control mode; When a communication interruption or data anomaly is detected and the battery's current abnormality level exceeds a preset alarm level, current limiting, active balancing, or emergency power-off measures are activated according to the abnormality type to obtain a protection strategy under the rule control mode; By defining the switching function to construct a sliding mode controller and setting the switching conditions, when the system switches from the model control mode to the rule control mode, the estimation error is guaranteed to meet the preset convergence conditions, and a dual-mode switching battery intelligent protection control system is obtained.

[0014] In a second aspect, the present invention provides an intelligent monitoring system for a new energy vehicle battery, the intelligent monitoring system for a new energy vehicle battery comprising: The acquisition module is used to obtain the target battery operation data set of the new energy vehicle battery and extract the time domain differential features and frequency domain energy loss features during the battery charging and discharging process to obtain a multidimensional state feature set; an electrochemical characteristic analysis module, configured to perform electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; A generation module is used to dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

[0015] In the technical solution provided by the present invention, the data quality of battery status monitoring is significantly improved through a high-precision sensor array and a three-stage filtering network, sampling noise and pulse interference are eliminated, and the continuous integrity of the data is achieved. Combined with Shannon entropy calculation and Laplace transform, the coordinated extraction of time domain differential features and frequency domain heat loss features is achieved, and the nonlinear change trend of battery capacity decay and the energy loss characteristics at different frequencies are fully captured. Through the recursive least squares parameter identification algorithm and the parameter correction coefficient matrix, a more accurate battery dynamic equivalent model is constructed, which overcomes the problem of nonlinear change of battery polarization characteristics and adapts to various complex working conditions. By using electrochemical impedance spectroscopy time domain mapping and complex domain convolution operation, combined with a deep learning framework, accurate identification of the microscopic degradation mechanism inside the battery is achieved, and a variable process noise covariance matrix and a measurement noise adaptive adjustment mechanism based on residual analysis are introduced. Combined with the recalibration of the charge and discharge inflection point, the problems of insufficient estimation accuracy and cumulative error in traditional methods are solved. Through a multi-level fault detection mechanism and a sliding mode controller, a smooth transition between model control and rule control is achieved, ensuring that the system can still maintain basic protection functions in the event of communication interruption or data anomaly, significantly improving the safety and reliability of battery use. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 Schematic diagram of an embodiment of an intelligent monitoring method for new energy vehicle batteries according to an embodiment of the present invention; Figure 2 The figure is a schematic diagram of an embodiment of an intelligent monitoring system for new energy vehicle batteries according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] An embodiment of the present invention provides an intelligent monitoring system and method for new energy vehicle batteries. The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the intelligent monitoring method for new energy vehicle batteries in an embodiment of the present invention includes: Step S101: obtaining a target battery operation data set of a new energy vehicle battery, and extracting the time domain differential features and frequency domain energy loss features during the battery charging and discharging process to obtain a multidimensional state feature set; It is understandable that the execution subject of the present invention can be an intelligent monitoring system for new energy vehicle batteries, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0020] Specifically, a high-precision sensor array is deployed within the battery management system to collect key operating parameters of the battery during operation. This sensor array includes voltage sensors for monitoring cell voltages, current sensors for detecting charge and discharge currents, a distributed thermal array for measuring battery surface temperature, and an external temperature sensor for monitoring ambient temperature. All sensors operate at a high sampling frequency, ensuring detailed capture of dynamic changes in the battery's operating state, resulting in high-fidelity, time-continuous raw battery parameter data. The collected raw battery signals contain noise and disturbances due to electromagnetic interference, sensor errors, and other factors. Therefore, multi-stage denoising is performed. A denoising method with multi-scale analysis capabilities is applied to filter out high-frequency noise contained in the signal, thereby retaining the principal component data of the battery's true response. To suppress pulse interference caused by switching of onboard electronic devices or transient interference, a median-based filtering algorithm combined with a dynamic threshold adjustment mechanism is used to correct for sudden changes in the data stream, improving signal stability and authenticity. Based on this processing, high-order interpolation techniques are used to complete the discontinuous data and reconstruct a complete and continuous battery operating trajectory. The sampling time delays between different sensors are synchronously corrected, and all physical quantities are normalized according to a unified numerical range to form a standardized target battery operation data set. The time domain differential characteristics and frequency domain energy loss characteristics of the battery during charging and discharging are extracted based on the target battery operation data set. On the time scale, the response rate and dynamic change trend of the battery during charging and discharging are analyzed to explore its energy conversion efficiency and polarization state evolution during operation; on the frequency scale, the impedance changes of the battery at different frequencies are combined to construct a frequency domain indicator reflecting its energy loss capacity, thereby identifying the energy loss characteristics caused by internal structural degradation. By fusing and compressing the features extracted in the two dimensions of time and frequency, a multidimensional state feature set with high information density is formed.

[0021] In time-domain analysis, voltage and current data are differentially processed to obtain the relationship between the rate of change of voltage response and capacity per unit time, i.e., voltage-capacity differential data. Voltage-capacity differential data is an important manifestation of the dynamic response of batteries during charge and discharge, reflecting the changing trends of the battery's internal polarization state, ion transport impedance, and active material utilization. To transform voltage-capacity differential data into quantifiable and comparable structured information, a grid discretization strategy is used to perform a two-dimensional discrete mapping. The continuous differential curve data is divided into equally spaced voltage and capacity intervals. The statistical distribution density within these discrete units is then calculated to construct a differential characteristic matrix. This matrix spatially characterizes the state response characteristics of the battery as it evolves over time, and its distribution pattern embodies the dynamic changes of the battery at different states of charge. To extract higher-order statistical properties, the Shannon entropy calculation function is introduced to measure the information entropy of the numerical distribution in this matrix, quantifying its structural complexity and uncertainty, and forming a set of time-domain characteristic vectors with statistical discriminant capabilities. This vector effectively identifies changes in macroscopic response characteristics caused by microscopic changes such as lithium dendrite formation, increased electrode polarization, or active material depletion, demonstrating strong sensitivity and stability. Simultaneously, the system performs frequency domain feature extraction. By applying a Laplace transform to the current signal in the target battery operating dataset, the time-evolving time-domain signal is mapped to frequency space, forming a frequency domain function representing the current spectrum distribution. In frequency space, signal responses in different frequency bands correspond to the battery's energy transfer behavior at different time scales. Therefore, the current amplitude response at each of these key pre-defined frequencies (e.g., low frequencies reflect polarization, mid-frequency reflects charge diffusion, and high frequencies reflect changes in contact impedance) is extracted. Combined with the equivalent internal resistance parameters stored in the battery equivalent model, the energy loss characteristics resulting from the interaction between current and impedance are calculated at each frequency point. This constructs a frequency domain feature vector that reflects the battery's thermal loss characteristics at different operating frequencies. This vector is suitable for identifying high-frequency impedance increases caused by mechanisms such as electrolyte decomposition, SEI film growth, or interface aging. The tensor outer product operation is used to combine the time domain eigenvector and the frequency domain eigenvector into a high-order tensor structure, and the redundant information is compressed and the structure is optimized through tensor decomposition technology to form a state feature set with unified structure and multi-dimensional expression capability.

[0022] Step S102: performing electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; Specifically, an equivalent modeling framework for the internal electrochemical behavior of a battery is established based on a multidimensional state feature set. This method uses the multidimensional state feature set obtained in the previous step as input. This feature set combines the time-domain differential response and frequency-domain energy loss characteristics extracted during the battery's charge and discharge processes, characterizing the battery's internal structure and dynamic behavior. Therefore, it is used to construct a physically meaningful equivalent circuit model. Based on these characteristic data, an initial set of battery polarization characteristic parameters is constructed, including the open-circuit voltage (OCV) reflecting the battery's static behavior, the ohmic internal resistance (OHM) representing the electronic conduction losses, and two parallel RC networks to describe electrochemical and concentration polarization phenomena. This model structure is more complete than a traditional first-order RC circuit and can capture both the transient and hysteretic responses of the battery at different charge and discharge rates. To ensure that the model parameters reflect the nonlinear variations of the battery under actual operating conditions, a prediction error function is applied to this initial parameter set to measure the deviation between the model output and the actual measured voltage. A recursive least squares algorithm is then used to iteratively update the model, enabling it to adaptively adjust to the new input data at each moment. This iterative process enables online model optimization and enhances the system's robustness and tracking capabilities under complex operating conditions. During each parameter update, the current state of charge and temperature information are incorporated as dynamic correction factors to generate a parameter correction coefficient matrix, enabling real-time adjustment of equivalent circuit parameters with state of charge and ambient temperature. This approach allows the model to adapt to changes in the battery's electrochemical behavior under various operating conditions, such as low temperature, high power, or aging, while avoiding structural mismatch and estimation bias caused by fixed parameters. Based on the updated equivalent circuit parameters, a state-space model is constructed, with the state vector consisting of the battery's state of charge and the polarization voltages on the two RC branches. The state transition equations describe its dynamic evolution over time, while simultaneously constructing a system output equation to characterize the functional relationship between the observed voltage and the internal state. This state-space structure connects the battery's internal dynamic processes, which are not directly observable, with external measurement data, enabling observable modeling of the electrochemical processes. To provide predictive capabilities for the model, state prediction rules and system output rules are defined within the structure, clarifying how to estimate the battery's response at the next moment based on the current state and how to convert the state vector into an observable terminal voltage, thereby completing the mathematical modeling of a closed-loop feedback system. This dynamic equivalent model analyzes the battery's electrochemical characteristics in real time. By continuously monitoring the evolution of state variables and the changes in model parameters, it identifies characteristic patterns that represent battery performance degradation. When the system detects phenomena such as an abnormal increase in the polarization time constant, a drift in the open-circuit voltage platform, or a significant increase in internal resistance, it determines that corresponding microscopic degradation mechanisms are occurring, such as abnormal growth of the SEI film, obstructed lithium-ion migration, electrolyte aging, or electrode structure collapse.By continuously updating the model status and accurately tracking the parameter change trajectory, the system can identify the degradation evolution process inside the battery with high resolution and form quantifiable degradation judgment results.

