Vehicle life cycle-oriented charging safety early warning threshold dynamic prediction method

Abstract

The application relates to the technical field of charging safety early warning, and particularly discloses a charging safety early warning threshold dynamic prediction method for the whole life cycle of a vehicle, which comprises the following steps: collecting basic cycle data, electrochemical characteristic data, user charging behavior data and multi-physical field parameters of the environment where the vehicle is located in real time; based on the basic cycle data and the electrochemical characteristic data, a quantitative judgment system is constructed, and the whole life cycle of the battery is divided into an initial decline stage, a middle decline stage and a non-decline stage. The application guarantees the data reliability through multi-source data acquisition and fusion preprocessing, combines the quantitative judgment of the battery aging stage and the differentiated SVR model adaptation, so that the early warning threshold can dynamically match the whole life cycle state of the battery from the initial decline to the final decline; the user charging behavior influence coefficient is introduced, the threshold is adapted to different use habits, the adaptation capability under complex working conditions is enhanced, and stable guarantee is provided for the charging safety of the whole life cycle of the vehicle.

Description

Technical Field

[0001] This invention relates to the field of charging safety early warning technology, and in particular to a dynamic prediction method for charging safety early warning thresholds throughout the entire life cycle of a vehicle. Background Technology

[0002] With the rapid development of the electric vehicle industry, the charging safety and reliability of power batteries, as core components, have increasingly become a focus of industry attention. During charging and discharging, safety issues such as thermal runaway, overcharging, and insulation failure can directly endanger the safety of the vehicle and its occupants. Therefore, establishing an effective charging safety early warning system to identify and intervene in potential risks at an early stage is of great practical significance.

[0003] Currently, there is considerable research and practice in battery charging safety early warning both domestically and internationally, primarily employing early warning strategies based on fixed thresholds. These methods typically set uniform safety thresholds for voltage, temperature, and current based on battery factory calibration parameters or experimental data under ideal operating conditions. When monitored data exceeds the preset threshold, an alarm or protection action is triggered. However, these methods have significant limitations: First, battery performance significantly degrades and fluctuates with increasing usage cycles, changes in environmental conditions, and differences in user charging habits. Fixed thresholds cannot adapt to the dynamically changing health status of the battery throughout its lifespan, easily leading to high false alarm rates in the early stages and a risk of missed alarms in the later stages of aging due to overly lenient thresholds. Second, existing early warning models often rely solely on one type of data (such as considering only voltage or temperature), failing to comprehensively consider the coupled effects of multiple dimensions such as battery aging, user behavior, and extreme environments, resulting in insufficient early warning accuracy. Third, while some solutions attempt to introduce environmental temperature compensation, they lack quantitative characterization and dynamic adaptation capabilities for complex extreme conditions such as high altitude and low pressure, and high humidity condensation, and their comprehensive impact on heat dissipation efficiency, insulation performance, and electrolyte activity. Summary of the Invention

[0004] This invention aims to at least partially address one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a dynamic prediction method for charging safety warning thresholds throughout the entire vehicle lifecycle, in order to improve the reliability and adaptability of vehicle charging safety warnings.

[0005] To achieve the above objectives, a first aspect of the present invention proposes a dynamic prediction method for charging safety warning thresholds throughout the entire lifecycle of a vehicle. The method includes the following steps: S1, real-time acquisition of basic cycle data, electrochemical characteristic data, user charging behavior data, and multi-physics parameters of the vehicle's environment; S2, based on the basic cycle data and electrochemical characteristic data, constructing a quantitative judgment system to divide the entire lifecycle of the battery into an initial aging stage, a medium aging stage, and a non-aging stage; the electrochemical characteristic data includes at least the battery internal resistance and static voltage rebound; S3, dynamically adapting differentiated support vector regression (SVR) model parameters and threshold correction coefficients according to the determined aging stage to generate a basic warning threshold based on the aging state; S4, parsing the user charging behavior data, calculating the behavioral influence coefficient used to characterize the risk level of user charging habits, and correcting the basic warning threshold; S5, based on the multi-physics parameters, calculating a multi-physics coupling factor used to quantify the impact of extreme environments on battery performance, and using the multi-physics coupling factor to further correct the warning threshold after correction by the behavioral influence coefficient, ultimately generating a dynamically predicted charging safety warning threshold.

[0006] To achieve the above objectives, a second aspect of the present invention proposes a dynamic prediction system for charging safety warning thresholds throughout the entire life cycle of a vehicle, comprising: a data acquisition module for real-time acquisition of basic cycle data, electrochemical characteristic data, user charging behavior data, and multi-physical field parameters of the vehicle's environment, wherein the electrochemical characteristic data includes at least the battery's internal resistance and static voltage rebound; and an aging stage determination module for constructing a two-dimensional quantitative determination system based on the basic cycle data and electrochemical characteristic data, dividing the battery's entire life cycle into the initial aging stage, the intermediate aging stage, and the non-aging stage. The aging stage includes a model parameter adaptation and basic threshold generation module, used to dynamically adapt differentiated support vector regression (SVR) model parameters and threshold correction coefficients based on the aging stage output by the aging stage determination module, generating a basic warning threshold based on the battery aging state; a behavior influence coefficient calculation and threshold correction module, used to parse the user charging behavior data, extract multiple charging behavior features and quantize and classify them, multiply the quantized values ​​of each feature level with their weight coefficients and accumulate them to obtain the behavior influence coefficient, and use this coefficient to correct the basic warning threshold; and a coupling factor calculation and final threshold generation module, used to calculate based on the multiphysics parameters. A multi-physics coupling factor reflecting changes in high-altitude heat dissipation efficiency, low-temperature electrolyte conductivity, and high-humidity insulation performance is used to further refine the warning threshold after behavioral influence coefficient correction, generating dynamically predicted charging safety warning thresholds. A data preprocessing submodule, integrated between the data acquisition module and subsequent modules, is used to establish a unified system time base for data synchronization timestamps, remove abnormal data through data range and rate of change threshold verification, compensate for missing data using interpolation algorithms, and output a standardized data sequence. An abnormal behavior filtering submodule, integrated into the behavioral influence coefficient calculation and threshold correction module, is used to establish abnormal behavior... The system includes a behavior recognition rule to clean raw charging behavior data and eliminate the influence of non-habitual abnormal behaviors; a stage switching trigger submodule, integrated into the aging stage determination module, for setting trigger conditions such as cycle count increment and capacity decay rate, automatically initiating aging stage re-determination when the conditions are met, updating system parameters and pushing status information when the stage changes; and a parameter self-learning calibration submodule for setting periodic calibration cycles and monitoring significant changes in battery performance, user behavior patterns, or environmental conditions, collecting recent operating data when triggered, re-optimizing SVR model parameters, behavior influence coefficient weights, and multiphysics coupling factors, and updating them to each module to achieve dynamic calibration.

[0007] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described dynamic prediction method for charging safety warning thresholds throughout the vehicle's entire lifecycle.

[0008] The dynamic prediction method for charging safety warning thresholds for the entire vehicle lifecycle, as described in this invention, ensures data reliability through multi-source data acquisition and fusion preprocessing. It combines quantitative determination of battery aging stages with differentiated SVR model adaptation, enabling the warning thresholds to dynamically match the battery's lifecycle state from initial degradation to final degradation. Furthermore, it introduces a user charging behavior influence coefficient to adapt the thresholds to different usage habits. By leveraging multi-physics coupling factors to quantify the impact of extreme environments on battery performance, it enhances adaptability under complex operating conditions. Simultaneously, through abnormal behavior filtering, adaptive stage switching, and online parameter self-learning calibration, it continuously optimizes warning accuracy, effectively reducing false alarms and missed alarms throughout the entire lifecycle, improving the reliability and adaptability of charging safety warnings, and providing stable protection for vehicle charging safety throughout its entire lifecycle. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the dynamic prediction method for charging safety early warning thresholds for the entire lifecycle of a vehicle provided by the present invention. Figure 2 This is a scatter plot of the two-dimensional aging stage division in the dynamic prediction method for charging safety early warning thresholds for the entire life cycle of vehicles provided by this invention. Figure 3 This is a schematic diagram illustrating the changes in SVR model parameters with battery aging stages in the dynamic prediction method for charging safety warning thresholds throughout the entire vehicle lifecycle provided by this invention. Figure 4 This is a schematic diagram comparing the effects of behavior influence coefficient calculation and abnormal behavior filtering in the dynamic prediction method for charging safety early warning thresholds for the entire life cycle of vehicles provided by this invention. Figure 5 This is a schematic diagram of the three-dimensional response surface of the multi-physics coupling factor in the dynamic prediction method for charging safety warning thresholds for the entire life cycle of vehicles provided by this invention. Figure 6 This is a schematic diagram of the stage switching triggering mechanism in the dynamic prediction method for charging safety early warning thresholds for the entire life cycle of vehicles provided by the present invention. Figure 7 This is a schematic diagram comparing the effects of online self-learning and dynamic calibration mechanisms of parameters in the dynamic prediction method for charging safety warning thresholds for the entire life cycle of vehicles provided by this invention. Figure 8 This is a schematic diagram of the structure of the dynamic prediction system for charging safety early warning thresholds for the entire life cycle of vehicles provided by the present invention; Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0011] The following description, with reference to the accompanying drawings, outlines a method, system, and electronic device for dynamically predicting charging safety warning thresholds throughout the vehicle's entire lifecycle.

