New energy automobile charging safety regulation and control method and system

Abstract

The invention relates to the technical field of charging control, in particular to a new energy automobile charging safety regulation and control method and system, and the method comprises the steps: collecting a multi-dimensional feature parameter set; acquiring battery health state and charge state information, and collecting multi-point temperature information on the surface of the battery; constructing the multi-dimensional feature parameter set into a multi-dimensional time sequence, performing statistical analysis on the multi-dimensional time sequence, and extracting dynamic feature indexes; performing structured fusion on the dynamic characteristic indexes, thermal field gradient parameters calculated according to multi-point temperature information and a multi-dimensional characteristic parameter set, and projecting the dynamic characteristic indexes, thermal field gradient parameters and the multi-dimensional characteristic parameter set to a key characteristic subspace through principal component analysis to obtain a charging response vector; generating a multi-dimensional security hyperplane through the dynamic security boundary model; continuously comparing the charging response vector with the multi-dimensional safety hyperplane, and calculating a safety margin scalar in real time; and inputting the safety margin scalar into the self-adaptive regulation and control model, and adjusting the charging current output. According to the invention, active predictive charging safety regulation and control are realized through multi-dimensional perception.

Description

Technical Field

[0001] This invention relates to the field of charging control technology, specifically to a method and system for safe regulation of charging of new energy vehicles. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the charging safety of power batteries has become a focus of industry attention. Currently, the mainstream charging control strategy for new energy vehicles is the constant current-constant voltage (CC-CV) mode. In this mode, charging safety mainly relies on the battery management system (BMS) to monitor the key physical parameters of the battery pack.

[0003] Existing safety monitoring methods typically set fixed protection thresholds, such as a total voltage limit, the maximum allowable temperature of a single cell, and the maximum charging current. During charging, the BMS collects the instantaneous values ​​of these parameters in real time. Once any parameter is detected to exceed the preset safety threshold, a protection mechanism is immediately triggered, such as reducing the charging current or directly cutting off the charging circuit, to prevent dangerous situations such as overcharging or overheating of the battery.

[0004] However, this passive protection strategy based on fixed thresholds typically sets its safety margin conservatively based on the worst operating conditions throughout the battery's lifespan, making it difficult to balance charging efficiency with the actual safety boundaries of batteries at different stages of aging. Furthermore, this type of method relies primarily on monitoring macroscopic, isolated parameters, offering limited sensitivity to the complex, dynamically changing electrochemical states within the battery. It usually only reacts after safety hazards have already manifested, lacking the ability for early warning and proactive regulation.

[0005] Therefore, a method and system for regulating the charging safety of new energy vehicles is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for safe charging control of new energy vehicles, which achieves proactive predictive charging safety control through multi-dimensional sensing. The method includes: collecting a multi-dimensional feature parameter set; acquiring battery health and state of charge information, and collecting multi-point temperature information on the battery surface; constructing a multi-dimensional time series from the multi-dimensional feature parameter set, performing statistical analysis on the multi-dimensional time series, and extracting dynamic feature indicators; structurally fusing the dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and the multi-dimensional feature parameter set, and projecting them onto a key feature subspace through principal component analysis to obtain a charging response vector; generating a multi-dimensional safety hyperplane through a dynamic safety boundary model; continuously comparing the charging response vector with the multi-dimensional safety hyperplane to calculate a safety margin scalar in real time; and inputting the safety margin scalar into an adaptive control model to adjust the charging current output.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for controlling the charging safety of new energy vehicles, comprising:

[0009] An excitation pulse is injected into the charging circuit, and a multi-dimensional feature parameter set triggered by the excitation pulse is collected; battery health status and state of charge information are obtained through the vehicle communication protocol, and multi-point temperature information on the battery surface is collected.

[0010] The multidimensional feature parameter set is constructed into a multidimensional time series, and statistical analysis is performed on the multidimensional time series to extract dynamic feature indicators that characterize short-term volatility and changing trends.

[0011] By constructing a state-trend fusion matrix, the current values ​​of dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and multi-dimensional feature parameter sets are structurally fused. The matrix is ​​then projected onto the key feature subspace using principal component analysis to obtain the charging response vector.

[0012] The battery health status and state of charge information are input into the dynamic safety boundary model to generate a multidimensional safety hyperplane; the charging response vector is continuously compared with the multidimensional safety hyperplane to calculate the safety margin scalar in real time; the safety margin scalar is input into the adaptive control model to dynamically adjust the charging current output.

[0013] Preferably, the process of acquiring a multi-dimensional feature parameter set triggered by an excitation pulse includes:

[0014] At the output end of the charging pile, the complete voltage transient response curve triggered by the excitation pulse is captured by high-frequency sampling. The voltage transient response curve is analyzed. The ohmic internal resistance is calculated by analyzing the initial vertical jump part of the curve. The polarization internal resistance is calculated by analyzing the nonlinear gradual change part of the rising stage of the curve. The relaxation time constant is calculated by analyzing the decay rate of the voltage fall-off stage after the excitation pulse ends. The voltage transient response curve is then subjected to a fast Fourier transform to extract the impedance amplitude and phase angle at a preset characteristic frequency point as frequency domain characteristic parameters.

[0015] Preferably, the process of collecting multi-point temperature information on the battery surface includes: simultaneously collecting temperature readings at multiple locations using a temperature sensor array arranged on the surface of the battery module, and obtaining the thermal gradient parameter by calculating the difference between the maximum and minimum values ​​of the temperature readings.

[0016] Preferably, the process of obtaining the charging response vector includes:

[0017] Within a preset time window, the time series of the multi-dimensional feature parameter set is calculated to obtain the time change rate as a dynamic feature index; a two-dimensional state trend fusion matrix is ​​constructed, where the rows of the matrix correspond to each physical quantity in the multi-dimensional feature parameter set, and the columns of the matrix correspond to the current instantaneous value and dynamic feature index of each physical quantity, respectively; the thermal field gradient parameters calculated based on multi-point temperature information are added to the matrix as additional dimension information; principal component analysis is performed on the state trend fusion matrix, and the top N principal component vectors whose cumulative variance contribution rate reaches the threshold are selected; the state trend fusion matrix is ​​projected onto the feature subspace spanned by the N principal component vectors to obtain a set of N-dimensional principal component scores; the N principal component scores are arranged in order of importance of the corresponding principal components to form a charging response vector.