[0023] Based on the dynamic equivalent model of the battery, a frequency-domain response-based electrochemical impedance spectroscopy (EIS) analysis process was designed to reveal signs of structural degradation at the microscopic level. This process uses pre-set, logarithmically divided frequency points as excitation nodes, selecting several representative frequency ranges from very low to high. When the battery is operating in a stable state, the system uses the dynamic equivalent circuit model to superimpose a set of small-amplitude, low-perturbation AC signals at each frequency point. The signal amplitude is set to a fraction of the rated capacity to ensure no destructive effects on the battery. The excitation frequency ranges from millihertz to kilohertz, thereby obtaining frequency response information. The system records the battery voltage response changes in real time after the excitation is applied, forming a time-domain voltage response data sequence. The acquired time-domain voltage response data is processed in the complex domain, and the raw signal is mapped to the frequency domain through Fourier transform and complex convolution operations, thereby reconstructing the frequency-domain EIS data. In the frequency domain, the real and imaginary parts of the impedance together reflect the changing trends of the battery's internal transport behavior, interfacial reaction rates, and ion diffusion. The frequency domain impedance spectrum describes the response characteristics of the battery under different frequencies and is an important basis for identifying internal aging mechanisms. On this basis, an impedance characteristic vector composed of the impedance response at each frequency point is constructed, and the difference analysis is performed with the pre-constructed healthy battery standard impedance template to calculate the impedance difference characteristic vector, thereby extracting the specific physical quantity changes that reflect the degradation of the internal structure and function of the battery. In order to more systematically reveal the formation process of the degradation mechanism, the above impedance difference characteristic vector is combined with the time-varying circuit parameter sequence in the dynamic equivalent circuit model to construct a multidimensional degradation feature data set reflecting the evolution of battery aging. This data set contains the frequency domain characteristics of the static structural response, and integrates the dynamic evolution trajectory of state variables such as resistance, capacitance, and polarization voltage over time, which can capture the development path of the degradation mechanism from the time series dimension. A dataset of battery degradation features is fed into a deep learning framework consisting of a bidirectional long short-term memory (LSTM) network and a multi-head self-attention mechanism. The bidirectional LSTM architecture simultaneously captures forward and backward dependencies in the time series, identifying both slow-changing trends and sudden changes in degradation. The self-attention mechanism automatically focuses on key features and regions of significant parameter change, improving the model's accuracy in detecting complex degradation patterns. After network training, the deep recognition model outputs a probability distribution describing the likelihood of six typical degradation mechanisms in the battery's current state. These mechanisms include SEI film growth caused by electrolyte side reactions, active material shedding leading to irreversible capacity loss, lithium metal precipitation due to potential unevenness, electrolyte decomposition leading to gas production and electrolyte failure, current collector performance degradation due to localized corrosion, and electrode structural collapse due to accumulated charge and discharge stress. Based on the probabilities corresponding to each degradation mechanism, the system comprehensively determines the current microscopic degradation state of the battery, generating a quantitative, clear, and reliable degradation state determination.

[0024] Post-processing analysis of the degradation mechanism probability distribution is performed. During this process, the probability values ​​corresponding to the six degradation mechanisms are compared one by one, and a set of discrimination thresholds that balance sensitivity and stability are set to screen the degradation factors with significant influence at the current moment. Only when the probability value of a mechanism exceeds the set threshold is it included in the set of candidate dominant degradation mechanisms. This ensures that misjudgments caused by occasional perturbations or model uncertainty are eliminated, forming a reliable subset of candidate mechanisms that includes SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse. Based on the candidate dominant degradation mechanism set, the probability values ​​corresponding to each mechanism within it are normalized, and the relative contribution of each mechanism to the current degradation state is calculated. This proportion reflects its dominance in the process of battery performance degradation. Based on the calculated results, the degradation mechanisms are arranged in descending order of relative contribution, forming a sequence of decreasing importance. The leading degradation mechanism ranked first in the sequence is selected as the core basis for life evolution modeling. An appropriate life decay function model is selected based on its physical evolution characteristics. For example, for SEI film growth, an exponential-logarithmic composite function is suitable to describe the gradual growth trend; for active material loss, a quadratic function structure with a nonlinear decreasing trajectory is more suitable; and for highly dynamic processes such as rapid lithium dendrite precipitation or drastic electrolyte decomposition, an exponential function is used to characterize the accelerated decay effect. The determined decay function structure is applied to key state variables associated with this leading mechanism in historical operating data, such as changes in battery internal resistance, voltage plateau drift, or changes in impedance spectrum eigenvalues. A minimum error fitting strategy is used to solve the parameters and obtain a set of degradation trend prediction parameters that match the current operating conditions. This parameter set reflects the degradation rate, trend, and long-term behavior of the battery under the current dominant mechanism. Based on this, the life decay function is quantitatively integrated, taking into account the current battery state of charge, temperature conditions, and cycle history. The remaining number of cycles during which the battery can maintain normal performance is derived, and the corresponding actual service life is estimated by combining indicators such as operating frequency and usage intensity. This calculation result constitutes the final output for determining the microscopic degradation state.

[0025] Step S103: Dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

[0026] Specifically, a state-space representation structure is constructed based on the dynamic equivalent modeling of the battery. This structure uses the battery's state of charge, electrochemical polarization voltage, and concentration polarization voltage as core state variables, establishing a functional relationship between state transitions and observations. This provides the mathematical basis for prediction and updating of the hybrid Kalman filter. Within this structure, the prediction equation describes the evolution of the state over time, while the update equation incorporates current measurement data to modify the state, enabling the filter to dynamically track the true state. The construction of this system of equations relies on the dynamic equivalent circuit parameters obtained from the previous identification, ensuring good physical interpretability and state observability of the constructed filter. Furthermore, a set of update rules for the variable process noise covariance matrix is ​​formulated based on the degradation mechanism probability distribution output by the microscopic degradation state determination module in the previous stage. Different degradation mechanisms, such as SEI film growth, lithium precipitation, or electrolyte decomposition, induce varying degrees of system dynamic changes, and thus the corresponding process noise uncertainty is adjusted accordingly. The probability values ​​of each mechanism are mapped to noise enhancement coefficients, and the original process noise covariance matrix is ​​modified before each filter iteration. This allows the prediction model to perceive the degree of system instability in the current state and adaptively expand or compress its uncertainty modeling capabilities for state transitions. Furthermore, to address the drift of measurement errors during long-term operation, an adaptive measurement noise adjustment mechanism based on residual analysis is introduced. This mechanism statistically analyzes the residual changes between the measured value and the predicted output at each moment during the filtering process, performs residual sequence analysis using a sliding window, and calculates its covariance matrix to construct a dynamically changing measurement noise adjustment matrix. Combining these adaptive adjustment mechanisms for process and measurement noise, a hybrid Kalman filter is used for prediction and update iterations. The filter gain and covariance propagation path are dynamically adjusted at each moment to generate an initial estimate of the battery state of charge (SOC). This estimate can be continuously updated as the battery state changes, demonstrating strong adaptability and real-time performance. To improve the accuracy of the estimate under extreme conditions, a SOC recalibration strategy based on inflection point identification in the voltage-current curve is designed. The system analyzes the battery's operating curve in real time, identifying periods of near-zero current, minimal fluctuation, and stable voltage as the resting phase, and periods of near-zero voltage as the fully charged phase. A baseline mapping relationship for the state of charge is established at these two time points, and previously estimated SOC values ​​are calibrated and corrected at these key nodes. Specifically, the nonlinear relationship between open-circuit voltage and SOC is used to infer the current true state of charge during the resting phase, while the SOC can be directly corrected to the full value during the fully charged phase, avoiding estimation offsets caused by error accumulation during long-term iterations. This results in an estimated battery state of charge.