[0012] Example 1: Figure 1 This is a flowchart illustrating a dynamic prediction method for charging safety warning thresholds throughout the entire lifecycle of a vehicle, according to an embodiment of the present invention. The method executes steps S1 to S5 sequentially, combining a dual-dimensional judgment system of macroscopic cycle characteristics and microscopic electrochemical features of the battery, and dynamically adapting through a differentiated support vector regression model, thereby achieving dynamic prediction of charging safety thresholds throughout the entire lifecycle of the battery, from the initial decay stage to the final decay stage. Each step will be explained in detail below: Figure 1 As shown, a dynamic prediction method for charging safety warning thresholds for the entire life cycle of a vehicle includes the following steps: S1, real-time collection of basic cycle data, electrochemical characteristic data, user charging behavior data, and multi-physical field parameters of the vehicle's environment, wherein: 1. Basic cycle data of the battery refers to long-term operating data reflecting the macroscopic use process of the battery, mainly including two core indicators: cycle number and capacity decay rate.

[0013] Loop count This refers to the cumulative number of times a battery completes one charge-discharge cycle. A complete cycle refers to the battery discharging from a fully charged state to the cutoff voltage and then recharging to a fully charged state. The number of cycles is an important parameter for measuring battery life.

[0014] For example, a certain type of ternary lithium battery has a rated cycle life of 2,000 cycles. When the vehicle has accumulated 500 cycles, the battery can be considered to be in the early aging stage; when the accumulated cycles exceed 1,500, it enters the late aging stage.

[0015] Capacity decay rate This refers to the difference between the battery's current actual capacity and its rated capacity, expressed as a percentage of the rated capacity. It reflects the degree of degradation in the battery's energy storage capacity. The calculation formula is: ;in: The actual current capacity of the battery (unit: kWh) is calculated in real time using the capacity estimation algorithm of the BMS. This is the rated capacity.

[0016] For example, a 60 kWh battery has a capacity of 60 kWh when the car is new; after 3 years of use, the capacity drops to 54 kWh, then the degradation rate is... .

[0017] Data acquisition method: Both the number of cycles and the capacity decay rate can be obtained through the recording and estimation functions in the battery management system.

[0018] 2. Electrochemical characteristic data reflects the internal microscopic mechanism of the battery. This embodiment mainly collects the following two types of data: current battery internal resistance. Battery internal resistance refers to the resistance exhibited by a battery during operation, comprising three parts: ohmic internal resistance, polarization internal resistance, and diffusion internal resistance. As the battery ages, the loss of active materials, increased polarization, and restricted diffusion all lead to an increase in internal resistance. Therefore, battery internal resistance is an important indicator reflecting the battery's health status.

[0019] For example, a new battery has a DC internal resistance of 2 milliohms, which rises to 3 milliohms after two years of use, indicating that the battery performance has deteriorated.

[0020] Static voltage rebound When a battery is left to stand for a period of time after it has completed charging and discharging, the terminal voltage will gradually rise or fall back to a stable value from the instantaneous value affected by polarization. The difference between the two is the resting voltage rebound, which reflects the degree of battery polarization and the reversibility of active materials.

[0021] For example, if a battery has a voltage of 4.2V when charging is stopped and drops to 4.15V after being left to stand for one hour, the rebound is 50 millivolts. If the rebound decreases to 20 millivolts with aging, it indicates that the battery polarization has intensified and the recovery ability of the active materials has declined.

[0022] 3. User charging behavior data refers to the charging habits formed by users during long-term vehicle use. Common behavioral data include: average charging frequency; fast charging ratio; average charging depth range, such as commonly charging in the 20%~80% range; nighttime charging ratio, etc. This data is collected jointly by the vehicle charging recorder and the battery management system.

[0023] 4. Environmental multi-physics parameters: Environmental parameters include external conditions such as temperature, humidity, and altitude. These conditions will be used in subsequent steps to correct the warning threshold after the behavior influence coefficient has been corrected.

[0024] For example, taking an electric vehicle as an example, after two years of operation, the vehicle has accumulated 800 cycles (out of a rated 2000 cycles). Its capacity has decreased from 60 kWh to 54 kWh, its internal resistance has increased to 1.3 times its initial value, and its static voltage rebound remains at 50 millivolts. At this point, the data collected by S1 will be used as input for subsequent steps.

[0025] S2. Based on the battery's basic cycle data and electrochemical characteristic data, a quantitative judgment system is constructed to divide the battery's entire life cycle into the initial decay stage, the intermediate decay stage, and the non-decay stage.

[0026] The quantitative judgment system is specifically based on a two-dimensional judgment model of the battery's macroscopic cycle characteristics and microscopic electrochemical state. The macroscopic cycle characteristics dimension includes the battery's cumulative cycle count and cumulative capacity decay rate; the microscopic electrochemical state dimension includes the battery's internal resistance and voltage rebound after resting.

[0027] The dual-dimensional judgment model uses the following judgment rules to divide the stages: the macroscopic preliminary judgment is based on the comparison results of the battery's cumulative cycle count, cumulative capacity decay rate and corresponding thresholds to make a preliminary stage division; then the microscopic correction judgment combines the growth multiple of the battery's internal resistance relative to the initial internal resistance and the comparison results of the static voltage rebound amount and the preset threshold to correct and confirm the preliminary division results.

[0028] Optionally, the macroscopic cycle characteristics dimension is based on the cumulative number of battery cycles and the cumulative capacity degradation rate, and a comprehensive degradation factor is defined here. as follows: ;in: : The current loop count, for example, 2000 times; Rated life cycle count, for example, 2000 cycles; Weighting coefficients, satisfying .

[0029] Judgment range: This is the initial decay stage, at which point the battery has fewer cycle times, lower capacity decay, and its performance is close to that of a new battery. This is the medium-degradation stage, where the number of cycles and capacity decay are moderate, and the battery performance is significantly degraded but can still be used normally. This is the final decay stage, at which point the number of cycles is high, capacity decay is significant, battery performance is severely degraded, and safety risks increase.

[0030] Optionally, to improve accuracy, a micro-correction mechanism needs to be introduced based on the macro-level judgment results. Here, an internal resistance growth factor is proposed. : ;in: Current battery internal resistance; Initial internal resistance.

[0031] In addition, a static voltage rebound threshold is set. Static voltage rebound threshold It is a critical value set based on a large amount of experimental data. When If this occurs, it indicates severe battery polarization, requiring an adjustment to the aging stage. In this embodiment... (Referencing aging experiments of ternary lithium batteries, when the static voltage rebound is 30mV, the recovery ability of active materials decreases significantly).

[0032] The specific rules for micro-correction are as follows: If or Then the initial macroscopic assessment result will be adjusted upwards by one stage, such as initial decay → intermediate decay, intermediate decay → final decay; if and If so, the initial macroeconomic assessment result will be lowered by one stage, such as the medium-term decline. Initial decay; other conditions ( and The preliminary macroeconomic assessment remains unchanged.

[0033] For example, 1. Stage determination for private user vehicle A: Macro-level preliminary determination: Knowing vehicle A's... Second-rate, Substituting into the comprehensive attenuation factor formula: ;because The macroscopic assessment initially identifies it as the initial decline stage; the microscopic correction involves calculating the current internal resistance growth factor. ( (Internal resistance increases gradually); static voltage rebound: (Good rebound, low polarization); According to the micro-correction rule, the initial macro-level judgment result should be maintained at the initial decay stage, without needing to be downgraded, since the macro-level is already at the initial decay stage; therefore, the final judgment result is at the initial decay stage.

[0034] 2. Stage determination for taxi B: Preliminary macro-level determination: Knowing user B's... Second-rate, Substitute into the formula: ;because The macro-level preliminary assessment indicates a mid-term decline; the micro-level correction involves calculating the current internal resistance growth factor. ( (Internal resistance increases by more than 50%); static voltage rebound: (Low rebound, severe polarization); According to the micro-correction rule, the preliminary macro-level judgment result needs to be adjusted up by one stage, that is, from the intermediate decay stage to the final decay stage; then the final judgment result is the final decay stage.

[0035] Furthermore, to verify the necessity of micro-correction, assuming the macroscopic data of a certain vehicle D are: N=600 times, D=7%, then the comprehensive attenuation factor is: It is preliminarily determined to be in the initial decay stage.