[0018] Preferably, the dynamic safety boundary model includes:

[0019] The input fuzzification layer receives battery health status and state of charge information as input variables, and maps the input variables to multiple fuzzy language sets through multiple built-in membership functions;

[0020] The fuzzy inference layer contains a fuzzy inference rule base generated by expert knowledge and data training. The trigger strength of each rule is calculated based on the fuzzy language set and normalized.

[0021] The output layer performs weighted calculations on the output conclusions of each rule based on the normalized trigger strength to obtain the maximum allowable charging current, the highest allowable terminal voltage, and the highest allowable surface temperature of the safety threshold, and constructs a multi-dimensional safety hyperplane.

[0022] Preferably, the calculation process of the safety margin scalar includes:

[0023] The multidimensional safety hyperplane is defined as the boundary of a repulsive potential field that applies a virtual repulsive force. Based on the Euclidean distance from the charging response vector to the boundary of the repulsive potential field, a position potential energy scalar determined by geometric distance is calculated. Based on the dynamic characteristic indices in the charging response vector, a state evolution velocity vector representing the direction and rate of system state evolution is constructed in the state space. The projection of the state evolution velocity vector onto the normal direction of the boundary of the repulsive potential field is calculated to obtain a kinetic energy risk scalar determined by the state approach rate. The position potential energy scalar and the kinetic energy risk scalar are input into a preset risk fusion model for nonlinear superposition to generate the safety margin scalar.

[0024] Preferably, the adaptive control model includes:

[0025] The error signal generation unit receives the safety margin scalar, compares it with the internally set target safety margin value in real time, and uses the difference between the two as the error signal.

[0026] The control component parallel processing unit inputs the error signal into three parallel mathematical operation channels; the proportional channel directly performs linear scaling on the error signal, the integral channel performs time accumulation operation on the error signal, and the differential channel calculates the rate of change of the error signal, and synchronously outputs three control components representing the current, historical and future predictions respectively.

[0027] The dynamic gain scheduling unit uses the externally input battery health status and temperature information as query coordinates, performs interpolation calculations on the preset multi-dimensional gain parameter surface, and outputs the proportional, integral, and differential gain coefficients with respect to the current actual battery operating conditions.

[0028] The instruction weighted synthesis unit receives the control component and the gain coefficient, multiplies the control component by the corresponding gain coefficient one by one and sums them to obtain the control output value, and converts the control output value into a specific adjustment instruction for the charging current.

[0029] A new energy vehicle charging safety control system includes:

[0030] The data acquisition module injects excitation pulses into the charging circuit and collects a set of multi-dimensional feature parameters triggered by the excitation pulses; it also obtains battery health status and state of charge information through the vehicle communication protocol and collects multi-point temperature information on the battery surface.

[0031] The feature extraction module constructs the multi-dimensional feature parameter set into a multi-dimensional time series, performs statistical analysis on the multi-dimensional time series, and extracts dynamic feature indicators that characterize short-term volatility and changing trends.

[0032] The feature mapping module constructs a state trend fusion matrix to structurally fuse dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and the current values ​​of multi-dimensional feature parameter sets. The matrix is ​​then projected onto the key feature subspace using principal component analysis to obtain the charging response vector.

[0033] The current adjustment module inputs the battery health status and state of charge information into the dynamic safety boundary model to generate a multi-dimensional safety hyperplane; it continuously compares the charging response vector with the multi-dimensional safety hyperplane to calculate the safety margin scalar in real time; and it inputs the safety margin scalar into the adaptive control model to dynamically adjust the charging current output.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. The dynamic safety boundary model employed in this invention takes into account the battery's health and state of charge to generate a safety boundary adapted to the battery's current state. This approach helps to set charging parameters more rationally while ensuring safety, thereby improving the balance between charging efficiency and safety.

[0036] 2. This invention, by introducing dynamic characteristic indicators and a risk assessment method based on potential field theory, increases the consideration of system state change trends. This helps to identify potential risk evolution trends before key parameters reach their limits, providing an earlier window for the system to take control measures.

[0037] 3. This invention enhances the characterization of battery state by introducing frequency domain characteristic parameters and thermal field gradient parameters. This enables the method to monitor internal electrochemical state changes and uneven temperature distributions that are difficult to observe using traditional methods, thereby improving the ability to identify specific types of potential risks. Attached Figure Description

[0038] Figure 1 A flowchart illustrating a method for controlling the charging safety of new energy vehicles, provided in an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of the hyperplane provided in an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of a new energy vehicle charging safety control system provided in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Please see Figures 1 to 3 This invention provides a method and system for controlling the charging safety of new energy vehicles, the technical solution of which is as follows:

[0043] Example 1:

[0044] A method for controlling the charging safety of new energy vehicles, the specific process is as follows: Figure 1 As shown, it includes:

[0045] An excitation pulse is injected into the charging circuit, and a multi-dimensional feature parameter set triggered by the excitation pulse is collected; battery health status and state of charge information are obtained through the vehicle communication protocol, and multi-point temperature information on the battery surface is collected.

[0046] The multidimensional feature parameter set is constructed into a multidimensional time series, and statistical analysis is performed on the multidimensional time series to extract dynamic feature indicators that characterize short-term volatility and changing trends.

[0047] By constructing a state-trend fusion matrix, the current values ​​of dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and multi-dimensional feature parameter sets are structurally fused. The matrix is ​​then projected onto the key feature subspace using principal component analysis to obtain the charging response vector.

[0048] The battery health status and state of charge information are input into the dynamic safety boundary model to generate a multidimensional safety hyperplane; the charging response vector is continuously compared with the multidimensional safety hyperplane to calculate the safety margin scalar in real time; the safety margin scalar is input into the adaptive control model to dynamically adjust the charging current output.