[0027] The probabilities of the six degradation mechanisms output by the degradation mechanism identification module are normalized to eliminate numerical offsets caused by differences in the original calculation scale, weighting model, or sample history. Using a standard normalization strategy, the probability values ​​of each degradation mechanism are mapped to a uniform, standardized interval, such as between 0 and 1, forming a standardized probability distribution vector. Based on this standardized probability distribution, a corresponding weight mapping is assigned to each degradation mechanism using an empirical model, a weighting strategy, or an optimization function driven by historical data. This functional relationship between the mechanism probability and the covariance weight is established. For example, if SEI film growth corresponds to slow system degradation, a smaller weight coefficient is assigned; whereas, lithium deposition or electrolyte decomposition, which correspond to more volatile state fluctuations, receive a significantly larger weight. By mapping mechanism-weight pairs into a mechanism weight table, the sensitivity to the uncertainty impact of each degradation source is adjusted under different degradation scenarios. Based on this weight table, a global adaptive coefficient is calculated, which represents the comprehensive impact of the current degradation environment on the system's process noise propagation capability. The adaptive coefficients are combined with the diagonal elements of the initial process noise covariance matrix to define a set of diagonal elements representing the fundamental perturbation level of each state variable, generating a basic noise covariance matrix. Based on the adaptive coefficients and the basic noise covariance matrix, a noise matrix generation function is created that dynamically adjusts to the degradation state. This function uses the adaptive coefficients as a dynamic weight source to map the mechanism effects to different process noise channels in the state-space equation. For example, the lithium precipitation mechanism is mapped to the enhanced state perturbation in the polarization branch, and the SEI film growth mechanism is projected to the increased uncertainty in state-of-charge drift. This ensures that the generated covariance matrix is ​​physically consistent with the true impact direction of the degradation mechanism. To prevent frequent updates from causing system fluctuations or increased computational overhead, an update trigger mechanism is set for the noise generation function. This triggers the regeneration of the covariance matrix only when the probability distribution of the degradation mechanism changes significantly within a short period of time, such as when the cumulative change exceeds a preset threshold. Through the above steps, the system is able to dynamically construct a process noise covariance matrix that is adapted to the actual degradation state of the battery, and implement structural corrections to the state transition uncertainty model in the hybrid Kalman filter, thereby improving the stability and accuracy of the battery state of charge estimation during the degradation evolution process.

[0028] Using the battery's estimated state of charge (SOC) and microscopic degradation state determination results as core input variables, a multi-dimensional abnormal state assessment system is constructed to reflect the battery's operational safety, health status, and functional stability. This system comprehensively analyzes whether the current SOC is within the safe range, whether the activity of different degradation mechanisms exceeds normal thresholds, and whether the battery's dynamic response exhibits hysteresis or instability. It then defines multiple indicator parameters in the feature space, such as voltage deviation, current mutation, temperature rise rate, impedance increase, and degradation mechanism weights. This weighted combination then forms a state assessment vector reflecting the battery's overall risk level. Based on this vector, the system establishes several grading rules, classifying abnormal states into four levels: minor, significant, severe, and urgent. Each level is assigned a corresponding response mechanism and threshold criteria, forming a battery abnormality classification standard. Based on this, the battery's estimated SOC and degradation state determination results are sampled in real time and fed into the multi-dimensional assessment system for online abnormality level determination. If the current state indicator is on the verge of a warning or if any indicator exceeds the set threshold, the system automatically determines the battery's current abnormality level and marks it with the corresponding state level result. If the communication channel is stable at this time and the data from various sensors is complete and reliable, the system will use the state space matrix defined in the model structure and the current state variables based on the battery dynamic equivalent model to construct a state feedback gain matrix through a linear feedback gain design method, thereby generating a dynamic control law for the current operating conditions. This control law calculates the deviation between the current state and the target desired state and outputs real-time control instructions for adjusting the current, power or voltage to maintain the stable operation of the battery system. This process is the implementation stage of the model control mode and has highly refined and adaptive adjustment capabilities. If the system detects communication interruption, sensor data distortion or loss during operation, and at the same time the current abnormality level of the battery has reached or exceeded the preset alarm level threshold, the model control mode will automatically fail and the system will immediately switch to the rule control mode. In this mode, the corresponding safety protection strategy is selected based on the identified anomaly type: if an overcurrent trend or extreme temperature rise is detected, the system activates the current limiting mechanism, dynamically adjusting the current limit value to suppress current peak fluctuations; if the voltage difference between the single cells is found to continue to increase, the system initiates the active balancing mechanism, using bypass resistors to release energy to the high-voltage cells; if severe overtemperature, overvoltage, or uncontrollable electrical shock occurs, an emergency power-off control is triggered, disconnecting the main circuit by disconnecting the high-voltage relay to ensure overall system safety. To achieve smooth switching between the two control modes, a sliding mode controller is embedded in the control structure, and a state switching function is defined to construct the switching trajectory based on the error between the current system state and the desired state.Before the system enters rule-based control, the sliding mode controller forces the error to remain within the convergence boundary and ensures that state disturbances and control offsets during the switching process do not exceed the set tolerance threshold. This ensures a certain degree of robustness and continuity in the transition between model-based control mode and rule-based control mode. This implements a dual-mode battery protection control mechanism that is primarily model-driven and supplemented by rule-based protection.

[0029] In an embodiment of the present invention, a high-precision sensor array and a three-stage filtering network are used to significantly improve the data quality of battery status monitoring, eliminate sampling noise and pulse interference, and achieve data continuity and integrity. Combined with Shannon entropy calculation and Laplace transform, the time domain differential features and frequency domain heat loss features are synergistically extracted to fully capture the nonlinear change trend of battery capacity decay and the energy loss characteristics at different frequencies. Through the recursive least squares parameter identification algorithm and parameter correction coefficient matrix, a more accurate battery dynamic equivalent model is constructed, overcoming the problem of nonlinear changes in battery polarization characteristics and adapting to various complex working conditions. By using electrochemical impedance spectroscopy time domain mapping and complex domain convolution operations, combined with a deep learning framework, accurate identification of the microscopic degradation mechanism inside the battery is achieved. The introduction of a variable process noise covariance matrix and a measurement noise adaptive adjustment mechanism based on residual analysis, combined with charge and discharge inflection point recalibration, solves the problems of insufficient estimation accuracy and cumulative error in traditional methods. Through a multi-level fault detection mechanism and a sliding mode controller, a smooth transition between model control and rule control is achieved, ensuring that the system can still maintain basic protection functions in the event of communication interruption or data anomaly, significantly improving battery safety and reliability.

[0030] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Collect the single cell voltage, charge and discharge current, surface temperature distribution and ambient temperature of new energy vehicle batteries to obtain the original battery parameter data; Performing denoising processing on the original battery parameter data to obtain battery signal data with sampling noise removed, and performing high-frequency interference processing on the battery signal data with sampling noise removed to obtain battery signal data with high-frequency interference eliminated; Reconstruct missing points of the battery signal data after eliminating high-frequency interference to obtain continuous and complete battery signal data. Perform time synchronization and standardization on the continuous and complete battery signal data to obtain the target battery operation data set. The time domain differential features and frequency domain energy loss features of the battery during charging and discharging are extracted according to the target battery operation data set to obtain a multidimensional state feature set.

[0031] Specifically, a high-precision sensor array is integrated into the battery management system to continuously and real-timely sample cell voltage, charge and discharge current, battery surface temperature distribution, and ambient temperature at high frequencies. During data acquisition, timestamps are uniformly assigned to each acquisition channel to ensure a consistent time-domain reference for subsequent multidimensional data fusion operations. Noise reduction is performed on the raw battery parameter data to effectively remove noise introduced by factors such as environmental electromagnetic interference, sensor quantization errors, and power supply instability. A multi-scale wavelet transform is preferentially used to decompose each signal channel. By selecting a mother wavelet basis function (such as DB4) suitable for the characteristics of battery time series signals, the signal is decomposed into approximate and detail components at different scales. A soft threshold function is applied to the detail components to achieve noise compression, reconstructing a purified signal dominated by true physical trends, effectively preserving the dynamic characteristics of the battery response. High-frequency interference signals are then identified and corrected. A sliding window-based median filter method combined with an adaptive threshold strategy identifies sudden pulses caused by motor commutation, power switching, and high-frequency switching of the electronic control system. These pulses are then smoothed and corrected, making the overall signal more continuous and stable, ensuring that the true representation of the battery's operating state is unaffected by abnormal pulse fluctuations. High-order spline interpolation reconstruction technology leverages the changing trends between adjacent points to achieve smooth reconstruction without destroying the original data structure, thus forming a complete and continuous battery signal data stream. After signal repair is complete, all parameter channels are time-synchronized to eliminate sampling offsets caused by sensor response speed differences and data frame delays. Standard normalization methods are then used to uniformly map different physical quantities (such as voltage, current, and temperature) to equivalent numerical ranges (e.g., [-1, 1]). This creates a target battery operating dataset with a unified format, time domain alignment, and consistent structure. Time-domain differential features and frequency-domain energy loss characteristics during the battery charging and discharging processes are extracted from this target battery operating dataset. In terms of time domain characteristics, the sensitivity of voltage response to charge and discharge current behavior is analyzed, the rate of voltage change caused by unit charge change is calculated, and a voltage-capacity differential sequence is constructed with the help of a sliding window mechanism. This differential characteristic reflects the internal ion migration efficiency and polarization voltage evolution trend of the battery, and is highly sensitive to degradation processes such as reduced electrode material activity and decreased charge capture ability. Through discretization mapping of a certain time interval, a differential characteristic matrix with graphical properties can be formed for further nonlinear behavior characterization. At the same time, power loss analysis is carried out synchronously in the frequency domain. The original current signal is mapped to the frequency domain through Laplace transform or fast Fourier transform. Combined with the equivalent internal resistance parameters corresponding to several characteristic frequency points of the battery (such as 0.01Hz, 0.1Hz, 1Hz, 10Hz, 100Hz), the frequency domain energy loss level at each frequency point is calculated.This analysis method can characterize the energy decay characteristics of the battery from the perspective of frequency response under the combined influence of multiple factors such as internal ohmic loss, polarization impedance, and diffusion impedance, providing important input for subsequent battery health evaluation, dynamic loss modeling, and thermal management risk assessment. The above-mentioned time-domain differential characteristics are fused with the frequency-domain loss characteristics, and a high-dimensional feature tensor is constructed through tensor outer product. Then, tensor decomposition technology is used to extract representative principal component dimensions, forming a multidimensional state feature set with high information density, strong scalability, and engineering interpretability.