[0036] However, the micro-level data is as follows: , According to the revised rules, it needs to be upgraded to the medium decay stage.

[0037] The above example illustrates that relying solely on macroscopic data can lead to the misjudgment of batteries that have already reached a moderate level of aging as being in the initial stage of degradation, resulting in an overly lenient warning threshold and increasing safety risks. A two-dimensional assessment can avoid such misjudgments.

[0038] Furthermore, to ensure that the stage assessment results are updated in real time as the battery ages, this embodiment sets up a stage reassessment triggering mechanism, which specifically includes: timed triggering: after every 100 complete cycles, the stage reassessment process is automatically started, such as once every 25 months for vehicle A and once every 2.8 months for vehicle B; update and push: after the stage assessment results are updated, the BMS automatically updates the aging stage parameters in the system and pushes a battery health status report through the vehicle display screen and user APP. If the battery has entered the medium aging stage, it is recommended to reduce the frequency of fast charging.

[0039] like Figure 2 This diagram illustrates a two-dimensional aging stage division, clearly distinguishing the three stages of initial decay (green), intermediate decay (yellow), and final decay (red) using different colored scatter points. The horizontal axis represents the number of cycles N (0-2000 times), and the vertical axis represents the capacity decay rate (0-40%). The color of each data point represents its final determined aging stage.

[0040] Figure 2 The diagram visually illustrates how microscopic electrochemical parameters correct macroscopic judgment results. Vehicle A (N=500 cycles, capacity decay rate=5%) is located in the lower left region of the diagram, and its macroscopic comprehensive decay factor is shown. =0.13 < 0.3, preliminarily determined to be in the initial decay stage; due to its good microscopic parameters ( =1.1≤1.2, =45mV≥40mV), ultimately maintaining the initial decay stage (green dot); Vehicle B (N=1800 cycles, capacity decay rate=20%) is located in the upper right area of ​​the figure, its =0.48 is in the intermediate decay range, but due to the deterioration of micro parameters ( =1.75>1.5, =20mV<30mV), according to the micro-correction rule, it is adjusted to the final decay stage (red dot); vehicle D (N=600 cycles, capacity decay rate=7%) is located in the middle left region of the figure, its =0.162<0.3 Macroscopically, it is determined to be the initial decay, but microscopic parameters show moderate aging. =1.4, =28mV<30mV) was corrected to the mid-decay stage (yellow dot), which proves the superiority of the two-dimensional judgment model over the single-dimensional method and avoids the risk of misjudgment caused by relying solely on macro data.

[0041] Figure 2The scattering of the dots shows a gradual change from the lower left to the upper right, indicating that as the number of cycles increases and capacity decay intensifies, the battery aging stage gradually transitions from initial decay to final decay, which is consistent with the actual aging pattern of batteries. Meanwhile, the presence of anomalous color points in certain areas (such as a few red dots in the lower left corner) demonstrates the actual effect of the microscopic correction rules. When the battery's microscopic electrochemical state is abnormal, even if the macroscopic parameters are good, it may still be judged as a higher level of aging, which enhances the reliability and safety of the early warning system.

[0042] S3. Based on the aging stage (early aging stage, middle aging stage, and no aging stage) determined in step S2, dynamically adapt the differentiated support vector regression (SVR) model parameters and threshold correction coefficients to generate a basic early warning threshold based on the aging state. This can solve the problem that fixed model parameters cannot adapt to the characteristics of batteries at different aging stages. For example, batteries at the end of their life cycle require higher model sensitivity to avoid false negatives.

[0043] The core idea of ​​the Support Vector Regression (SVR) model is to map input features to a high-dimensional feature space through a mapping function, constructing the optimal regression hyperplane to achieve accurate prediction of continuous values. In this embodiment, the SVR model is used for: Input: Key external factors affecting battery safety (cumulative cycle count) Cumulative mileage Ambient temperature Output: Current battery safety status parameters (internal resistance increase value) Capacity decay value Predicting using the SVR model , This provides accurate aging status input for subsequent threshold calculations, avoiding the lag of directly using measured data.

[0044] The SVR model also includes three key concepts: input feature weights, penalty parameters, and kernel function parameters.

[0045] Input feature weights are coefficients in the SVR model used to measure the influence of different input features on the predicted battery aging state (such as internal resistance growth and capacity decay). The sum of the weights of all input features is always 1. The higher the weight, the greater the contribution of that feature to the model output. In this embodiment, the input features of the SVR model are selected around the driving factors of battery aging, and the weights need to be adjusted according to the aging stage to match the core aging causes of different stages.

[0046] The penalty parameter in the SVR model is used to adjust the model's tolerance for abnormal data (such as sudden voltage fluctuations or instantaneous increases in internal resistance during a charging cycle). A larger penalty parameter value means the model penalizes abnormal data more strictly, avoiding missed detections due to ignoring anomalies. Conversely, a smaller value indicates a higher tolerance for abnormal data, making the model more inclined to fit the overall data trend and avoiding misjudgments due to excessive focus on anomalies. In this embodiment, the core of adjusting the penalty parameter is to match the safety risk levels of different aging stages.

[0047] The kernel function parameter in the SVR model controls the accuracy of the model's fitting to the nonlinear relationships of the data (the SVR model uses the kernel function to map linearly inseparable data to a high-dimensional space, achieving linear fitting). A larger kernel function parameter value indicates a stronger ability of the model to fit local nonlinear features of the data. This means it can accurately capture subtle fluctuations in the data, such as the dramatic fluctuations in battery capacity during the final stages of aging. A smaller value indicates that the model tends to fit the overall linear trend of the data, such as the steady decline in battery capacity during the initial stages of aging, with less attention paid to local fluctuations. In this embodiment, the kernel function parameter is adjusted based on the degree of nonlinearity of battery characteristics at different aging stages.

[0048] Optionally, the specific adaptation rules for SVR model parameters will be shown below to ensure that the model can accurately match the characteristics of batteries at different aging stages, providing adaptability support for the subsequent generation of basic warning thresholds. The specific content is as follows: I. SVR model parameter adaptation in the initial aging stage. In the initial aging stage, the battery aging is relatively mild, and its performance is close to that of a new battery. Aging is mainly driven by macroscopic factors such as cycle count and mileage, and the nonlinear fluctuations in battery characteristics are relatively small. Therefore, the focus of SVR model parameter adaptation in this stage is low sensitivity and coverage of basic features.

[0049] 1. Input Feature Weights: Only three macroscopic features—cycle count, mileage, and ambient temperature—are included, and assigned relatively low weights: the weights for cycle count and mileage are relatively basic, while the weight for ambient temperature is the lowest, to avoid secondary factors excessively affecting the model output; 2. Penalty Parameters: Set to a low level, the penalty for abnormal data identified by the model is lighter, preventing false judgments triggered by occasional minor data fluctuations in new batteries and reducing unnecessary warnings; 3. Kernel Function Parameters: Set to basic values, because the linearity of the characteristics of early-age batteries is relatively high, and there is no need for excessively high local fitting ability. Basic kernel function parameters are sufficient to meet the model's prediction accuracy for battery aging trends.

[0050] II. SVR Model Parameter Adaptation in the Mid-Aging Stage. In the mid-aging stage, battery performance exhibits significant degradation, the driving effect of cycle number on aging becomes significantly stronger, and nonlinear fluctuations in battery characteristics begin to appear. Parameter adaptation in this stage focuses on moderate sensitivity and enhancing the impact of core characteristics.

[0051] 1. Input Feature Weights: Keep the three input features—cycle count, mileage, and ambient temperature—unchanged, but increase the weights of cycle count and mileage. The weight of cycle count is increased the most to highlight the accelerated aging effect of high-frequency use, while maintaining the basic weight of ambient temperature to ensure that the impact of environmental factors is not ignored. 2. Penalty Parameters: Moderately increased compared to the initial aging stage, strengthening the penalty for abnormal data. This avoids missed detections due to fluctuations in battery characteristics without being overly strict and causing false alarms, balancing the accuracy and stability of the warning. 3. Kernel Function Parameters: Slightly increased compared to the initial aging stage, enhancing the model's ability to fit the nonlinear characteristics of the battery, adapting to the irregular fluctuations in parameters such as battery internal resistance and capacity that begin to appear in the middle aging stage, and improving the accuracy of aging state prediction.

[0052] III. SVR Model Parameter Adaptation in the Final Decay Stage. In the final decay stage, battery performance degrades significantly. Besides macroscopic factors like cycle count and mileage, microscopic polarization (such as resting voltage rebound) has a significantly increased impact on safety risks, and the nonlinear fluctuations of battery characteristics become drastic. Parameter adaptation in this stage focuses on high sensitivity and comprehensive feature coverage.