[0049] Furthermore, the process of acquiring the multi-dimensional feature parameter set triggered by the excitation pulse includes:

[0050] At the output end of the charging pile, the complete voltage transient response curve triggered by the excitation pulse is captured by high-frequency sampling. The voltage transient response curve is analyzed. The ohmic internal resistance is calculated by analyzing the initial vertical jump part of the curve. The polarization internal resistance is calculated by analyzing the nonlinear gradual change part of the rising stage of the curve. The relaxation time constant is calculated by analyzing the decay rate of the voltage fall-off stage after the excitation pulse ends. The voltage transient response curve is then subjected to a fast Fourier transform to extract the impedance amplitude and phase angle at a preset characteristic frequency point as frequency domain characteristic parameters.

[0051] Specifically, an excitation pulse is superimposed onto the charging circuit. In this embodiment, the excitation pulse is a constant current square wave pulse with an amplitude of 5% of the main charging current and a duration of 100 milliseconds to ensure that this diagnostic step does not significantly interfere with the normal charging process. During the entire cycle of excitation pulse injection and subsequent relaxation, the terminal voltage of the charging circuit is synchronously sampled at a high sampling frequency of 5 kHz to capture the complete voltage transient response curve triggered by the excitation pulse, including the rising and falling edges. This curve is recorded in the form of high-resolution time series data for subsequent analysis and processing.

[0052] In addition, the excitation pulse can also utilize an adaptive composite excitation signal, which is a composite current signal formed by the linear superposition of sine waves of multiple key frequencies. In the offline stage, a characteristic frequency database of the operating condition is established through experiments, and a mapping model of the K-nearest neighbor algorithm is trained based on the database. During actual charging, the real-time battery health status, state of charge, and temperature values ​​are input into the model, and the model will output a set of the most suitable advantageous frequency points for diagnosis. The sine waves of these frequency points are digitally synthesized, and the peak-to-average power ratio of the synthesized signal is minimized through a phase optimization algorithm to ensure that the total peak current is strictly controlled between 1% and 5% of the main charging current. The signal is finally accurately superimposed on the main charging current through the power control module of the charging device.

[0053] Based on the acquired voltage transient response curves, multi-dimensional characteristic parameters are calculated through a series of analytical steps. First, by analyzing the step change in voltage at the instant of applying a current pulse, such as at time t0, the ohmic internal resistance reflecting the pure resistive characteristics of the battery is calculated. Second, the nonlinear slowly varying part of the voltage during the pulse duration is analyzed. This part mainly reflects the charge transfer and double-layer effect in the electrochemical reaction. By modeling this process, the polarization internal resistance is obtained by quantification.

[0054] After the excitation pulse ends, the decay curve of the terminal voltage gradually decreasing is analyzed. By fitting the rate of change of the curve with an exponential function, the relaxation time constant is obtained. This constant characterizes the reaction speed of the electrochemical polarization process.

[0055] Finally, to obtain deeper state information, the collected voltage and current time series were subjected to fast Fourier transform. By calculating the relationship between the two in the frequency domain, the electrochemical impedance spectrum of the battery was obtained. To meet the real-time requirements, only two preset characteristic frequency points, such as the impedance amplitudes Z_1000 and Z_1 at 1000 Hz and 1 Hz, and the phase angles phi_1000 and phi_1, were extracted from the spectrum as frequency domain characteristic parameters that can characterize the battery interface state and reaction kinetics.

[0056] A multidimensional fingerprint of the battery's electrochemical state can be obtained using a single small perturbation. By combining time-domain analysis (distinguishing between ohmic and polarization resistance) and frequency-domain analysis, different internal processes can be decoupled more clearly. This makes it possible to assess the state of deep characteristics such as battery interface reactions and ion migration, and provides richer feature evidence for identifying early signs of aging.

[0057] Furthermore, the process of collecting multi-point temperature information on the battery surface includes: simultaneously collecting temperature readings at multiple locations using a temperature sensor array arranged on the surface of the battery module, and obtaining the thermal field gradient parameter by calculating the difference between the maximum and minimum values ​​of the temperature readings.

[0058] Specifically, a temperature sensor array consisting of eight NTC thermistors is pre-positioned on the surface of the battery module. The sensors are strategically positioned to cover key areas of the module, such as near the main electrode terminals and evenly distributed on the battery surface at the center and corners of the module, ensuring effective capture of the module's temperature distribution. During charging, the sensor array is periodically sampled at a frequency of 1Hz, and at each sampling point, the real-time temperature readings of all eight temperature sensors are simultaneously acquired, collectively forming a temperature snapshot characterizing the current thermal field distribution.

[0059] An optimization calculation is performed on the temperature snapshot data obtained at each sampling time point. This calculation process traverses all temperature values ​​in the dataset to find the highest and lowest temperature readings. Then, by subtracting the highest and lowest temperature readings, a temperature difference is obtained. This difference is the thermal field gradient parameter at the current time, which is used in the subsequent charging response vector construction step.

[0060] By introducing the thermal field gradient parameter, spatial dimension safety monitoring is added. Compared with traditional single-point temperature measurement, this parameter directly quantifies the uniformity of battery heat distribution. Since local overheating is a key precursor to thermal runaway, this method can identify local abnormal risks such as internal micro-short circuits earlier than overall temperature exceeding limits, thus improving the early warning capability and precision of thermal safety management.

[0061] Furthermore, the process of obtaining the charging response vector includes:

[0062] Within a preset time window, the time series of the multi-dimensional feature parameter set is calculated to obtain the time change rate as a dynamic feature index; a two-dimensional state trend fusion matrix is ​​constructed, where the rows of the matrix correspond to each physical quantity in the multi-dimensional feature parameter set, and the columns of the matrix correspond to the current instantaneous value and dynamic feature index of each physical quantity, respectively; the thermal field gradient parameters calculated based on multi-point temperature information are added to the matrix as additional dimension information; principal component analysis is performed on the state trend fusion matrix, and the top N principal component vectors whose cumulative variance contribution rate reaches the threshold are selected; the state trend fusion matrix is ​​projected onto the feature subspace spanned by the N principal component vectors to obtain a set of N-dimensional principal component scores; the N principal component scores are arranged in order of importance of the corresponding principal components to form a charging response vector.