[0032] In a specific embodiment, the step of extracting the time-domain differential features and frequency-domain energy loss features during the battery charging and discharging process based on the target battery operation data set to obtain a multidimensional state feature set may specifically include the following steps: Perform differential calculation on the voltage and current data in the target battery operation data set to obtain voltage-capacity differential data; Perform grid discretization on the voltage-capacity differential data to obtain the differential characteristic matrix; The Shannon entropy calculation function is used to quantify the characteristic distribution of the differential characteristic matrix to obtain the time domain characteristic vector; Perform Laplace transform on the current signal in the target battery operation data set to obtain current spectrum data. Based on the current spectrum data and the impedance value of the battery equivalent internal resistance at the preset characteristic frequency point, the frequency domain eigenvector is calculated. The time domain feature vector and the frequency domain feature vector are fused and tensor decomposition is performed to obtain a multidimensional state feature set.

[0033] Specifically, the time-distributed voltage and current series in the target dataset are differentiated. By pairing the voltage change in adjacent time intervals with the current integral within those time intervals, voltage-capacity differential data are constructed, reflecting the voltage response rate to a unit charge change. The differential results reveal the dynamic response amplitude of the battery under unit load and can sensitively reflect detailed changes in electrochemical processes such as decreased electrode activity, enhanced polarization, and increased ion transport resistance. To transform this differential data into structured information for modeling and distribution analysis, a two-dimensional mapping method is used to partition the voltage and capacity variation space into a number of equally spaced intervals, forming a discrete grid structure in two-dimensional space. The number of data points occurring in each grid interval is counted to generate a differential characteristic matrix reflecting the distribution characteristics of the dynamic response. The Shannon entropy function is used to quantify the characteristic distribution of the differential characteristic matrix. The probability distribution of each grid cell in the characteristic matrix is ​​used as an input variable, and the entropy value is calculated to characterize the complexity and uncertainty of the system state. Specifically, the more uniform and dispersed the distribution, the higher the entropy value, indicating that the system state response is more complex and the distribution is wider; conversely, if the matrix is ​​concentrated in a small area, it indicates that the response behavior tends to be stable or rigid, indicating a certain degradation trend or response inactivation. Through Shannon entropy mapping, the entire two-dimensional distribution feature is compressed into a time domain feature vector with statistical significance. At the same time, the dynamic loss characteristics of the battery are synchronously constructed in the frequency domain to supplement the energy conversion mechanism that cannot be characterized by simple time domain differentiation. In the frequency domain analysis, the current signal in the target data set is selected as the research object, and a Laplace transform or Fourier transform operation is performed on it to map the original time series current signal into a frequency domain signal, capturing the differences in the battery's electrochemical response capabilities at different frequencies, thereby revealing the energy transmission efficiency and loss structure in the frequency channel. After the transformation is completed, a set of current spectrum data is obtained, where the amplitude at each frequency point represents the energy proportion occupied by the frequency band. Combined with the equivalent internal resistance parameters at several typical frequency points (such as low-frequency polarization points, medium-frequency ion migration points, and high-frequency interface impedance points) in the battery equivalent circuit model, the power distribution in the spectrum is weighted to calculate a set of frequency-domain energy loss eigenvalues, thereby forming a frequency-domain eigenvector containing multiple key frequency components. The time-domain eigenvector and the frequency-domain eigenvector are combined into a high-order tensor structure using tensor operations. A three-dimensional tensor is constructed by performing an outer product operation on the two eigenvectors, in which one dimension represents the time-domain response structure, the other dimension represents the frequency response characteristics, and the third dimension is used to store sample sequences under different time slices or working conditions, thus forming a tensor data structure with complete characterization capabilities.In order to avoid redundancy and computational burden caused by excessive tensor dimension, tensor decomposition methods such as high-order singular value decomposition or tensor principal component extraction algorithm are used to perform structured dimensionality reduction on the original tensor and retain the principal components with the highest information density, ultimately obtaining a multidimensional state feature set containing several highly expressive principal axes.

[0034] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Based on the multi-dimensional state feature set, the initial values ​​of the battery polarization characteristic parameters including open circuit voltage, ohmic internal resistance and two RC parallel networks are constructed; The prediction error function is set for the initial value of the battery polarization characteristic parameter and iteratively calculated through the recursive least squares algorithm to obtain the parameter vector iterative update formula; The parameter vector iterative update formula is combined with the current battery state of charge and temperature to generate a parameter correction coefficient matrix, and the battery equivalent circuit parameters are adjusted in real time according to the working conditions. Based on the battery equivalent circuit parameters that are adjusted in real time according to the working conditions, a state vector equation including the state of charge and polarization voltage is established to obtain the battery state space model structure; According to the battery state space model structure, the calculation rules of the state prediction equation and the system output equation are set to obtain the battery dynamic equivalent model; The battery dynamic equivalent model is used to analyze the electrochemical characteristics of new energy vehicle batteries and obtain the results of microscopic degradation state determination.

[0035] Specifically, a battery equivalent circuit model with a well-defined physical structure is constructed based on a multidimensional state feature set. This model includes open-circuit voltage and ohmic internal resistance, reflecting the characteristics of the battery terminals, and considers more detailed dynamic behavior, particularly two RC parallel network branches that reflect the electrochemical polarization and concentration polarization processes. The open-circuit voltage, as a static function of SOC, reflects the battery's energy storage capacity in the open-circuit state. The ohmic internal resistance reflects the resistive losses introduced by the battery's internal materials, electrolyte, and connection contacts. The two RC branches correspond to the slow voltage recovery behavior caused by the hysteresis of electrochemical interface reactions and changes in material concentration gradients. The former has a direct impact on high-frequency dynamic response, while the latter affects the smoothness of long-cycle charge and discharge voltage curves. After setting the initial parameter structure, to ensure that the model can dynamically adjust to real-time changes in the battery state during actual operation, a recursive least squares algorithm is introduced as the core mechanism for parameter estimation, and a prediction error function is designed to guide the iteration. This function defines the objective to minimize, based on the difference between the actual battery output voltage and the voltage predicted by the current model output. The sum of squared errors is used to measure the accuracy of the model's current parameters. As the battery's runtime progresses, the system receives new input data in real time and feeds the prediction error back to the parameter update mechanism, enabling the model parameters to be adjusted based on the voltage response at each time point. By weightedly accumulating the prediction error and constructing an information gain function, a recursive least squares algorithm generates an iterative update formula, which is used to modify the parameter vector in each round, gradually approximating the true electrochemical state. This update mechanism offers real-time performance, stability, and noise immunity, making it suitable for tracking the state of new energy vehicle batteries under complex dynamic operating conditions. Because battery parameters exhibit significant nonlinear fluctuations across different states of charge, temperature environments, and even degradation levels, updates based solely on historical voltage and current data are insufficient to cover all operating conditions. Therefore, a parameter correction coefficient matrix based on the current state of charge estimate and temperature sensor data is introduced to perform lateral compensation corrections on the original parameter estimates. The correction coefficient is constructed by a preset two-dimensional lookup table. The input variables are the current SOC and temperature values, and the corresponding output is the correction multiple of each parameter, so as to realize dynamic adjustment of the parameters as the operating environment changes. This processing method ensures that the model parameters can maintain physical consistency and response accuracy even when the battery undergoes extreme operating conditions such as deep discharge, extremely low temperature start-up or high temperature fast charging, thereby improving the stability and adaptability of the overall model throughout its life cycle. Based on the circuit parameters that have been corrected in real time, a state vector structure containing core state variables such as state of charge and RC branch voltage is established, and the evolution relationship of the state vector in the time domain is set. The state transfer equation is based on the physical circuit model, and the change in SOC is integrally mapped to the input current. At the same time, an exponential decay factor is applied to the RC branch voltage to reflect its delayed response characteristics.The state vector equation describes the evolutionary logic of the system's implicit variables and transmits external control influences through the input vector (i.e., the current value), forming a mapping model of the system's internal and external responses. Based on the functional relationship between this state transition structure and the real-time measured voltage output, the system's output equation is established, allowing the battery terminal voltage to be predicted as a function of the internal state variables and the current input. This results in a complete state-space model encompassing both the state transition equation and the system's output equation. With this state-space model as the core, a dynamic equivalent model of the battery is constructed, serving as a unified mathematical framework for all subsequent electrochemical behavior analysis, state estimation, and degradation mechanism identification. This model is both structured, real-time, and scalable. It can not only accept external intervention variable inputs but also superimpose a degradation factor model on top of it, further enhancing its ability to model the degradation evolution of batteries over long-term use. This model is compared with historical parameter change curves, slow RC branch response trends, and patterns of sustained internal resistance growth to identify systematic errors between the model output and the actual response under identical inputs and trace these errors back to microscopic degradation mechanisms. After the error source analysis is complete, a set of corresponding relationships between characteristics and mechanisms is constructed based on the dynamic model, combining the impact of each degradation mechanism on the model parameters, such as the slow increase in internal resistance caused by SEI film growth, the drift of RC parameters caused by active material shedding, and the impact of lithium dendrite formation on the voltage hysteresis amplitude. By detecting the change trend of specific parameters over time, the intensity of nonlinear fluctuations, and the distribution pattern of model prediction errors, the system infers the dominant degradation mechanism currently occurring in the battery and forms a structured micro-degradation state judgment result. This judgment result has the function of mechanism classification and can also express the evolution rate of the degradation process, the potential impact range, and the future warning level in the form of probability or trend parameters.