[0053] 1. Input Feature Weights: In addition to cycle count, mileage, and ambient temperature, static voltage rebound is added as an input feature and assigned a high weight. This captures the microscopic aging characteristics of increased internal polarization and decreased recovery ability of active materials in the battery. At the same time, the high weights of cycle count and mileage are maintained to ensure that both macroscopic and microscopic aging factors are fully considered by the model. 2. Penalty Parameter: Set to the highest level, the penalty for abnormal data identified by the model is strongest, ensuring that the model is highly sensitive to small abnormal data of the battery at the end of its decay (such as a sudden increase in internal resistance or a sudden drop in capacity), avoiding missed detections due to lenient thresholds and preventing safety risks. 3. Kernel Function Parameter: Set to the highest level, maximizing the model's ability to fit local nonlinear features, adapting to the drastic fluctuations in parameters such as battery internal resistance and capacity at the end of the decay stage, ensuring that the model can accurately predict the battery aging state and providing reliable support for the stringent basic warning threshold.

[0054] For example, the basic early warning threshold The calculation can be expressed by the following formula: ;in: The charging safety warning benchmark threshold for a new battery (initially in the un-degraded stage) under standard operating conditions (25°C, 1 atm atmospheric pressure, 50% relative humidity); : Aging parameter normalization function, used to normalize the aging parameters , The formula is to uniformly map to the effective range of 0.8 to 1.2. ;and It is an industry-recognized critical value for battery failure and complies with the battery failure standard in the national standard GB / T31484-2015. and They are respectively and The weighting coefficients, and and Complementary ( ); This is the threshold correction coefficient, in the non-decay stage. Mid-decay stage Final decay stage The coefficient value was calibrated through 1000+ battery cycle tests.

[0055] It is important to note the basic early warning threshold. It can represent a broad concept, such as when to warn of charging when the remaining power is at a certain level (charging safety warning), or when the charging temperature reaches a certain level to limit the charging power. Here and in the following text, we will use the direction of charging safety warning to explain and demonstrate the effect of this method, but it should not be understood as a limitation or constraint on this method.

[0056] like Figure 3 The diagram illustrates the changes in SVR model parameters as the battery ages. Four sub-figures illustrate the variations in input feature weights, penalty parameters, kernel function parameters, and threshold correction coefficients at different aging stages. The top-left sub-figure (a) shows the changes in input feature weights. As the battery progresses from the initial aging stage to the final aging stage, the feature weight allocation changes significantly: the initial aging stage (green) only includes three macroscopic features: cycle count (weight 0.4), mileage (weight 0.4), and ambient temperature (weight 0.2), with relatively low weights; the middle aging stage (yellow) increases the weights of cycle count (weight 0.5) and mileage (weight 0.4), while decreasing the weight of ambient temperature (0.1), highlighting the aggravating effect of high-frequency use on battery aging; the final aging stage (red) adds static voltage rebound as an input feature (weight 0.3), while maintaining relatively high weights for cycle count (weight 0.3) and mileage (weight 0.3), ensuring that both macroscopic and microscopic aging factors are fully considered by the model.

[0057] The upper right subplot (b) shows the trend of the penalty parameter, which is used to adjust the model's sensitivity to outlier data. The larger the value, the stronger the penalty for outlier data. The figure clearly shows that the penalty parameter gradually increases from 1.0 in the initial aging stage to 3.5 in the final aging stage. This is completely consistent with the adjustment strategy that matches the safety risk level of different aging stages: the initial aging stage is set to a lower level (1.0) to prevent false alarms caused by occasional small data fluctuations in new batteries; the middle aging stage is moderately increased (2.0) to balance the accuracy and stability of the warning; the final aging stage is set to the highest level (3.5) to ensure that the model is highly sensitive to small outlier data in the final aging battery and avoid false negatives due to a lenient threshold.

[0058] The lower left subplot (c) shows the trend of the kernel function parameter, which controls the model's fitting accuracy to the nonlinear relationships of the data. A larger value indicates a stronger fitting ability to local nonlinear features. The figure shows that the kernel function parameter gradually increases from 0.5 in the initial decay stage to 2.0 in the final decay stage. This is completely consistent with the adjustment strategy to match the degree of nonlinearity of battery characteristics at different aging stages: the initial decay stage is set to the base value (0.5) because the linearity of the battery characteristics in the initial decay stage is relatively high, and there is no need for excessively high local fitting ability; the middle decay stage is slightly increased (1.2) to enhance the model's fitting ability to the nonlinear characteristics of the battery; the final decay stage is set to the highest level (2.0) to maximize the model's fitting ability to local nonlinear features and adapt to the drastic fluctuations in parameters such as battery internal resistance and capacity in the final decay stage.

[0059] The lower right subplot (d) shows the trend of the threshold correction coefficient, which is used to adjust the basic warning threshold to adapt to the safety risk level at different aging stages. The figure shows that the threshold correction coefficient gradually increases from 0.9 in the initial aging stage to 1.2 in the final aging stage, which is completely consistent with the coefficient value calibrated through 1000+ battery cycle experiments: [The last part, "no aging stage," appears to be an unrelated fragment and is omitted from the translation.] =0.9, intermediate decay stage =1.0, final decay stage =1.2. This gradual increase reflects the need for a more stringent warning threshold to address higher safety risks as batteries age more.

[0060] S4. Analyze user charging behavior data, calculate the behavioral impact coefficient to characterize the risk level of user charging habits, and adjust the basic warning threshold. Specifically: Extract risk features such as charging frequency, charging depth, charging method, and charging time from the user charging behavior data collected in step S1. Divide each risk feature into multiple levels according to its risk level, assign weights according to the degree of influence of the feature, and accumulate the weighted sums to obtain the behavioral impact coefficient. Filter out short-term abnormal behaviors to ensure that the behavioral impact coefficient reflects true habits. Use the behavioral impact coefficient to adjust the basic warning threshold; if the risk is high, the basic warning threshold is tightened; if the risk is low, it is slightly loosened, balancing warning accuracy and user experience.

[0061] S5. Based on multi-physics parameters, a multi-physics coupling factor is calculated to quantify the impact of extreme environments on battery performance, including three key dimensions: altitude, temperature, and humidity. When calculating the single-field coupling factor: the altitude factor increases with increasing altitude, reflecting decreased heat dissipation; the temperature factor increases at both low and high temperatures, reflecting electrolyte conduction and SEI film issues; the humidity factor increases at high humidity, reflecting insulation failure. Multiplying the single-field coupling factors results in a multi-physics coupling factor, reflecting the superposition of environmental risks. This factor is used to correct the threshold after behavioral influence coefficient correction; the higher the environmental risk, the stricter the threshold. Ultimately, this achieves dynamic adaptation of early warnings throughout the battery's entire lifecycle and under complex operating conditions, generating dynamically predicted charging safety early warning thresholds.

[0062] Example 2: The behavioral influence coefficient in step S4 is not an isolated value, but a comprehensive, numerical measure of the strength of habit or preference reflected in a user's historical charging behavior in a specific charging scenario (e.g., a combination of location, time of day, and vehicle status). A higher coefficient indicates a more stable and regular charging behavior pattern in that scenario, meaning stronger habituality. Conversely, a lower coefficient suggests that the user's charging behavior in that scenario is more random and accidental, and does not constitute a stable habit.

[0063] In practical applications, this coefficient is a key decision-making basis for subsequent personalized charging guidance strategies (such as pushing charging reminders at specific times, recommending discounted electricity packages for specific charging stations, or predicting acceptable discharge periods for users in V2G (Vehicle-to-Grid) applications). A high-coefficient charging habit will be reinforced and catered to by the system; while for low-coefficient behaviors, the system may attempt to optimize guidance or use it as a secondary reference.

[0064] The calculation of the behavior influence coefficient includes the following steps: extracting multiple charging behavior features from the user's historical charging behavior data; quantifying and classifying each charging behavior feature; and multiplying the quantified value of each feature's level by its corresponding weight coefficient and summing the results to obtain the final behavior influence coefficient. Specifically, this includes the following: 1. Charging behavior features refer to multiple dimensions or attributes that can be extracted from the user's historical charging behavior data to describe and characterize the user's charging habits. These features are the basic elements for building user profiles. To comprehensively and accurately depict user habits, in this embodiment, the charging behavior features may include at least, but are not limited to, the following: 1. Charging start time feature: recording the specific time point at which the user connects to the charging pile for each charging behavior. This feature is mainly used to analyze the user's charging time preferences throughout the day, for example, whether they are accustomed to charging immediately after returning home (during evening peak hours), charging during off-peak hours at night, or charging at work.

[0065] 2. Vehicle status at the start of charging: This mainly refers to the remaining percentage of the vehicle's battery charge, or State of Charge (SOC). This feature reflects the user's range anxiety or usage plan. For example, some users tend to charge only when the SOC is below 20%, while others may look for opportunities to recharge when the SOC is above 50%.