[0063] Specifically, in an offline preparation phase, a principal component analysis projection model with generalization capabilities needs to be established. This process begins with collecting a large-scale, multi-condition battery feature dataset, which should cover different battery health states, states of charge, and ambient temperatures. For each historical data point recorded in this dataset, the following operations are performed: the time change rate of its multi-dimensional feature parameter set is calculated as a dynamic feature index; these dynamic feature indices, along with the corresponding instantaneous parameter values, thermal gradient parameters, and their change rates, are used to construct a structured state trend fusion matrix. Tens of thousands of such matrix samples generated from all historical data points are aggregated to form a training database. Subsequently, global principal component analysis is performed on this database to calculate all principal component vectors that can represent the main change directions of the entire dataset. Based on their variance contribution rate, the top N most important principal component vectors that can explain 99% of the data changes are selected. These N principal component vectors ultimately form a fixed projection transformation matrix, which is then stored in the memory of the charging control device. Before performing principal component analysis, in order to eliminate the influence of differences in units and dimensions of different physical quantities on the analysis results, each row of the state trend fusion matrix (i.e., corresponding to each physical quantity) needs to be Z-score standardized.

[0064] In the actual real-time charging control process, an online calculation step is performed. In each calculation cycle, for example, every 5 seconds, the latest multi-dimensional feature parameter set and thermal field gradient parameters are first obtained. Then, within a preset time window, for example, including the 5 most recent consecutive data points, the current time change rate of each parameter is calculated using methods such as linear regression, as a real-time dynamic feature indicator.

[0065] Then, these latest instantaneous values, the real-time calculated dynamic characteristic indicators, and the thermal gradient parameters and their rate of change are used to construct a two-dimensional state trend fusion matrix for the current moment. Each row of this matrix corresponds to a physical quantity (such as ohmic internal resistance, thermal gradient, etc.), the first column is filled with the instantaneous value of the physical quantity, and the second column is filled with its dynamic characteristic indicators.

[0066] Finally, the projection transformation matrix generated and stored in the offline stage is invoked. Through a matrix multiplication operation, the current state trend fusion matrix is ​​projected onto the feature subspace composed of N principal component vectors. The result of this projection operation is a set of N-dimensional projection coordinate values, which are the principal component scores at the current moment. These N principal component scores are arranged in descending order of importance (i.e., variance contribution rate) of their corresponding principal components. The resulting N-dimensional vector is defined as the compact and information-dense charging response vector at the current moment.

[0067] Before constructing the state-trend fusion matrix, a multi-source data cross-validation step is added. A pre-trained offline Kalman filter is built in, which predicts the theoretical range of thermal gradient parameters based on real-time values ​​and rates of change of ohmic resistance, charging current, and ambient temperature. In each calculation cycle, the actual collected thermal gradient parameters are compared with the theoretical range predicted by the model. If the actual values ​​continuously deviate from the theoretical prediction range and exceed a preset threshold for several consecutive cycles, the temperature sensor array is deemed faulty. The weight of the thermal gradient parameters in subsequent principal component analysis is proactively reduced, and a sensor maintenance recommendation is reported to the vehicle's diagnostic system. This mechanism ensures that an anomaly of a single sensor will not seriously interfere with the final safety margin assessment, significantly improving the fault tolerance of the entire control system.

[0068] By constructing a state matrix that integrates instantaneous values, rates of change, and multi-physics information, and using principal component analysis for dimensionality reduction and feature extraction, the beneficial effect is the generation of a charging response vector with high information density and strong robustness. This not only comprehensively characterizes the battery state but also helps remove redundancy and noise from the data, providing high-quality input for subsequent accurate and reliable risk assessment.

[0069] Furthermore, the dynamic security boundary model includes:

[0070] The input fuzzification layer receives battery health status and state of charge information as input variables, and maps the input variables to multiple fuzzy language sets through multiple built-in membership functions;

[0071] The fuzzy inference layer contains a fuzzy inference rule base generated by expert knowledge and data training. The trigger strength of each rule is calculated based on the fuzzy language set and normalized.

[0072] The output layer performs weighted calculations on the output conclusions of each rule based on the normalized trigger strength to obtain the maximum allowable charging current, the highest allowable terminal voltage, and the highest allowable surface temperature of the safety threshold, and constructs a multi-dimensional safety hyperplane.

[0073] Specifically, the model's construction and training are completed in an offline preparation phase, requiring a training database containing inputs and desired outputs. This database is obtained through extensive battery safety boundary experiments or precise electrochemical simulations. Each record in the database contains a set of inputs (battery health state and state of charge) and corresponding verified safe output thresholds (maximum allowable charging current, maximum allowable terminal voltage, and maximum allowable surface temperature). Using this database, the membership function parameters and fuzzy rule base of the adaptive neurofuzzy inference system model are iteratively optimized and trained through a hybrid learning algorithm. The trained model is then stored and ready for online real-time retrieval.

[0074] In the input fuzzification layer, the current battery health status (e.g., 92%) and state of charge (e.g., 75%) are obtained in real time from the vehicle communication protocol as input variables. This layer maps these two input values ​​to multiple fuzzy language sets through built-in multiple sets of membership functions, such as Gaussian membership functions. For example, when the battery health status is 92%, it is mapped to "healthy" membership degree of 0.8 and "sub-healthy" membership degree of 0.2; when the state of charge is 75%, it is mapped to "medium" membership degree of 0.4 and "high" membership degree of 0.6.