[0036] In a specific embodiment, the process of performing electrochemical characteristic analysis on a new energy vehicle battery using a battery dynamic equivalent model to obtain a microscopic degradation state determination result may specifically include the following steps: Based on the battery dynamic equivalent model, a small AC excitation signal is applied at the preset logarithmically divided frequency points to obtain the battery voltage response time domain data; Perform complex domain convolution operation on the battery voltage response time domain data to obtain frequency domain electrochemical impedance spectroscopy data; Constructing an impedance characteristic vector based on the frequency-domain electrochemical impedance spectroscopy data and calculating the difference with the standard impedance spectrum of a healthy battery to obtain an impedance difference characteristic vector; Construct a battery degradation feature dataset based on the impedance difference feature vector and the circuit parameter time series in the battery dynamic equivalent model; The battery degradation feature dataset is input into the bidirectional LSTM layer and the multi-head self-attention mechanism to calculate the probability distribution and obtain the degradation mechanism probability distribution; According to the probability distribution of degradation mechanisms, the microscopic degradation state judgment results of six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion and electrode structure collapse are determined.

[0037] Specifically, based on the dynamic electrochemical characteristics within the battery, an inference mechanism is constructed that integrates frequency-domain impedance analysis and deep learning recognition. The core of this method lies in leveraging the controllability of the battery's dynamic equivalent model to actively apply a set of structured, small perturbation signals. Without affecting the battery's normal operation, this method induces the response behavior of its internal electrochemical structure to different frequency excitations, thereby obtaining frequency-domain features that can characterize its health state or degradation trend. To this end, the system presets a set of logarithmically divided frequency points, covering multiple orders of magnitude from extremely low frequency to high frequency, ensuring that the dynamic characteristics of the battery at different time scales can be effectively stimulated and collected. At each frequency point, the system applies a small-amplitude sinusoidal excitation signal and, based on the response equation output by the battery's dynamic equivalent model, records the voltage response curve caused by the excitation current, forming a set of frequency-dependent time-domain response sequences. The time-domain voltage response data is mapped to the complex domain to fully preserve the mapping relationship between the excitation signal and the response signal in terms of amplitude and phase. Complex domain convolution processing is performed on the input current and output voltage pairs at each set of frequency points, and the time domain signal is converted into a frequency domain expression using Fourier transform or Laplace transform. The complex impedance value corresponding to each frequency point is calculated by complex ratio calculation. The real part of this value represents the resistive impedance of the battery in this frequency band, that is, the ohmic resistance generated by the current passing through the conductive material, electrolyte and contact structure; the imaginary part represents the capacitive or inductive response of the battery, reflecting complex processes such as interfacial electrochemical reaction hysteresis, charge accumulation and dissociation, and diffusion barriers. These frequency-domain electrochemical impedance data are combined into an impedance spectrum vector, whose distribution structure reflects the overall dynamic mechanism of the battery from high-frequency fast behavior to low-frequency slow polarization process. In order to measure the degree of change in the impedance response of the current battery state compared to the healthy state, a standard impedance spectrum template is introduced as a reference. The template is measured by a new battery of the same model under consistent test conditions or established by long-term database statistics. The frequency difference between the current impedance spectrum vector and the standard impedance spectrum is calculated, and the difference vector is normalized to generate an impedance difference feature vector, which represents the degree of deviation of the battery from the ideal state in each frequency response dimension. For batteries in different degradation stages, their impedance difference characteristics show discernible structural characteristics in the high-frequency band, the medium-frequency band, and the low-frequency band. For example, the growth of the SEI film causes a slow increase in the real impedance in the high-frequency band, the precipitation of lithium metal causes an increase in the imaginary part fluctuation in the medium-frequency band, and the shedding of active materials or electrode collapse is often seen in a strong rise in the low-frequency impedance. At the same time, a parameter sequence related to the time evolution is extracted from the battery dynamic equivalent model, including the ohmic internal resistance, the resistance and capacitance parameters of the two RC networks, the polarization voltage evolution, and the state of charge estimation trajectory. These parameters are combined with the impedance difference characteristics to construct a degradation feature dataset that covers the static impedance offset information and dynamic modeling response behavior of the battery.This dataset, with time series on the horizontal axis and feature dimensions on the vertical axis, exhibits typical multivariate time series properties. Parameter trends, oscillation patterns, and hysteresis amplitudes may all hint at specific degradation pathways. To accurately identify the dominant degradation mechanism from this complex sequence, a deep recognition network architecture was designed, centered around a bidirectional LSTM (Long Short Term Memory) and supplemented by a multi-head self-attention mechanism for key feature extraction and mechanism distribution modeling. In practice, the bidirectional LSTM network utilizes a forward and backward recursive structure to simultaneously read the degradation feature dataset, capturing the dynamic evolution of parameters in the temporal direction. This allows for the identification of degradation patterns that only manifest over long timescales, such as slow capacitance drop and periodic polarization voltage drift. The multi-head self-attention mechanism establishes importance scores across different feature channels, weighting the most significant dimensions that influence the final recognition decision. This automatically focuses on key frequency bands, parameters, and nodes, while effectively suppressing interference from redundant dimensions and noise. This deep architecture ultimately outputs a probability distribution vector for the degradation mechanism, where each dimension represents the likelihood of a particular degradation mechanism occurring under the current state. The system uses this probability distribution as the final degradation mechanism identification result, and based on it, confirms which mechanism or combination of mechanisms drives the current battery degradation state, dominated by SEI film growth, active material loss, lithium metal precipitation, electrolyte decomposition, current collector corrosion, or electrode structure collapse. In addition, based on this distribution result, continuous tracking and prediction are carried out to determine the transition trend or combined effect between mechanisms, such as the evolution from SEI film growth to lithium dendrite precipitation, or the transition from active material shedding to structural collapse.

[0038] In a specific embodiment, the process of determining the microscopic degradation state determination results of six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse according to the degradation mechanism probability distribution may specifically include the following steps: Based on the probability distribution of degradation mechanisms, threshold screening is performed to obtain a set of candidate dominant degradation mechanisms, which include six degradation mechanisms: SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse. Based on the set of candidate dominant degradation mechanisms, the relative contribution ratio of each dominant degradation mechanism is calculated and ranked to obtain a degradation mechanism sequence ranked by importance. Assigning a life decay model in the form of an exponential-logarithmic composite function, a quadratic function, or an exponential function to the degradation mechanism ranked first in the sequence of degradation mechanisms sorted by importance; The parameter fitting calculation of the life decay model is performed to obtain the degradation trend prediction parameters. The remaining number of cycles and service life are calculated based on the degradation trend prediction parameters and the current battery status to obtain the microscopic degradation status judgment result.

[0039] Specifically, a six-dimensional probability vector output by a deep neural network (such as a bidirectional LSTM and a multi-head attention mechanism) is used for screening. This probability distribution covers six typical battery degradation mechanisms: SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse. To prevent low-probability perturbations from interfering with subsequent modeling and improve model focus and evolutionary rationality, a threshold screening mechanism is introduced at this stage. A lower probability limit is set to filter out mechanisms with extremely low weights in the current operation cycle. The screening criteria use either static thresholds or adaptive strategies (such as dynamic adjustment based on standard deviation). The remaining set of mechanisms constitutes the candidate set of dominant degradation mechanisms. This set reflects the core mechanisms that dominate battery performance degradation in the current state and has a clear causal basis. Based on this, the importance of each degradation mechanism in the candidate set is quantified to reflect its relative contribution to the overall degradation level. By constructing a multi-factor weighted fusion model, various information such as the original probability value, characteristic channel activity, and model residual sensitivity are fused into a single contribution index, and all candidate mechanisms are normalized according to this index to obtain the relative contribution ratio of each dominant mechanism. These ratios are used to sort and generate a sequence of degradation mechanisms arranged in descending order of importance, clearly indicating which mechanism is the main cause of degradation, which mechanisms are secondary factors, and which are potential evolution paths in the current battery operation stage. The system focuses on the first mechanism in the sorting sequence - that is, the dominant mechanism with the most significant current degradation trend, and assigns the corresponding life attenuation model based on its physical behavior characteristics. Different degradation mechanisms exhibit different time evolution behavior characteristics, so the use of a unified mathematical model will lead to poor adaptation, prediction deviation, and even failure. For example, SEI film growth exhibits an evolutionary trend of rapid growth initially followed by a slow, stabilizing phase in the later stages, making it suitable for fitting using an exponential-logarithmic composite function. Active material loss, on the other hand, occurs primarily in the middle and late stages and exhibits a slow, linear accumulation trend, making it more suitable for modeling using a quadratic function. Lithium dendrite precipitation is accompanied by sudden capacity loss and exhibits the typical exponential growth characteristic, making a single exponential function more suitable for describing its degradation trajectory. Therefore, upon mechanism confirmation, the system selects a mathematical expression from a library of pre-set function models that dynamically matches the mechanism and uses this expression as the foundation for the lifetime decay curve. Once the function model is determined, a parameter fitting process is performed. This process uses macroscopic degradation indicators such as battery capacity change, internal resistance growth, and voltage plateau drift recorded during historical operation cycles as target variables, while using time, cycle number, and state-of-charge changes as independent variables. The parameters in the selected function model are fitted using a least-squares method to obtain the parameter combination that best matches the actual battery's current operating trajectory.To improve fitting accuracy and robustness, a genetic algorithm or particle swarm optimization algorithm is introduced to perform a global search of the parameter space. A sliding window and recursive error minimization method are then combined to smooth short-term data fluctuations. This results in a set of degradation trend prediction parameters that are dynamically stable and physically well-defined. These parameters reflect the growth rate of the dominant degradation mechanism, the inflection point of the change curve, and the final level of degradation approach. They also imply the evolutionary rhythm of internal battery material aging, interfacial reaction deterioration, or structural damage. The parameterization of the lifetime decay model is then combined with the current battery's real-time operating state, including remaining capacity, voltage limit, and safety threshold, to estimate the remaining number of cycles required to maintain normal function under the prevailing mechanisms. By performing a reverse solution or curve integration on the model, an upper limit for the current remaining effective number of cycles is estimated. Furthermore, the number of cycles is mapped to the expected remaining useful life, taking into account user characteristics such as frequency of use, average daily energy consumption, and charging rate, generating an interpretable lifetime metric in days, months, or hours of use. The mechanism identification results, contribution ranking, attenuation model type, prediction parameters, remaining life and number of cycles are packaged together to form a micro-degradation state judgment result. The result points out the main causes and trends of the current battery performance decline, and provides users with targeted optimization suggestions, such as reducing fast charging frequency, reducing deep discharge behavior, and adjusting high-temperature usage strategies.