[0066] 3. Charging geographical location characteristics: Record the specific geographical location of each charging behavior, such as private charging piles at home, public charging piles at work, fast charging stations in specific commercial areas, etc. This characteristic is the most important basis for distinguishing different charging scenarios (such as home charging, workplace charging, emergency charging).

[0067] 4. Charging duration or target SOC characteristic: Records the duration of a single user charging session, or the target SOC preset by the user in the charging app or vehicle settings, such as charging to 90% or 100%. This characteristic helps determine the user's charging speed requirements and battery maintenance habits.

[0068] 5. Charging frequency characteristics: This refers to the number of times a user initiates charging behavior within a certain time window (e.g., one week or one month). This characteristic reflects the user's vehicle usage intensity and dependence on the vehicle's range.

[0069] 6. Electricity Price Sensitivity Characteristics: By analyzing the distribution of users' charging behavior during different electricity price periods (such as peak, flat, and off-peak electricity prices), we can assess users' sensitivity to charging costs. For example, if the vast majority of a user's charging behavior occurs during off-peak hours, then their electricity price sensitivity is high.

[0070] II. Quantitative grading refers to the process of transforming the extracted, typically continuous or discrete, raw feature data into unified, standardized grade scores. Directly using raw data (such as timestamps down to the second or SOC accurate to 0.01%) for calculation is complex and unstable; even small fluctuations can lead to significant differences in results. Quantitative grading aims to map this raw data onto a finite, ordered set of grades, thereby capturing its core, statistically significant patterns.

[0071] For example, the charging start time characteristic can be divided into multiple meaningful time periods, and each time period can be assigned a level, such as: Level 5: Deep off-peak electricity period at night (00:00-06:00); Level 4: Evening to nighttime shallow off-peak period (22:00-24:00); Level 3: Evening peak household charging period (18:00-22:00); Level 2: Daytime flat price period (07:00-18:00); Level 1: Daytime peak electricity price period (specific time, such as 10:00-12:00).

[0072] Third, weighting coefficients are numerical values ​​assigned to each charging behavior feature to represent its importance in comprehensively evaluating a user's charging habits. Different features contribute differently to defining a habit. For example, for identifying the core habit of charging at home at night, the importance of charging location features (whether the user is at home) and charging start time features (whether it is at night) is usually much higher than the SOC feature at the start of charging. Therefore, higher weighting coefficients need to be assigned to the former. These weighting coefficients can be preset by domain experts based on experience, or they can be adaptively learned and dynamically adjusted based on a large amount of user data through machine learning algorithms.

[0073] Fourth, it also includes an abnormal behavior filtering mechanism: This mechanism is a crucial preliminary step to ensure the accuracy of the behavior influence coefficient calculation. Its core idea is that users' historical behavior data includes both habitual behaviors that reflect their long-term, stable preferences and short-term abnormal behaviors caused by special circumstances. If all data is used for calculation without distinction, abnormal behaviors will contaminate the data, distort the final analysis results, and lead to misjudgments of user habits by the system.

[0074] Habitual behavior refers to behaviors that are highly repetitive, regular, and predictable over a relatively long period of time. For example, a commuter might charge their phone in their garage 90% of the time, usually between 9 pm and 11 pm.

[0075] Short-term abnormal behavior refers to behaviors that significantly deviate from a user's usual patterns, triggered by accidental, non-repetitive events. For example, the aforementioned commuter may have charged their car twice during a long road trip at a fast-charging station in a highway service area; or a friend may have charged their car at a charging station in another city because the vehicle was lent out. These behaviors do not represent the user's inherent charging habits.

[0076] Therefore, it is necessary to establish a clear set of rules for identifying abnormal behavior. Before feature extraction and quantification, the raw charging behavior data should be cleaned to identify and exclude these abnormal behavior events, ensuring that the data used for calculation can reflect the user's true intentions and habits to the greatest extent.

[0077] For example, take Mr. Zhang, a typical electric vehicle owner, who lives in District A of a city and works in District B. He commutes by car every day and has installed a private charging station at his residence.

[0078] The system retrieved Mr. Zhang's charging records from the backend database over the past 90 days, totaling 40 records. To simplify the explanation, we list 8 representative records:

[0079] 1 2025-06-10 21：35 City Area A - Family Garage 28% 2 2025-06-15 22：05 City Area A - Family Garage 35% 3 2025-06-20 21：50 City Area A - Family Garage 31% 4 2025-06-28 14：10 C City - Highway Service Area 15% 5 2025-07-05 10：15 City Area A - Family Garage 25% 6 2025-07-12 10：30 City B Zone - Company Parking Lot 45% 7 2025-07-18 21：45 City Area A - Family Garage 33% 8 2025-07-25 00：30 City Area A - Family Garage 22% ... ... ... ... ... 40 ... ... ... ...

[0080] Before calculating the impact coefficient of behavior, abnormal behavior filtering is first performed. The following rules can be set: Rule 1 (Geographical change rule): If the geographical location of a certain charging behavior is more than 50 kilometers away from the location where more than 80% of the charging behaviors in history occurred (i.e., high-frequency location, in this case, city A area - family garage), and the number of times that location has appeared in the past 90 days is less than 2, then it is judged as abnormal behavior.

[0081] Rule 2 (Time Pattern Deviation Rule): Analyze the main distribution range of a user's charging start time at high-frequency locations. If the start time of a charging session at a high-frequency location deviates from the center of the main time distribution range by more than 6 hours, it is considered abnormal behavior.

[0082] Rule 3 (Location and Time Combination Rule): If a charging behavior occurs in a non-high-frequency location (such as the company or shopping mall) and the charging time occurs during the user's usual inactive period (such as late at night), it is judged as abnormal behavior (for example, it may be a designated driver or a friend's temporary car use).

[0083] Next, the data is cleaned. For example, record 4: the charging location is a highway service area in City C. This location is more than 50 kilometers away from Mr. Zhang's family garage, and it only occurred once. According to rule 1, the system marks record 4 as abnormal behavior and excludes it. This record is likely an emergency charging incident by Mr. Zhang during a long trip and does not reflect his daily habits.

[0084] Record 6: The charging location is the company parking lot in District B of the city. Although this location may appear multiple times, the charging time is 10:30 AM, which is daytime. Assuming the system analysis reveals that 95% of Mr. Zhang's charging behavior occurs after 8:00 PM, this daytime charging significantly deviates from his primary time pattern. However, for more refined judgment, the system will first count the number of times charging is done in the company parking lot. Suppose that out of all 40 records, Mr. Zhang only charged at the company twice, both times during weekday daytime. This might be a backup habit, not an anomaly. However, if such charging is very rare, such as only once, the system can classify it as a low-weight event or a potential anomaly based on a variation of Rule 2 (frequency and time analysis for secondary locations). In this example, to highlight the filtering effect, we assume this is because he forgot to charge one night and urgently charged at the company the next day; the system classifies this as an abnormal behavior and excludes it.

[0085] After cleaning, the effective dataset used to calculate the impact coefficient of charging at home overnight will no longer include records 4 and 6. Assuming that this mechanism eliminates 5 outlier records out of all 40, leaving 35 valid records, based on these 35 cleaned valid records, the following three core features are extracted and quantified for the scenario of charging at home overnight:

[0086] (1) Characteristics of charging start time Data analysis: Analysis of the start time of 35 valid records revealed that the vast majority were concentrated between 21:30 and 01:00.

[0087] Quantitative grading rules: Grade 5 (G=5): 22:00-02:00 (deep habitual period, off-peak electricity price); Grade 4 (G=4): 21:00-22:00 or 02:00-04:00 (habitual buffer period); Grade 3 (G=3): 19:00-21:00 (regular hours); Grade 2 (G=2): 04:00-07:00 (early morning hours); Grade 1 (G=1): other times; Quantitative results: Count the number of times each of the 35 records falls into each grade and calculate its weighted average grade. Assuming 25 times fall into Grade 5, 8 times into Grade 4, and 2 times into Grade 3, then the quantitative grade value of this feature is: .

[0088] (2) SOC characteristics at the start of charging Data analysis: Analysis of the initial SOC of 35 valid records revealed that it was generally low, mainly ranging from 20% to 440%.

[0089] Quantitative grading rules: Level 5 (G=5): SOC < 30% (strong demand range); Level 4 (G=4): 30% ≤ SOC < 40% (relatively strong demand range); Level 3 (G=3): 40% ≤ SOC < 60% (moderate demand range); Level 2 (G=2): 60% ≤ SOC < 80% (weak demand range); Level 1 (G=1): SOC ≥ 80% (opportunistic recharging); Quantitative results: Assuming the statistical results are: 18 times falling into Level 5, 12 times falling into Level 4, and 5 times falling into Level 3, then the quantitative level value of this feature is: .