[0075] Specifically, the membership function is a Gaussian membership function. For example, for the state of battery health (SOH), it can be divided into three fuzzy sets: "healthy," "sub-healthy," and "degraded," and the center point and width parameters of the Gaussian function corresponding to each set are given. For the fuzzy inference rule base, specific rule examples are as follows: "Rule 1: If the battery health state is 'healthy' and the state of charge is 'high,' then the safety threshold is 'conservative' (e.g., the maximum allowable charging current is 60% of the rated value)"; "Rule 2: If the battery health state is 'healthy' and the state of charge is 'low,' then the safety threshold is 'aggressive' (e.g., the maximum allowable charging current is 100% of the rated value)." This rule base obtains training data through a large number of battery safety boundary experiments or precise electrochemical simulations, and is iteratively optimized and trained using a hybrid learning algorithm of the Adaptive Neural Fuzzy Inference System (ANFIS), ultimately being embedded in the controller.

[0076] Furthermore, in the fuzzy inference layer, these fuzzy membership values ​​are fed into a fixed fuzzy inference rule base. This rule base contains several expert rules in the form of "If the battery health state is healthy and the state of charge is high, then the safety threshold is low". Based on the membership of each input variable, the trigger strength of each rule is calculated, and the trigger strength of all rules is normalized to obtain the weight of each rule in the final decision.

[0077] Finally, at the clarification output layer, based on the normalized trigger strength calculated by the fuzzy inference layer, a weighted summation is performed on the output conclusions (i.e., specific current, voltage, and temperature thresholds) corresponding to each rule. This process transforms the fuzzy inference results back into precise values. This calculation is performed independently and in parallel three times, yielding the maximum allowable charging current, the highest allowable terminal voltage, and the highest allowable surface temperature under the current state. These three dynamically generated safety thresholds, generated in real time, together construct a multi-dimensional safety hyperplane in the state space, serving as the boundary benchmark for subsequent risk assessment. An example of the hyperplane is shown below. Figure 2 As shown.

[0078] As a preferred implementation, the multidimensional safety hyperplane generated by the dynamic safety boundary model is a nonlinear surface that characterizes the multi-parameter coupling effect. This surface is defined by an offline-trained support vector machine model. The model's training data comes from battery abuse experiments and simulations, enabling it to learn under what multidimensional parameter combinations the battery will enter an irreversible dangerous state. In real-time control, the model uses the collected multidimensional feature parameters as input and directly outputs a scalar value representing the distance from the current state point to the true safety boundary. This boundary definition accurately captures the synergistic effects between parameters, allowing for a relatively higher battery temperature at lower charging currents, thus constructing a smooth and nonlinear safety boundary that more closely approximates the battery's true electrochemical stability domain.

[0079] It can generate dynamically changing safety boundaries based on the battery's real-time health and state of charge. This overcomes the shortcomings of traditional fixed thresholds that cannot adapt to battery aging and changes in operating conditions, and provides personalized protection strategies that better meet the actual safety needs of batteries in different states, helping to improve the overall performance of the charging process while ensuring safety.

[0080] Furthermore, the calculation process of the safety margin scalar includes:

[0081] The multidimensional safety hyperplane is defined as the boundary of a repulsive potential field that applies a virtual repulsive force. Based on the Euclidean distance from the charging response vector to the boundary of the repulsive potential field, a position potential energy scalar determined by geometric distance is calculated. Based on the dynamic characteristic indices in the charging response vector, a state evolution velocity vector representing the direction and rate of system state evolution is constructed in the state space. The projection of the state evolution velocity vector onto the normal direction of the boundary of the repulsive potential field is calculated to obtain a kinetic energy risk scalar determined by the state approach rate. The position potential energy scalar and the kinetic energy risk scalar are input into a preset risk fusion model for nonlinear superposition to generate the safety margin scalar.

[0082] Specifically, after obtaining the charging response vector and the multidimensional safety hyperplane at the current moment, the risk assessment calculation based on potential field theory is immediately executed. The multidimensional safety hyperplane generated by the dynamic safety boundary model is defined in the state space as a repulsive potential field boundary that applies virtual repulsive force.

[0083] First, the position potential energy scalar representing static risk is calculated by finding the shortest Euclidean distance from the current charging response vector to the boundary of the repulsive potential field hyperplane. The smaller the distance value, the closer the state point is to the safety boundary. This distance value is then transformed by an inverse proportional function, converting the geometric distance into a potential energy scalar. Thus, the closer the state point is to the boundary, the larger the calculated position potential energy scalar, representing a higher static risk.

[0084] Meanwhile, to assess the dynamic evolution trend of the state, it is necessary to construct a state evolution velocity vector. Each component of this vector is determined by performing least squares linear regression fitting on the data points of the N principal component scores that constitute the charging response vector within the most recent time window (e.g., the most recent 3 calculation cycles), and taking the slope of the fitted line as the current rate of change. This vector clearly characterizes the direction and speed of movement of the current state point in the N-dimensional feature subspace.

[0085] Next, the kinetic risk scalar, representing dynamic risk, is calculated. First, the normal direction of the repulsive potential field boundary at the point closest to the charging response vector is determined. Then, the state evolution velocity vector constructed in the previous step is projected onto this normal direction. The result of this projection is a scalar characterizing the approach rate of the state point perpendicular to the boundary. A positive rate indicates that the state point is approaching the boundary; a negative rate indicates that it is moving away. This approach rate value is input into a nonlinear function, such as one that is proportional only to its square when the rate is positive, to obtain the kinetic risk scalar. A faster approach rate will produce an exponentially growing kinetic risk scalar.

[0086] Finally, the calculated position potential energy scalar and kinetic energy risk scalar are superimposed using a predefined nonlinear fusion function. In this embodiment, a weighted and saturated sigmoid function can be used to map the weighted sum of the two risk terms into a range of 0 to 1. The dimensionless value output by this function is the safety margin scalar at the current moment. The closer this value is to 1, the higher the overall risk level and the lower the safety margin; conversely, the closer it is to 0, the safer it is.

[0087] Specifically, the nonlinear superposition is implemented using a sigmoid function with adjustable weights. First, the calculated position potential energy scalar and kinetic energy risk scalar are weighted and summed. The weighting coefficients are preset values ​​used to adjust the importance of static and dynamic risks in the final assessment. For example, in this embodiment, the weight of static risk can be set to 0.4, and the weight of dynamic risk to 0.6, to place greater emphasis on early warning of dynamic risk trends. Then, this weighted sum is used as input and nonlinearly mapped through the sigmoid function, transforming and outputting a dimensionless value within the standardized range of 0 to 1, which is the final safety margin scalar.