[0040] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Construct the prediction and update equations of the hybrid Kalman filter based on the state space equations in the battery dynamic equivalent model; Creating a variable process noise covariance matrix update rule based on the probability values ​​of different degradation mechanisms in the microscopic degradation state determination results; Perform sliding window analysis on the measurement residual sequence generated during the filtering process and calculate the residual covariance to obtain the measurement noise adaptive adjustment matrix; Based on the variable process noise covariance matrix update rule and the measurement noise adaptive adjustment matrix, the hybrid Kalman filter performs prediction and update iterative calculations to obtain an initial estimate of the battery state of charge; By monitoring the inflection point characteristics of the battery voltage-current curve, the static stage and the fully charged stage are identified, and the state of charge recalibration time point is obtained; The initial estimated value of the battery state of charge is corrected at the state of charge recalibration time point to obtain an estimated value of the battery state of charge.

[0041] Specifically, a state-space model is constructed based on a second-order RC equivalent circuit model. The state vector contains three core variables: the battery state of charge (SOC), the electrochemical polarization branch voltage, and the concentration polarization branch voltage. The input is the charge and discharge current of the current sampling cycle, and the output is the measured battery terminal voltage. The system establishes a state transition equation based on the integral characteristic of the current with respect to the SOC change. A recursive expression of the two RC branch voltages is constructed using an exponential decay form. The output equation connects all internal states with the externally observable voltage values, completing the structural modeling of the battery's internal electrochemical behavior and providing theoretical support and a framework for the filter's prediction and update. In the design of the hybrid Kalman filter, the state recursive relationship and covariance propagation path are defined in the prediction step. Specifically, the state is extrapolated using the prior state estimate and the process noise covariance matrix. Subsequently, in the update step, the Kalman gain is calculated based on the difference between the current measured battery terminal voltage and the predicted voltage, and the prior state estimate is corrected to generate the posterior estimate. In standard Kalman filters, the process noise covariance matrix and the measurement noise covariance matrix are often set as fixed constants. However, this is difficult to adapt to the degradation-evolving battery system. Because battery internal parameters continuously change with aging, a single noise assumption can easily lead to accumulated estimation bias or model mismatch. To address this issue, a degradation-mechanism-driven process noise adaptive adjustment mechanism is introduced. In the previous stage, the system used frequency-domain impedance analysis and deep learning to identify the probability distributions of six microscopic degradation mechanisms (such as SEI film growth, lithium precipitation, and active material loss) for the current battery. The system normalized the probabilities of these mechanisms and correlated them with a preset mechanism strength mapping table to generate a set of weight coefficients that characterize the influence of different degradation mechanisms on the state evolution path. These weight coefficients are combined into a global adaptive adjustment coefficient and, combined with the basic process noise covariance matrix, a variable process noise matrix is ​​constructed. This allows the filter to prioritize state diffusion in the electrochemical polarization branch when the probability of SEI film growth is high and enhance its sensitivity to nonlinear changes in the SOC evolution path when lithium precipitation dominates, thus achieving degradation-aware dynamic modeling of state perturbations. This mechanism enhances the filter's ability to detect periods of unstable state evolution, resolving the difficulty in tracking state drift in aging batteries. Furthermore, to compensate for fluctuations in observation errors caused by sensor aging, measurement bias, or external disturbances, a sliding window residual analysis module is designed within the filter to perform statistical analysis on the measurement residual sequence over several consecutive filter iterations. The system records the error between the actual voltage measurement and the model's predicted output and uses these residual values ​​as input. Using a sliding window strategy, the residual mean and covariance are calculated to determine the current level of measurement uncertainty.This covariance is defined as an adaptive measurement noise matrix, which dynamically replaces the static observation noise assumption. This allows the filter to adaptively expand observation uncertainty when the sensor state is unstable, avoiding overweighting anomalous observations. It also reduces the observation noise tolerance when the sensor state is good, achieving high-precision state correction. Driven bidirectionally by the variable process noise covariance matrix and the adaptive measurement noise matrix, the system performs recursive calculations of prediction and update equations at each time step to provide a preliminary estimate of the battery's SOC and related electrochemical state. To further mitigate estimation bias caused by model drift or residual accumulation over long-term use, a SOC recalibration mechanism is designed. This mechanism, based on the characteristic variation patterns of the battery's voltage-current curve, identifies two key operating conditions in real time: the quiescent phase, where the current approaches zero, the voltage remains stable, and internal reactions change slowly. This is a suitable timeframe for reverse calibration of the current estimate using the OCV-SOC mapping table; and the fully charged phase, where the voltage approaches the upper limit and the current gradually decreases. During this phase, the SOC is forcibly calibrated to the full value to prevent the estimated value from drifting long-term from the true full charge state. By jointly analyzing the slope, amplitude, and duration of the voltage-current curve, the system automatically marks the SOC recalibration point when the static or fully charged conditions are met, using this as a trigger to correct the current estimate. When in the static phase, the system consults a table to obtain the theoretical SOC value at the corresponding voltage and uses this value to correct the current estimate. When a fully charged characteristic is detected, the system directly calibrates the SOC to 1 to ensure that the upper boundary conditions are met.

[0042] In a specific embodiment, the process of executing the step of creating a variable process noise covariance matrix update rule according to the probability values ​​of different degradation mechanisms in the microscopic degradation state determination result may specifically include the following steps: The probability values ​​of the six degradation mechanisms (SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse) in the microscopic degradation state determination results were normalized to obtain the standardized probability distribution of each degradation mechanism. Based on the standardized probability distribution, the weight coefficient mapping relationship corresponding to different degradation mechanisms is designed to obtain the mechanism weight table; The adaptive coefficients are calculated according to the mechanism weight table to obtain the adaptive coefficients reflecting the combined effects of multiple degradation mechanisms, and the diagonal elements of the initial process noise covariance matrix are set to obtain the basic noise covariance matrix; A noise matrix generating function is created based on the adaptive coefficients and the basic noise covariance matrix, which is dynamically adjusted according to the degradation state. An update trigger condition is set for the noise matrix generation function. When the change amplitude of the degradation mechanism probability distribution exceeds the preset target value, the update calculation is performed to obtain the update rule of the variable process noise covariance matrix.