[0090] (3) Geographical location characteristics of charging Data Analysis: Since we are calculating the behavioral impact coefficient in the scenario of being at home, all 35 valid data points occurred in this location, and the consistency of this feature is very high.

[0091] Quantitative grading rules: Level 5 (G=5): The proportion of charging times occurring in the target scenario (home garage) accounts for >90% of the total effective charging times; Level 4 (G=4): The proportion is between 75% and 90%; Level 3 (G=3): The proportion is between 50% and 75%; Level 2 (G=2): The proportion is between 25% and 50%; Level 1 (G=1): The proportion is <25%; Quantitative result: In the 35 cleaned records, 100% occurred in the home garage. Therefore, the quantitative grade value of this feature is... Then, the behavioral impact coefficient is calculated using a weighted summation method to obtain the final behavioral impact coefficient.

[0092] The behavioral influence coefficient given in this embodiment The calculation formula can be expressed as: ;in: : Index of charging behavior characteristics; : The total number of behavioral features used for calculation (in this example, ); : No. The weight coefficient corresponding to each behavioral feature. This coefficient is either preset or dynamically learned, reflecting the importance of that feature in evaluating overall habits. The sum of all weight coefficients. The value is usually 1; if it is 1, the denominator can be omitted. : No. The grade value is obtained by quantifying and classifying each behavioral feature; : An optional scaling factor used to map the calculation results to a more business-meaning numerical range, such as 0 to 100. Here we set... This ensures the result ranges from 0 to 100 (because the highest level is 5, 5 * 20 = 100).

[0093] Based on expert experience, the importance of defining household nighttime charging habits is ranked as follows: Location > Time > State of Charge (SOC). Therefore, the following weighting coefficients are assigned: Weight of charging location characteristics. Weighting of charging start time feature Weights of SOC features at the start of charging Weighted sum: .

[0094] Substitute the previously obtained feature quantization level values ​​and weight coefficients into the formula: ; Calculations show that Mr. Zhang's behavior of charging his phone at home overnight has an impact coefficient of 95.44 (out of 100). This is a very high value, strongly demonstrating that this behavior is an extremely stable and ingrained charging habit for Mr. Zhang.

[0095] To more clearly demonstrate the technical effect of the abnormal behavior filtering mechanism in this invention, a comparative experiment was conducted: calculations were performed directly using the original data without filtering, that is, calculations were performed using all 40 original data entries.

[0096] First, re-quantify and classify: geographic location characteristics. Of the 40 records, 38 were from the home (assuming the other two abnormal records include one on the highway and one at the company), the proportion is 38 / 40 = 95%. According to the classification rules, the level remains 5. .

[0097] Charging start time characteristics Now we need to add the times for records 4 (14:10) and 6 (10:30), both of which fall in the lowest level, 1. (For simplicity, only two level 1 records will be added): ; As can be seen, due to the inclusion of abnormally low data, the average rank of the time feature dropped significantly from 4.66 to 4.125.

[0098] SOC characteristics at the start of charging Similarly, SOCs for records 4 (15%) and 6 (45%) were added, with 15% belonging to level 5 and 45% belonging to level 3.

[0099] The average grade of the SOC feature also decreased from 4.37 to 4.025.

[0100] Recalculate the behavioral impact coefficient : ; By comparison, the unfiltered value decreased by 4.59. This means that without an abnormal behavior filtering mechanism, a few accidental, non-habitual charging behaviors (such as emergency charging during long-distance travel) would significantly dilute the final behavior impact coefficient, leading to a lower calculated coefficient. However, by introducing an abnormal behavior filtering mechanism, data noise can be accurately identified and removed, resulting in a purer dataset that better reflects the user's core habits. The final behavior impact coefficient (95.44) is higher than the unfiltered one (90.85), more accurately quantifying the strength of user habits. It precisely and robustly extracts and quantifies the user's true charging habits, providing a solid data foundation and technical support for achieving truly personalized, efficient, and intelligent charging guidance services.

[0101] like Figure 4 This diagram compares the original behavior impact coefficient calculated by the battery management system with the behavior impact coefficient obtained after filtering for abnormal behavior. This comparison demonstrates how the system uses anomaly detection algorithms to eliminate the impact of irregular behavior (such as occasional high charging speeds or temporary irregular charging), thus ensuring the accuracy and stability of the calculation results. The red curve (original behavior impact coefficient, unfiltered) represents the change of the originally calculated behavior impact coefficient over time without any abnormal behavior filtering. The data points in the graph show significant spikes in certain time periods (such as days 6, 11, and 15). These spikes indicate abnormal user behavior, such as frequent fast charging or irregular charging habits. Such abnormal data often misleads subsequent battery status assessments and reduces the accuracy of the warning system. For example, the sudden increases on days 6 and 15 reflect extreme deviations in user charging methods or charging times.

[0102] The green curve (filtered behavior impact coefficient) represents the behavior impact coefficient obtained after processing by the abnormal behavior filtering mechanism. By defining reasonable thresholds and pattern detection algorithms, the system removes abnormal data points (such as outliers on day 6 and day 15), and the remaining data shows a smoother and more stable trend. This indicates that the filtered data more accurately reflects the user's actual charging behavior, avoiding the impact of sudden anomalies on battery assessment. Therefore, the green curve in the figure demonstrates a more predictable and physically consistent battery behavior pattern, providing more accurate input for subsequent battery life prediction and risk assessment.

[0103] pass Figure 4The positive impact of abnormal behavior filtering on the calculation of behavior influence coefficients in the battery management system is clearly evident. The red curve shows fluctuations and outliers in the original data, while the green curve shows the stable trend after abnormal behavior filtering. This filtering mechanism can effectively improve the accuracy of the early warning system and avoid misleading judgments caused by occasional abnormal behaviors.

[0104] In obtaining the behavioral influence coefficient After that, interval division is required to determine the basic early warning threshold. To make corrections, for example, the following is a specific four-level judgment range and adjustment scheme: Level 1: Highly predictable users, behavioral impact coefficient range: Risk assessment: Extremely low risk. Users in this range have extremely stable and consistent charging habits. For example, they will charge their phones overnight if the battery level drops below 40% after arriving home each day. Users can be highly trusted to follow their habits independently.

[0105] Adjustment strategy: Significantly relax the early warning threshold, threshold adjustment amount : ; Corrected warning threshold : For these types of users, the system won't issue a low battery warning until the vehicle's battery level drops to 15%. This minimizes unnecessary interruptions. A user with good habits, even with a SOC at 18%, is likely already on their way home and planning to charge within a few hours. An early warning (at 20%) would be redundant information for them and could even be annoying. By relaxing the threshold to 15%, the system demonstrates trust in the user's habits, significantly improving the user experience.

[0106] Level Two: Users with good habits, behavioral influence coefficient range: Risk assessment: Low risk. Users in this range have developed relatively stable, but occasionally changing, charging habits. Adjustment strategy: Moderately relax the warning threshold, with the threshold adjustment amount... : ; Corrected warning threshold : The system issues an alert when the SOC drops to 18%. This also reduces the frequency of alerts, but retains a higher safety redundancy than for Tier 1 users. This moderate relaxation reflects personalized care while ensuring that the system can still provide timely reminders when users occasionally deviate from their habits, such as returning home later than usual due to overtime work.

[0107] Level 3: Users with fluctuating behavior, range of behavioral impact coefficient: Risk Assessment: Moderate to high risk. Users in this range exhibit irregular charging behavior or are in the process of forming a habit. They are highly likely to forget to charge. Adjustment Strategy: Moderately tighten the warning threshold; threshold adjustment amount... : ; Corrected warning threshold : The system will issue an early warning when the State of Charge (SOC) drops to 25%. For these users, ensuring a high success rate of charging takes precedence over avoiding interruptions. Early warnings give users more time to react, plan their next steps, or find a charging station. For example, if a user receives a notification at 25% battery, they might still be far from home or work. They have more options: whether to drive directly to a charging station or continue driving home. This strategy effectively reduces the risk of users getting stuck in a low-battery situation.

[0108] Level 4: Random or new users, behavioral impact coefficient range: Risk Assessment: High risk. Users in this range generally lack identifiable charging habits, or, due to being newly registered users, the system's historical data is insufficient for effective judgment. Adjustment Strategy: Significantly tighten the warning threshold; threshold adjustment amount... : ; Corrected warning threshold : The system will issue a warning when the State of Charge (SOC) drops to a very conservative level of 30%. This is the highest level of warning strategy, designed to provide users with maximum safety. For a user whose behavior patterns are completely unknown, the system must take the most cautious approach. Issuing a warning at 30% battery ensures that even if the user is in a remote area or unfamiliar with nearby charging facilities, there is sufficient time and range to cope with any unexpected situations. While this may cause more inconvenience initially, it is crucial for building basic user trust in the system and avoiding catastrophic consequences (such as the vehicle breaking down en route).