[0088] By integrating positional potential energy, representing static distance, and kinetic risk, representing dynamic approach rate, this method establishes a comprehensive risk assessment model that simultaneously considers "position" and "velocity." This model endows safety margin assessment with a certain predictive capability, helps to identify rapidly developing risk trends in advance, compensates for the lag in traditional static judgments, and provides a more comprehensive decision-making basis for subsequent regulation.

[0089] Furthermore, the adaptive control model includes:

[0090] The error signal generation unit receives the safety margin scalar, compares it with the internally set target safety margin value in real time, and uses the difference between the two as the error signal.

[0091] The control component parallel processing unit inputs the error signal into three parallel mathematical operation channels; the proportional channel directly performs linear scaling on the error signal, the integral channel performs time accumulation operation on the error signal, and the differential channel calculates the rate of change of the error signal, and synchronously outputs three control components representing the current, historical and future predictions respectively.

[0092] The dynamic gain scheduling unit uses the externally input battery health status and temperature information as query coordinates, performs interpolation calculations on the preset multi-dimensional gain parameter surface, and outputs the proportional, integral, and differential gain coefficients with respect to the current actual battery operating conditions.

[0093] The instruction weighted synthesis unit receives the control component and the gain coefficient, multiplies the control component by the corresponding gain coefficient one by one and sums them to obtain the control output value, and converts the control output value into a specific adjustment instruction for the charging current.

[0094] Specifically, upon receiving the calculated safety margin scalar, the adaptive control process is immediately initiated. First, in an error signal generation step, the real-time input safety margin scalar (a value between 0 and 1, with higher values ​​indicating greater risk) is compared with an internally set target safety margin value (e.g., 0.2) representing the ideal safety state. By calculating the difference between the two, i.e., the actual margin minus the target margin, a real-time changing error signal is generated.

[0095] The internally set target safety margin value is not fixed, but adaptively adjusted according to the charging scenario and battery aging level. Specifically, users can select different charging modes, such as the fastest charging mode and the battery maintenance mode. When the fastest charging mode is selected, the target safety margin value is set at a relatively high level (e.g., 0.3) to allow for more aggressive charging in areas close to the safety boundary. When the battery maintenance mode is selected, the target value is set at a more conservative level (e.g., 0.15) to reserve a larger safety redundancy. In addition, the target value is also related to the battery's health status. A built-in lookup table makes the target safety margin value decrease accordingly as the SOH decreases, so as to provide a gentler protection strategy for aging batteries.

[0096] The error signal is input in parallel to three independent mathematical operation channels for processing. In the proportional channel, the current error signal is linearly scaled to generate a proportional control component proportional to the current error magnitude. In the integral channel, the values ​​of the error signal over a set period of time are accumulated and summed to generate an integral control component capable of eliminating steady-state errors. In the derivative channel, the rate of change of the error signal between the two most recent calculation cycles is calculated to generate a derivative control component capable of predicting the trend of error changes. These three control components are output synchronously.

[0097] At the same time, a dynamic gain scheduling step is executed. Based on the real-time acquired battery health status and the highest battery surface temperature, a query is performed on a multi-dimensional gain parameter surface that has been established in advance through experiments or simulations. This surface stores the optimal proportional, integral, and differential gain coefficient sets under different health status and temperature combinations. By interpolating the parameters around the query point, a set of dynamic gain coefficients that accurately match the current battery operating conditions are obtained.

[0098] Specifically, the preset multidimensional gain parameter surface is a lookup table generated through offline optimization and calibration. Its construction process is as follows: First, a high-precision co-simulation model of the battery charging system is established. Then, using a particle swarm optimization algorithm, different combinations of battery health states and temperatures are used as operating points. For each operating point, with the optimization objective of minimizing overshoot and settling time during the safety margin adjustment process, a set of optimal proportional, integral, and derivative gain coefficients is automatically searched and determined. All operating points and their corresponding optimal gain coefficient combinations are stored, thus forming the multidimensional gain parameter surface. During real-time scheduling, the system uses a bilinear interpolation algorithm to calculate the most suitable gain coefficient for the current operating condition based on the real-time acquired battery health state and temperature values ​​from this lookup table.

[0099] Finally, in a weighted synthesis step, the three control components obtained above—proportional, integral, and derivative—are multiplied one by one by the three corresponding proportional, integral, and derivative gain coefficients generated by dynamic scheduling. Then, the results of these three multiplications are summed to obtain a final control output value. This control output value is interpreted as a specific adjustment amount to the charging current. For example, a positive output value corresponds to a proportional decrease in the charging current, while a negative output value corresponds to a proportional increase in the charging current within a safe range. This final adjustment instruction is sent to the power control module of the charging device for execution, thereby completing the dynamic, closed-loop, and safe regulation of the charging current.

[0100] By using dynamic gain scheduling, the control parameters can be matched with the battery's health status and temperature in real time, overcoming the shortcomings of fixed parameter controllers that have poor response under battery aging or different operating conditions. By using a complete proportional-integral-derivative structure, the adjustment process of the safety margin is fast, accurate and stable, achieving more refined and reliable closed-loop safety management throughout the battery's entire life cycle.

[0101] By fusing time-frequency domain electrochemical characteristics with spatial thermal field gradient information, a multi-dimensional, multi-physics-based deep state perception system was constructed, overcoming the limitations of traditional monitoring. Based on this, principal component analysis was used to extract robust response vectors from high-dimensional data containing dynamic trends, and these vectors were compared with a dynamic safety boundary that adaptively adjusts based on the battery's real-time health and state of charge. This comparison process employs risk assessment based on potential field theory, considering both the "position" and "velocity" of the state, achieving proactive risk quantification and compensating for the lag in traditional passive protection. Finally, by implementing closed-loop control of this predictive safety margin, more refined and proactive safety management was achieved, improving the balance between charging safety and performance throughout the battery's entire lifecycle.