[0043] Specifically, the degradation mechanism identification model outputs a set of probability distributions corresponding to six typical internal battery degradation mechanisms: SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse. The raw probability values ​​are normalized to form a standardized probability distribution vector that sums to 1. This normalization employs max-min normalization, sum normalization, or a softmax function transformation to ensure that the relative strengths of different mechanisms are consistent and comparable on a numerical scale. The normalized probabilities are mapped to process noise impact factors. Because different degradation mechanisms have fundamentally different impacts on state diffusion during battery dynamic behavior, for example, SEI film growth primarily affects changes in ohmic internal resistance, causing output voltage drift; whereas lithium dendrite precipitation directly threatens the stability of the battery polarization process and causes abrupt changes in the state transition path. Therefore, its contribution to state perturbations should be given a higher weight. To this end, a mechanism-weight mapping table is constructed, in which each mechanism is assigned a static or dynamic weight coefficient. These coefficients are set based on experimental data, electrochemical mechanism models, or data-driven optimization strategies. These weight coefficients are element-wise multiplied by the normalized probability vector and then weighted summed to produce an adaptive coefficient that reflects the combined influence of multiple degradation mechanisms. This coefficient is used to control the global amplification or contraction of the process noise covariance matrix. The process noise covariance matrix is ​​adjusted based on the adaptive coefficients. Based on the initial process noise covariance matrix set during the system modeling phase, its diagonal elements are extracted and structured to construct a basic noise covariance matrix. Each element on the diagonal of this matrix corresponds to the fundamental perturbation intensity of each core state variable in the state space, while the off-diagonal elements are set to zero to maintain the independence assumption between noise sources. The adaptive coefficients are applied to the overall structure of this basic matrix as scaling factors, and the dynamic process noise covariance matrix at the current time point is generated through matrix scaling. To enhance the system's ability to differentiate between different state variables, mechanism-sensitive mappings are introduced for different state dimensions. Specifically, some state dimensions are affected only by specific mechanisms. Consequently, a structural adjustment term is introduced, multiplying certain elements of the basic matrix by mechanism-specific function coefficients. This significantly amplifies the perturbation propagation capability when a specific mechanism dominates, while contracting the state diffusion channel when the mechanism is weakened. To ensure dynamic adaptability and responsiveness in this process, a noise matrix generation function was constructed. This function takes a normalized probability vector, a mechanism weight table, adaptive coefficients, and a base matrix as input variables. Upon invocation, it returns the process noise covariance matrix to be used at the current moment. This matrix is ​​used to calculate the prediction covariance and adjust the Kalman gain of the hybrid Kalman filter. Internal logic within the function determines whether the dominance of the mechanism has substantially changed and whether an update of the state perturbation model is necessary. This avoids the surge in system overhead or unstable state estimates caused by frequent updates.To this end, an update trigger condition is introduced. This condition monitors the magnitude of change in the probability distribution of the degradation mechanism between two consecutive time steps. If this magnitude of change (such as the Euclidean distance or the maximum element difference) exceeds a preset threshold, it is considered that the battery state has undergone a significant evolution, thereby triggering a new round of calculations in the noise matrix generation function. Specifically, the change trigger threshold is set to a fixed range. When the probability change of any mechanism exceeds this range, or the deviation of the overall structure of the standardized probability distribution exceeds the set distance, the re-evaluation process is initiated and a new process noise covariance matrix is ​​output as the Q matrix input to the next round of prediction equations of the filter. If the magnitude of the change is below the threshold, the system maintains the covariance matrix generated in the previous round unchanged, avoiding estimation oscillations caused by frequent switching of noise models during the stable state phase.

[0044] In a specific embodiment, executing the intelligent monitoring method for new energy vehicle batteries further includes the following steps: A multi-dimensional evaluation system for battery abnormality is constructed based on the battery state of charge estimation value and the micro-degradation state determination results, and abnormality level classification standards are set based on the multi-dimensional evaluation system for battery abnormality; Perform real-time monitoring of the battery state of charge estimation value and microscopic degradation state determination results according to the abnormality level classification standard to obtain the current abnormality level determination result of the battery; Under the condition that communication is normal and sensor data is reliable, a state feedback gain matrix is ​​constructed based on the battery dynamic equivalent model to obtain the control law under the model control mode; When communication interruption or data anomaly is detected and the battery's current abnormality level exceeds the preset alarm level, current limiting, active balancing, or emergency power-off measures are activated according to the abnormality type to obtain a protection strategy under the rule control mode; By defining the switching function to construct a sliding mode controller and setting the switching conditions, when the system switches from the model control mode to the rule control mode, the estimation error is guaranteed to meet the preset convergence conditions, and a dual-mode switching battery intelligent protection control system is obtained.

[0045] Specifically, based on battery state estimation and degradation mechanism identification, several key indicators are extracted to construct a multidimensional abnormality judgment vector. This vector must simultaneously reflect the battery's energy storage state, electrochemical activity level, internal structural integrity, and proximity to the safe operating boundary. The system uses the estimated SOC value, SOC estimation residual amplitude, internal resistance growth rate, RC branch response time drift, degradation mechanism probability intensity, impedance spectrum deviation amplitude, surface temperature gradient, and voltage and current anomaly factors as its primary input dimensions to construct a well-structured state indicator matrix. Using a multidimensional mapping function or normalization method, each indicator is normalized to an abnormality degree indicator within a uniform numerical range, which is then combined into a unified abnormal state evaluation vector. Based on this evaluation vector, an abnormality classification standard is designed, including level 1 (minor abnormality), level 2 (moderate abnormality), level 3 (severe abnormality), and level 4 (urgent abnormality). Numerical thresholds and multi-indicator linkage judgment logic are set for each level. Based on the set classification criteria, the state data within the current sampling period is judged in real time. Combining the output of the prediction model with sensor feedback, a current abnormality level indicator is quickly generated, providing a logical entry point for control mode selection. Under conditions where the abnormality level is low, the communication link is unobstructed, and the sensors are operating stably, the system prioritizes a model-based control mode driven by a dynamic equivalent model to manage battery energy and regulate operating conditions. In this mode, a linear or nonlinear feedback control law is constructed using the relationship between the state vector defined in the state-space model and the observed output. To this end, the state error is calculated based on the deviation between the current state estimate and the target reference trajectory (such as desired SOC or optimal polarization voltage). A state feedback gain matrix is ​​then generated using robust control design methods (such as LQR or minimum gain control). This gain matrix is ​​used to calculate the control input correction term, dynamically adjusting the charge and discharge current or power scheduling strategy to achieve closed-loop regulation of the system state, ensuring that the battery operates within the optimal and safe operating range. This mode offers high precision and low error response, making it suitable for normal operation and mild degradation. If the system detects a communication anomaly, missing sensor data, or the anomaly level reaches or exceeds a set alarm threshold, the model control mechanism is terminated and the system automatically switches to a rule-based control mode to ensure battery safety. In this rule-based control mode, the system no longer relies on model feedback predictions, but instead quickly activates predefined safety protection strategies based on the current anomaly type.Specifically, if the monitored current exceeds the set safety current limit and is accompanied by an increased probability of active material loss, the current limiting mechanism is activated, dynamically reducing the maximum allowable charge and discharge current. If uneven voltage distribution and a continuously increasing voltage differential between cells are detected, an active balancing strategy is implemented, adjusting the load of individual cells through bypass circuits to maintain balance within the group. If the system enters an extreme state, such as when any of the voltage, current, or temperature indicators triggers a power-off threshold, the system initiates an emergency power-off protection, directly shutting off the power supply by controlling a high-voltage relay or switch module, thereby forcibly protecting the battery and the entire vehicle system. To achieve efficient switching between model-based and rule-based control and ensure smooth transitions between system states, a sliding mode controller is designed, and a state switching function is defined as the mode transition trigger. The sliding mode controller uses the system state error as input to construct a switching plane. During operation, the system continuously determines whether the current state error enters the sliding mode range. When the error gradually approaches the set limit, the abnormality level remains high, or a communication failure persists, the mode switch condition is triggered, and the system switches from model-based control to rule-based control. During the sliding-mode switching process, the system smoothes the system by controlling the ratio between the error derivative and the state transition term. This ensures that the system state does not suddenly change at the moment of switching, maintains a continuous estimation process, and prevents drastic fluctuations in the control input from causing current surges, voltage spikes, and other hazards. The system also applies an exponential convergence function to the sliding-mode trajectory, ensuring that the state error decays to an acceptable range at a steady rate, ultimately completing the mode switch and entering a stable, regular control trajectory.

[0046] The above describes the intelligent monitoring method for new energy vehicle batteries in the embodiment of the present invention. The following describes the intelligent monitoring system for new energy vehicle batteries in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an intelligent monitoring system for new energy vehicle batteries includes: An acquisition module 201 is used to acquire a target battery operation data set of a new energy vehicle battery and extract time-domain differential features and frequency-domain energy loss features during the battery charge and discharge process to obtain a multi-dimensional state feature set; The electrochemical characteristic analysis module 202 is used to analyze the electrochemical characteristics of the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; The generating module 203 is used to dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

[0047] Through the collaborative cooperation of the above components, through a high-precision sensor array and a three-stage filtering network, the data quality of battery status monitoring is significantly improved, sampling noise and pulse interference are eliminated, and the continuity and integrity of the data are achieved. In combination with Shannon entropy calculation and Laplace transform, the time domain differential characteristics and frequency domain heat loss characteristics are extracted in a coordinated manner, fully capturing the nonlinear change trend of battery capacity decay and the energy loss characteristics at different frequencies. Through the recursive least squares parameter identification algorithm and parameter correction coefficient matrix, a more accurate battery dynamic equivalent model is constructed, overcoming the problem of nonlinear changes in battery polarization characteristics and adapting to various complex working conditions. By utilizing electrochemical impedance spectroscopy time domain mapping and complex domain convolution operations, combined with a deep learning framework, the internal microscopic degradation mechanism of the battery is accurately identified. The introduction of a variable process noise covariance matrix and a measurement noise adaptive adjustment mechanism based on residual analysis, combined with charge and discharge inflection point recalibration, solves the problems of insufficient estimation accuracy and cumulative error in traditional methods. Through a multi-level fault detection mechanism and sliding mode controller, a smooth transition between model control and rule control is achieved, ensuring that the system can still maintain basic protection functions in the event of communication interruption or data anomaly, significantly improving the safety and reliability of battery use.