[0109] Example 3: Optionally, the multiphysics coupling factor in step S5 aims to quantify multiple external environmental physical quantities (physical fields) affecting the performance and safety of the electric vehicle battery system into a unified index that can be directly used to correct system control parameters (such as warning thresholds) through a mathematical model. It does not measure the environment itself, but rather the degree of negative impact the environment has on the battery system. The term "coupling" refers to the complex relationship of interaction and mutual influence between these external physical fields (such as atmospheric pressure, temperature, and humidity fields) and the electrochemical, thermodynamic, and electrical stress fields inside the battery. A high multiphysics coupling factor indicates a greater impact of the current external environment on the battery system, increasing the potential risk to the system.

[0110] The multiphysics coupling factors specifically include: 1. Altitude coupling factor, which is a sub-coefficient specifically used to quantify the negative impact of high-altitude environments on the heat dissipation performance of battery systems. Due to the reduced air density and decreased heat dissipation efficiency, under the same charging and discharging conditions, the temperature rise rate of the battery system will be faster and the temperature point to reach thermal equilibrium will be higher when the vehicle is driving or charging in high-altitude areas. This not only triggers the thermal limitation of charging power and prolongs charging time, but more seriously, it increases the risk of battery thermal runaway; 2. Low temperature coupling factor, which is a sub-coefficient specifically used to quantify the negative impact of low-temperature environments on the electrochemical performance of batteries. Under low-temperature conditions, the viscosity of the electrolyte inside the battery will increase sharply. The increased electrolyte viscosity leads to a significant increase in the resistance (i.e., migration impedance) of lithium ions shuttling through it, and a slower ion diffusion rate. Specifically, in vehicles, this manifests as a significant decrease in available discharge power, slow charging speed or inability to charge, and a larger voltage drop at the battery terminals under load, resulting in a large deviation in the remaining charge SOC estimated by the vehicle system based on voltage; 3. Anti-humidity coupling factor, which is a sub-coefficient specifically used to quantify the negative impact of high-humidity environments on the insulation performance of the vehicle's high-voltage electrical system. In high-humidity environments, water molecules in the air can penetrate almost any pore. When the ambient temperature changes, especially in humid environments transitioning to air-conditioned parking lots or other temperature-differential scenarios, water vapor easily condenses into droplets or films on the cold surfaces of electrical components. While pure water is a poor conductor, the condensation dissolves impurities in the air and contaminants on the component surfaces, creating a conductive path. This significantly reduces the insulation resistance between high-voltage components and ground. Once the insulation resistance drops below safety standards, it will trigger the vehicle's insulation fault alarm, potentially leading to a high-voltage system power outage or, in extreme cases, the risk of electric shock or short circuits.

[0111] For example, a set of calculation methods is proposed here to convert environmental physical quantities into final coupling factors, wherein each field coupling factor is designed as a dimensionless number between 0 and 1, where 0 represents no negative impact and 1 represents reaching the set maximum negative impact.

[0112] Altitude coupling factor : ;in: : The current altitude of the vehicle's location (unit: meters); Based on the altitude, altitudes below this level are considered to have no significant impact on heat dissipation. In this embodiment, the altitude is set as follows: rice; The critical altitude is defined as the altitude at which heat dissipation is considered to be at its maximum, requiring the highest level of protection. In this embodiment, the altitude is set as follows: rice.

[0113] Low temperature coupling factor : ;in: : Current temperature of the battery pack or environment (unit: degrees Celsius); Reference temperature: Temperatures above this level are considered to have no significant negative impact on battery performance. In this embodiment, it is set as follows: ; Critical low temperature: Below or equal to this temperature, battery performance is considered to have reached its maximum degradation. In this embodiment, it is set as follows: The index of 1.5 is used to reflect the trend that the performance deteriorates non-linearly as the temperature decreases.

[0114] Humidity coupling factor : ;in: : Current relative humidity (unit: percentage); Reference humidity: humidity levels below this are considered to have no significant impact on insulation performance. In this embodiment, the humidity level is set as follows: ; Critical humidity: A humidity level above or equal to this value is considered to pose the greatest electrical safety risk. In this embodiment, it is set as follows: Index 2 is used to reflect the physical characteristic that the risk of condensation increases sharply as the relative humidity approaches saturation.

[0115] Multiphysics coupling factor The calculation can be obtained by weighted summation of each sub-factor to reflect the differences in importance of different environmental factors when targeting a specific early warning objective. ;in, , , These are the weight coefficients of the three sub-factors, and These weights can be dynamically adjusted based on the different targets of the warning.

[0116] For example, the weight for a charging thermal runaway warning can be set as follows: , , For winter range warnings, the weight can be set as follows: , , For high-voltage insulation fault early warning, the weight can be set as follows: , , .

[0117] Based on the real-time calculated integrated multiphysics coupling factor (Range 0-1), it is divided into four levels, and corresponding early warning and guidance strategies are implemented for different levels: Level 1: Ideal environment, multiphysics coupling factor range: This range corresponds to the vehicle's operating conditions in plains areas (below 1000 meters above sea level), during mild spring and autumn seasons (temperatures between 5°C and 25°C), and in dry weather (relative humidity below 80%). This is the operating range where the battery system performs optimally and has the greatest safety margin. Within this range, the actual usable range of the battery is highly consistent with the instrument display, and the charging system can operate at its maximum efficiency.

[0118] Threshold adjustment strategy: moderately relax the threshold, the system believes that the current environment is friendly and the user has enough time and choice to respond.

[0119] SOC adjustment amount ( ): Regarding the revised warning threshold After further revisions, the final charging safety warning threshold was obtained. : This value indicates that the battery level is below 18%, the current driving environment is ideal, the battery performance is stable, and you can continue driving as planned. Charge when convenient. It supports the use of any public charging station, including super-fast charging. By slightly lowering the warning threshold, premature interruptions to users are reduced, improving the user experience in good road conditions. At the same time, users are clearly informed that fast charging is available, giving them maximum charging freedom.

[0120] Level 2: Moderate stress environment, multiphysics coupling factor range: This range corresponds to hilly or plateau areas (altitude 1000-2500 meters), or southern cities in winter (temperature 0℃ to 5℃), or the high-humidity plum rain season (humidity 80%-90%). Under these conditions, the actual usable battery life will decrease to some extent due to the low temperature, and high altitude and high humidity also pose initial challenges to charging safety.

[0121] Threshold adjustment strategy: Standard warning with additional prompts; SOC adjustment amount ( ): Regarding the revised warning threshold After further revisions, the final charging safety warning threshold was obtained. : This value indicates that the battery level is below 23%. Due to the current low temperature and high altitude, the remaining driving range may be slightly shorter than expected. It is recommended to start planning a charge. To ensure stable charging and battery health, AC charging stations are the preferred option. This early warning provides users with a buffer time to mitigate the impact of reduced range. The introduction of charging method recommendations guides users to choose the gentler AC charging in less-than-ideal environments. This ensures a higher charging success rate (fast charging is prone to tripping at low temperatures) and better protects the battery.

[0122] Level 3: High-pressure environment, multiphysics coupling factor range: This range corresponds to driving in high-altitude areas (2500-4000 meters above sea level), experiencing severe winters in the north (temperatures -15°C to 0°C), or in extreme high-humidity weather such as heavy rain or snow. Under these conditions, the actual usable battery range will significantly decrease, charging capacity will be severely limited at low temperatures, and high humidity and high altitude will pose a serious threat to electrical insulation and heat dissipation.

[0123] Threshold adjustment strategy: Significantly advance and force recommendations; SOC adjustment amount ( ): Regarding the revised warning threshold After further revisions, the final charging safety warning threshold was obtained. : Examples of this value could be:

Warning

Charging Recommendation

[0124] Issuing a warning at a relatively safe battery level of 30% significantly enhances user safety. The warning is no longer just a suggestion, but a mandatory guide. Clearly informing users that fast charging is limited avoids the predicament of arriving at a fast charging station only to find they cannot charge, directly guiding them to find the most reliable charging method – a clear demonstration of prioritizing safety.

[0125] Level 4: Extremely hazardous environment, multiphysics coupling factor range: This range corresponds to the superposition of multiple harsh conditions, such as the Sichuan-Western Plateau in winter (altitude above 4000 meters, temperature below -15℃, accompanied by high humidity caused by wind and snow). Under this range, the battery is at its performance freezing point, the driving range is extremely unreliable, and it may lose power at any time due to low temperature protection. The charging behavior itself is also accompanied by an extremely high risk of thermal runaway or insulation failure.