[0102] Example 2:

[0103] To further enhance the risk perception and proactive safety control capabilities during the charging process of new energy vehicles, a new energy vehicle charging safety control system of the present invention is introduced, as shown in the schematic diagram below. Figure 3 As shown.

[0104] A new energy vehicle's power battery, diagnosed by the onboard BMS, currently has a State of Health (SOH) of 85%, with some degree of cell capacity degradation and a significantly higher internal resistance compared to a new battery. The vehicle is connected to a DC fast charging station at an ambient temperature of 35 degrees Celsius, with an initial State of Charge (SOC) of 20%.

[0105] Once charging begins, periodic diagnostics are initiated every 5 seconds. During the initial charging phase (SOC from 20% to 70%), the battery state is relatively stable. The collected multi-dimensional characteristic parameters exhibit baseline levels consistent with its aging state; for example, the ohmic internal resistance remains stable at 0.8 milliohms, the polarization internal resistance changes slowly around 2.5 milliohms, and the thermal gradient parameter remains at a low level of around 1.5 degrees Celsius, indicating relatively uniform heat generation. During this period, the Adaptive Neural Fuzzy Inference System (ANFIS) model generates a more conservative safety hyperplane in real time than that of a new battery, based on an 85% health state and the continuously increasing state of charge. For example, its maximum permissible charging current is limited to 200 amps, instead of the 250 amps of a new battery. Based on stable input parameters, the calculated safety margin scalar remains consistently around the target value of 0.2, and the charging process proceeds stably at the maximum permissible current under the current operating conditions.

[0106] When charging reaches approximately 75% of its state of charge, the difficulty of lithium ion insertion into the graphite anode increases at this high state of charge, and the polarization effect is significantly enhanced. For this aged battery, a certain cell region with poor consistency (possibly due to slight lithium deposition or uneven electrolyte wetting) becomes the bottleneck for current charging efficiency, causing a sharp increase in local current density and polarization overpotential in this region, with a heat generation rate far exceeding that of other regions. In the following control cycles, the multi-dimensional sensing steps of the present invention captured this series of abnormal signs: First, the thermal field gradient parameter rapidly climbed from 1.5 degrees Celsius to 4.0 degrees Celsius within 30 seconds, clearly indicating the formation of local hot spots; at the same time, the polarization internal resistance value resolved from the excitation pulse increased by nearly 20%, and the impedance amplitude at the 1 Hz frequency point also showed a significant nonlinear increase, all of which reflect that the electrochemical environment in this local region is rapidly deteriorating.

[0107] These anomalous parameter changes are immediately reflected in the state trend fusion matrix. In particular, the second column, representing the rate of change, shows a significant numerical jump. Although the instantaneous values ​​of all parameters are still within the safety boundaries set by the model, the principal component analysis (PCA) process is extremely sensitive to such coordinated and dramatic changes in multiple parameters. This can be summarized as a rapid drift of the projected coordinates (principal component scores) on a key principal component axis. This drift causes the state evolution velocity vector to show a tendency to rush towards the safety hyperplane at high speed. Therefore, in the risk assessment based on potential field theory, the "position potential energy scalar" representing static risk only increases moderately, but the kinetic risk scalar representing dynamic risk, due to its proportionality to the square of the approach rate, increases by nearly tenfold. The safety margin scalar obtained by the final fusion rapidly climbs from a stable 0.2 to a high-risk region of 0.75 within just a few cycles.

[0108] The high-risk signal immediately triggered precise intervention from the adaptive control model. A large error signal was generated between the safety margin scalar of 0.75 and the target value of 0.2, driving the PID controller to generate a strong response demand. At the same time, the dynamic gain scheduling unit, based on the current 85% health status and the battery surface temperature that had climbed to a maximum of 48 degrees Celsius, selected a set of relatively smooth and stability-oriented PID gain coefficients to avoid current oscillations caused by over-adjustment. Finally, the command weighted synthesis unit calculated a clear control output value based on this set of gains and the large error signal, and converted it into a command requiring the charging pile power module to smoothly and gradually reduce the charging current from 180 amps to 120 amps within 30 seconds.

[0109] As the charging current decreases, the overall heat generation rate of the battery and the degree of electrochemical polarization in local areas are effectively mitigated. In the following control cycles, the thermal gradient parameter stops rising and begins to slowly fall back to below 3.0 degrees Celsius, and the growth trend of electrochemical parameters such as polarization resistance also returns to normal. The safety margin scalar then returns to the target safety level of 0.2. Finally, the charging process continues at a safer and lower current until charging is complete. In contrast, traditional BMS that rely solely on a fixed temperature threshold of 55 degrees Celsius do not take any measures at this stage, which may allow the local overheating problem to worsen. The system of this invention successfully identifies and actively intervenes in this potential thermal runaway risk in advance, realizing intelligent, precise, and safe charging of aging batteries under harsh operating conditions.

[0110] By simultaneously analyzing the spatial anomalies of the thermal field gradient and the temporal trends of electrochemical parameters, early characteristics of local thermal runaway were accurately captured. Its predictive risk assessment mechanism can proactively intervene before absolute limits are reached, mitigating crises through adaptive smooth current regulation. This demonstrates significant advantages over traditional passive protection strategies in improving refined safety management and full-lifecycle charging reliability.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling the charging safety of new energy vehicles, characterized in that, include: Inject excitation pulses into the charging circuit and collect a set of multi-dimensional feature parameters triggered by the excitation pulses; The battery health status and state of charge information are obtained through the vehicle communication protocol, and the temperature information of multiple points on the battery surface is collected. The multidimensional feature parameter set is constructed into a multidimensional time series, and statistical analysis is performed on the multidimensional time series to extract dynamic feature indicators that characterize short-term volatility and changing trends. By constructing a state-trend fusion matrix, the current values ​​of dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and multi-dimensional feature parameter sets are structurally fused. The matrix is ​​then projected onto the key feature subspace using principal component analysis to obtain the charging response vector. The battery health status and state of charge information are input into the dynamic safety boundary model to generate a multidimensional safety hyperplane; the charging response vector is continuously compared with the multidimensional safety hyperplane to calculate the safety margin scalar in real time. The safety margin scalar is input into the adaptive control model to dynamically adjust the charging current output.