[0048] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0049] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling an intelligent monitoring device for new energy vehicle batteries (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0050] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent monitoring method for new energy vehicle batteries, characterized in that: include: Obtain a target battery operation dataset for new energy vehicle batteries, extract the time-domain differential features and frequency-domain energy loss features during the battery charging and discharging process, and obtain a multidimensional state feature set; Performing electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; Dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

2. The intelligent monitoring method for new energy vehicle batteries according to claim 1, characterized in that: The target battery operation data set of the new energy vehicle battery is obtained, and the time domain differential characteristics and frequency domain energy loss characteristics of the battery during charging and discharging are extracted. Get a multi-dimensional state feature set, including: Collect the single cell voltage, charge and discharge current, surface temperature distribution and ambient temperature of new energy vehicle batteries to obtain the original battery parameter data; Performing denoising processing on the original battery parameter data to obtain battery signal data from which sampling noise has been removed, and performing high-frequency interference processing on the battery signal data from which sampling noise has been removed to obtain battery signal data from which high-frequency interference has been eliminated; Reconstructing missing points on the battery signal data from which high-frequency interference has been eliminated to obtain continuous and complete battery signal data, and performing time synchronization and standardization processing on the continuous and complete battery signal data to obtain a target battery operation data set; The time domain differential features and frequency domain energy loss features of the battery during charging and discharging are extracted according to the target battery operation data set to obtain a multi-dimensional state feature set.

3. The intelligent monitoring method for new energy vehicle batteries according to claim 2, characterized in that: The time domain differential features and frequency domain energy loss features of the battery during charging and discharging are extracted according to the target battery operation data set. Get a multi-dimensional state feature set, including: performing differential calculation on the voltage and current data in the target battery operation data set to obtain voltage-capacity differential data; Performing grid discretization processing on the voltage-capacity differential data to obtain a differential characteristic matrix; Using the Shannon entropy calculation function to quantify the characteristic distribution of the differential characteristic matrix to obtain a time domain characteristic vector; Performing Laplace transform on the current signal in the target battery operation data set to obtain current spectrum data, and calculating the impedance value of the battery equivalent internal resistance at a preset characteristic frequency point based on the current spectrum data to obtain a frequency domain eigenvector; The time domain feature vector and the frequency domain feature vector are fused and tensor decomposition is performed to obtain a multi-dimensional state feature set.

4. The intelligent monitoring method for new energy vehicle batteries according to claim 1, characterized in that: The electrochemical characteristics analysis of the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result includes: Constructing initial values ​​of battery polarization characteristic parameters including open circuit voltage, ohmic internal resistance, and two RC parallel networks based on the multidimensional state feature set; Setting a prediction error function for the initial value of the battery polarization characteristic parameter and performing iterative calculations using a recursive least squares algorithm to obtain a parameter vector iterative update formula; Generate a parameter correction coefficient matrix based on the parameter vector iterative update formula combined with the current battery state of charge and temperature value to obtain battery equivalent circuit parameters that are adjusted in real time according to the operating conditions; Based on the battery equivalent circuit parameters adjusted in real time according to the operating conditions, a state vector equation including the state of charge and polarization voltage is established to obtain a battery state space model structure; Setting calculation rules of the state prediction equation and the system output equation according to the battery state space model structure to obtain a battery dynamic equivalent model; The battery dynamic equivalent model is used to analyze the electrochemical characteristics of the new energy vehicle battery to obtain a microscopic degradation state determination result.

5. The intelligent monitoring method for new energy vehicle batteries according to claim 4, characterized in that: The electrochemical characteristics analysis of the new energy vehicle battery using the battery dynamic equivalent model to obtain a microscopic degradation state determination result includes: Applying a small AC excitation signal at a preset logarithmically divided frequency point based on the battery dynamic equivalent model to obtain battery voltage response time domain data; Performing a complex domain convolution operation on the battery voltage response time domain data to obtain frequency domain electrochemical impedance spectroscopy data; constructing an impedance characteristic vector based on the frequency-domain electrochemical impedance spectroscopy data and calculating the difference from a standard impedance spectrum of a healthy battery to obtain an impedance difference characteristic vector; Constructing a battery degradation feature dataset based on the impedance difference feature vector and a circuit parameter time series in the battery dynamic equivalent model; Input the battery degradation feature dataset into the bidirectional LSTM layer and the multi-head self-attention mechanism to perform probability distribution calculation to obtain the degradation mechanism probability distribution; According to the probability distribution of the degradation mechanism, the microscopic degradation state judgment results of six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion and electrode structure collapse are determined.

6. The intelligent monitoring method for new energy vehicle batteries according to claim 5, characterized in that: The determination of the microscopic degradation state results of the six degradation mechanisms including SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion and electrode structure collapse according to the degradation mechanism probability distribution includes: Performing threshold screening based on the degradation mechanism probability distribution to obtain a set of candidate dominant degradation mechanisms, wherein the candidate dominant degradation mechanism set includes six degradation mechanisms: SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse; Calculating the relative contribution ratio of each dominant degradation mechanism based on the set of candidate dominant degradation mechanisms and sorting them to obtain a degradation mechanism sequence sorted by importance; assigning a life decay model in the form of an exponential-logarithmic composite function, a quadratic function, or an exponential function to the degradation mechanism ranked first in the sequence of degradation mechanisms sorted by importance; The lifetime decay model is used to perform parameter fitting calculations to obtain degradation trend prediction parameters, and the remaining number of cycles and service life are calculated based on the degradation trend prediction parameters and the current battery state to obtain a microscopic degradation state determination result.

7. The intelligent monitoring method for new energy vehicle batteries according to claim 6, characterized in that: The dynamically adjusting the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generating a battery state of charge estimation value includes: Constructing a prediction and update equation group of a hybrid Kalman filter based on the state space equations in the battery dynamic equivalent model; Creating a variable process noise covariance matrix update rule according to the probability values ​​of different degradation mechanisms in the microscopic degradation state determination result; Perform sliding window analysis on the measurement residual sequence generated during the filtering process and calculate the residual covariance to obtain the measurement noise adaptive adjustment matrix; Executing prediction and update iterative calculations of a hybrid Kalman filter based on the variable process noise covariance matrix update rule and the measurement noise adaptive adjustment matrix to obtain an initial estimate of the battery state of charge; By monitoring the inflection point characteristics of the battery voltage-current curve, the static stage and the fully charged stage are identified, and the state of charge recalibration time point is obtained; The initial estimated value of the battery state of charge is corrected at the state of charge recalibration time point to obtain an estimated value of the battery state of charge.

8. The intelligent monitoring method for new energy vehicle batteries according to claim 7, characterized in that: The step of creating a variable process noise covariance matrix update rule according to the probability values ​​of different degradation mechanisms in the microscopic degradation state determination result includes: Normalizing the probability values ​​of the six degradation mechanisms of SEI film growth, active material loss, lithium precipitation, electrolyte decomposition, current collector corrosion, and electrode structure collapse in the microscopic degradation state determination results to obtain a standardized probability distribution of each degradation mechanism; Based on the standardized probability distribution, a mapping relationship of weight coefficients corresponding to different degradation mechanisms is designed to obtain a mechanism weight table; Calculating the adaptive coefficient according to the mechanism weight table to obtain the adaptive coefficient reflecting the comprehensive influence of multiple degradation mechanisms, and setting the diagonal elements of the initial process noise covariance matrix to obtain the basic noise covariance matrix; Creating a noise matrix generating function that is dynamically adjusted according to the degradation state based on the adaptive coefficient and the basic noise covariance matrix; An update trigger condition is set for the noise matrix generation function, and an update calculation is performed when the variation range of the degradation mechanism probability distribution exceeds a preset target value to obtain an update rule for the variable process noise covariance matrix.

9. The intelligent monitoring method for new energy vehicle batteries according to claim 1, characterized in that: The intelligent monitoring method for new energy vehicle batteries also includes: Constructing a multi-dimensional evaluation system for battery abnormality based on the battery state of charge estimation value and the microscopic degradation state determination result, and setting an abnormality level classification standard based on the multi-dimensional evaluation system for battery abnormality; Performing real-time monitoring on the battery state of charge estimation value and the microscopic degradation state determination result according to the abnormality level classification standard to obtain a current abnormality level determination result of the battery; Under the condition that communication is normal and sensor data is reliable, a state feedback gain matrix is ​​constructed based on the battery dynamic equivalent model to obtain a control law in a model control mode; When a communication interruption or data anomaly is detected and the battery's current abnormality level exceeds a preset alarm level, current limiting, active balancing, or emergency power-off measures are activated according to the abnormality type to obtain a protection strategy under the rule control mode; By defining the switching function to construct a sliding mode controller and setting the switching conditions, when the system switches from the model control mode to the rule control mode, the estimation error is guaranteed to meet the preset convergence conditions, and a dual-mode switching battery intelligent protection control system is obtained.

10. An intelligent monitoring system for new energy vehicle batteries, characterized in that: For implementing the intelligent monitoring method for a new energy vehicle battery according to any one of claims 1 to 9, the intelligent monitoring system for a new energy vehicle battery comprises: The acquisition module is used to obtain the target battery operation data set of the new energy vehicle battery and extract the time domain differential features and frequency domain energy loss features during the battery charging and discharging process to obtain a multidimensional state feature set; an electrochemical characteristic analysis module, configured to perform electrochemical characteristic analysis on the new energy vehicle battery based on the multi-dimensional state feature set to obtain a microscopic degradation state determination result; A generation module is used to dynamically adjust the parameter configuration of the hybrid Kalman filter according to the microscopic degradation state determination result and generate a battery state of charge estimation value.

Images