2. The method for regulating the charging safety of new energy vehicles according to claim 1, characterized in that, The process of acquiring a multi-dimensional feature parameter set triggered by an excitation pulse includes: At the output end of the charging pile, the complete voltage transient response curve triggered by the excitation pulse is captured by high-frequency sampling. The voltage transient response curve is analyzed. The ohmic internal resistance is calculated by analyzing the initial vertical jump part of the curve. The polarization internal resistance is calculated by analyzing the nonlinear gradual change part of the rising stage of the curve. The relaxation time constant is calculated by analyzing the decay rate of the voltage fall-off stage after the excitation pulse ends. The voltage transient response curve is then subjected to a fast Fourier transform to extract the impedance amplitude and phase angle at a preset characteristic frequency point as frequency domain characteristic parameters.

3. The method for controlling the charging safety of new energy vehicles according to claim 1, characterized in that, The process of collecting multi-point temperature information on the battery surface includes: simultaneously collecting temperature readings at multiple locations using a temperature sensor array arranged on the surface of the battery module; and obtaining the thermal field gradient parameters by calculating the difference between the maximum and minimum values ​​of the temperature readings.

4. The method for regulating the charging safety of new energy vehicles according to claim 1, characterized in that, The process of obtaining the charging response vector includes: Within a preset time window, the time series of the multi-dimensional feature parameter set is calculated to obtain the time change rate as a dynamic feature index; a two-dimensional state trend fusion matrix is ​​constructed, where the rows of the matrix correspond to each physical quantity in the multi-dimensional feature parameter set, and the columns of the matrix correspond to the current instantaneous value and dynamic feature index of each physical quantity, respectively; the thermal field gradient parameters calculated based on multi-point temperature information are added to the matrix as additional dimension information; principal component analysis is performed on the state trend fusion matrix, and the top N principal component vectors whose cumulative variance contribution rate reaches the threshold are selected; the state trend fusion matrix is ​​projected onto the feature subspace spanned by the N principal component vectors to obtain a set of N-dimensional principal component scores; the principal component scores are arranged in order of importance to form a charging response vector.

5. The method for regulating the charging safety of new energy vehicles according to claim 1, characterized in that, The dynamic security boundary model includes: The input fuzzification layer receives battery health status and state of charge information as input variables, and maps the input variables to multiple fuzzy language sets through multiple built-in membership functions; The fuzzy inference layer contains a fuzzy inference rule base generated by expert knowledge and data training. The trigger strength of each rule is calculated based on the fuzzy language set and normalized. The output layer performs weighted calculations on the output conclusions of each rule based on the normalized trigger strength to obtain the maximum allowable charging current, the highest allowable terminal voltage, and the highest allowable surface temperature of the safety threshold, and constructs a multi-dimensional safety hyperplane.

6. The method for regulating the charging safety of new energy vehicles according to claim 1, characterized in that, The calculation process for the safety margin scalar includes: The multidimensional safety hyperplane is defined as the boundary of a repulsive potential field that applies a virtual repulsive force. Based on the Euclidean distance from the charging response vector to the boundary of the repulsive potential field, a position potential energy scalar determined by geometric distance is calculated. Based on the dynamic characteristic indices in the charging response vector, a state evolution velocity vector representing the direction and rate of system state evolution is constructed in the state space. The projection of the state evolution velocity vector onto the normal direction of the boundary of the repulsive potential field is calculated to obtain a kinetic energy risk scalar determined by the state approach rate. The position potential energy scalar and the kinetic energy risk scalar are input into a preset risk fusion model for nonlinear superposition to generate the safety margin scalar.

7. The method for controlling the charging safety of new energy vehicles according to claim 1, characterized in that, The adaptive control model includes: The error signal generation unit receives the safety margin scalar, compares it with the internally set target safety margin value in real time, and uses the difference as an error signal. The control component parallel processing unit inputs the error signal into three parallel mathematical operation channels; the proportional channel directly performs linear scaling on the error signal, the integral channel performs time accumulation operation on the error signal, and the differential channel calculates the rate of change of the error signal, and synchronously outputs three control components representing the current, historical and future predictions respectively. The dynamic gain scheduling unit uses the externally input battery health status and temperature information as query coordinates, performs interpolation calculations on the preset multi-dimensional gain parameter surface, and outputs the proportional, integral, and differential gain coefficients with respect to the current actual battery operating conditions. The instruction weighted synthesis unit receives the control component and the gain coefficient, multiplies the control component by the corresponding gain coefficient one by one and sums them to obtain the control output value, and converts the control output value into a specific adjustment instruction for the charging current.

8. A new energy vehicle charging safety control system, characterized in that, include: The data acquisition module injects excitation pulses into the charging circuit and collects a set of multi-dimensional feature parameters triggered by the excitation pulses. The battery health status and state of charge information are obtained through the vehicle communication protocol, and the temperature information of multiple points on the battery surface is collected. The feature extraction module constructs the multi-dimensional feature parameter set into a multi-dimensional time series, performs statistical analysis on the multi-dimensional time series, and extracts dynamic feature indicators that characterize short-term volatility and changing trends. The feature mapping module constructs a state trend fusion matrix to structurally fuse dynamic feature indicators, thermal field gradient parameters calculated based on multi-point temperature information, and the current values ​​of multi-dimensional feature parameter sets. The matrix is ​​then projected onto the key feature subspace using principal component analysis to obtain the charging response vector. The current adjustment module inputs the battery health status and state of charge information into the dynamic safety boundary model to generate a multi-dimensional safety hyperplane; it continuously compares the charging response vector with the multi-dimensional safety hyperplane to calculate the safety margin scalar in real time. The safety margin scalar is input into the adaptive control model to dynamically adjust the charging current output.

Images