[0126] Threshold adjustment strategy: highest level warning and provision of emergency guidance; SOC adjustment amount ( ): Regarding the revised warning threshold After further revisions, the final charging safety warning threshold was obtained. : An example of this value could be: [Critical Alert - Immediate Action] Battery power 40%! You are in an extremely dangerous environment, and your vehicle's power may be limited at any time! Please stop immediately in a safe location and seek assistance or use emergency power! Please contact a service center for support if needed.

[0127] Issuing the highest level alarm when the battery level reaches 40% is no longer a simple reminder about remaining battery life, but a survival safety warning, thereby maximizing the protection of the user and vehicle safety.

[0128] like Figure 5 The visualization verification results of the multiphysics coupling factor calculation are presented. Figure 5 The x-axis represents altitude (0-5000m); the y-axis represents ambient temperature (-30-15℃); and the z-axis represents the multiphysics coupling factor. The surface color gradually changes from blue (0-0.25) to red (0.75-1.0), corresponding to a four-level risk classification: ideal environment → moderate pressure environment → high pressure environment → extreme danger environment. From the surface morphology, when the altitude exceeds 4500m, the z-value rapidly approaches 1 (e.g., at H=5000m and T=-20℃). =0.98), which conforms to the piecewise logic that the heat dissipation effect is maximized when the altitude is above the critical value; when the temperature is below -20℃, the z value increases non-linearly (e.g., when H=3000m, T=-30℃). =0.85), verifying the exponential function design (exponent 1.5 in the formula) that the electrolyte conductivity deteriorates nonlinearly at lower temperatures; while the humidity is fixed at 95%, the result is calculated using the humidity coupling factor formula. =((95-80) / (98-80))²≈0.69, after superimposing altitude and temperature factors, the final coupling factor accurately reflects the multi-field collaborative risk; the extreme point marked in black (H=4500m, T=-20℃) corresponds to =0.98, which is in the red risk zone. It perfectly matches the application scenarios where the coupling factor is ≥0.75 in extremely dangerous environments and the highest level of early warning is required (such as triggering an alarm when SOC=40%), thus avoiding the limitations of traditional single-environment compensation.

[0129] Example 4: Optionally, this method also includes an adaptive triggering mechanism for stage switching. This mechanism is a core function designed to address the natural performance degradation of electric vehicle power batteries throughout their entire lifecycle. Battery aging is not a linear process; key parameters such as capacity, internal resistance, and charge / discharge characteristics exhibit different rates and characteristics of change at different stages of their lifespan. A warning model or charging strategy designed for a brand-new battery may be completely inapplicable to an old battery that has been used for five years and has experienced a 20% capacity degradation, and may even pose safety risks.

[0130] Therefore, the core idea of ​​this mechanism is to no longer treat the battery as a static component, but rather to divide its lifecycle into multiple logical stages, such as the healthy stage, the sub-healthy stage, and the degradation stage. The system no longer relies on a single model, but instead presets or dynamically generates a set of optimal system parameters for each stage, such as warning thresholds, charging power curves, and weights for behavioral patterns. The adaptive triggering mechanism is responsible for detecting whether the battery has migrated from one stage to the next and automatically switching parameters at the appropriate time.

[0131] The adaptive triggering mechanism for phase switching includes the following: setting multiple phase re-determination trigger conditions, including the cycle count increment condition and the capacity decay rate condition; when any trigger condition is met, the aging phase re-determination process is automatically started; when the determination phase changes, the system parameters are automatically updated and the status change information is pushed.

[0132] The following simulates a real-world scenario to describe this mechanism: Mr. Li purchased a new electric vehicle. For the first two years, the vehicle's range was very stable, and the charging speed was fast. The system determined the battery to be in a healthy state. At this time, various system parameters were relatively aggressive. For example, the low battery warning threshold was dynamically adjusted to a lower level, such as 18% SOC, based on Mr. Li's good driving habits, to reduce unnecessary disturbances; simultaneously, the system allowed for fast charging at maximum power under suitable conditions.

[0133] Comparison 1: Without this mechanism. After three years of use, Mr. Li noticed a significant reduction in his vehicle's range, especially in winter. With the same battery percentage, the actual mileage was far less than before. However, his in-car system still only issued a warning when the State of Charge (SOC) dropped below 18%. This caused a series of problems. For example, he might leave work with an SOC of 20%, expecting ample time to get home, but the range would drastically decrease en route, ultimately causing the vehicle to break down not far from home, resulting in considerable inconvenience.

[0134] Comparison 2: When this mechanism is involved, the system uses the loop count increment condition and capacity decay rate condition as the triggering conditions for stage re-determination.

[0135] 1. Cycle Count Increment Condition: The system continuously tracks the equivalent full charge-discharge cycle count of the battery. When the system detects that the battery's cycle count has increased by, for example, 150 cycles since the last comprehensive evaluation, this acts as a milestone, indicating that the battery has undergone significant wear and tear. At this point, regardless of the battery's performance, the system will automatically trigger a stage reassessment process.

[0136] Capacity degradation condition: Simultaneously, the system continuously monitors the battery's State of Health (SOH), i.e., its actual maximum usable capacity, through the Battery Management System (BMS). In the third year of Mr. Li's vehicle use, the system accurately calculated through a complete charging process that the battery's SOH had decreased from 95% at the beginning of the previous stage to 88%. This significant degradation of over 5% immediately triggered the stage reassessment process.

[0137] When any of the above conditions are triggered, the system will automatically run a more complex diagnostic algorithm in the background during idle periods such as nighttime charging, to conduct a comprehensive check on the battery's internal resistance, consistency, voltage plateau, and other parameters. Based on the diagnostic results, the system determines that the battery has officially entered a sub-healthy stage.

[0138] Subsequently, the system automatically performs the following operations: Automatically updates system parameters: The entire system's basic settings are switched to a sub-healthy stage mode library. For example, the calculation model for the basic warning threshold automatically becomes more conservative; even if Mr. Li's habits remain unchanged, the warning may be triggered earlier at 25% SOC. Simultaneously, the maximum allowable power of DC fast charging is automatically reduced by 15% to mitigate the impact on the aging battery.

[0139] Push notification of status change: The next time the vehicle is started, the central control screen will display a clear notification: "Dear user, the system has detected that your vehicle's power battery has entered a new performance phase. To optimize its long-term health and your driving safety, we have automatically adjusted the range prediction model and charging protection strategy for you. You may notice that the range display is more conservative, which is normal and designed to provide you with more reliable trip planning."

[0140] This mechanism enables intelligent tracking and adaptation to the battery aging process. Unlike traditional systems where fixed models lead to later warning failures and increased user risk, this mechanism is able to adapt to changing times and proactively adjust itself at inflection points where battery performance undergoes fundamental changes. This ensures the effectiveness of warnings and the safety of charging strategies throughout the battery's entire lifecycle, significantly improving long-term reliability and user experience.

[0141] like Figure 6 The diagram illustrates the change in battery health status (SOH) over the number of usage cycles, and demonstrates the switching mechanism from a healthy to a degraded stage using threshold indicators. The three dashed lines in the diagram represent the thresholds for different stages, helping the system to adjust management strategies and trigger relevant warnings in a timely manner during battery aging.

[0142] The blue curve represents the gradual decrease in battery health status (SOH) as the number of usage cycles increases. Battery health typically decreases with increasing charge-discharge cycles, and the curve shows a linear trend in this change. When the battery's usage cycles reach a certain threshold, the battery's SOH will enter different stages, and the system will make corresponding adjustments based on these changes.

[0143] The green dashed line (health stage threshold) indicates the SOH threshold (85%) required for the battery to enter the healthy stage. When the SOH is greater than this threshold, the battery is considered to be in a healthy state. The green area in the figure corresponds to the initial use stage of the battery, during which the system does not require much intervention and can be used normally.

[0144] The yellow dashed line (sub-health stage threshold) represents the SOH threshold (50%) at which the battery enters the sub-health stage. When the battery's SOH is below 85% but above 50%, the battery enters the sub-health stage. The system will then begin monitoring the battery's usage and prompt the user to perform necessary maintenance or checks to prevent further degradation of battery performance.

[0145] The red dashed line (degradation stage threshold) marks the State of Harshness (SOH) threshold (20%) at which the battery enters the degradation stage. When the SOH drops below this threshold, battery performance deteriorates severely, and the system will trigger a strong warning, reminding the user to replace the battery or take emergency measures. At this stage, the battery no longer has reliable performance and may malfunction or become unusable.

[0146] This diagram visually illustrates the changes in battery state of health (SOH) at different stages of use, clearly identifying the three stages of healthy, sub-healthy, and degraded battery through various thresholds. Using these thresholds, the battery management system can monitor battery status in real time and trigger corresponding management strategies or alerts based on changes in battery health, ensuring the safety and efficiency of the battery during use.

[0147] Optionally, this method also includes a multi-source data fusion preprocessing step. In real vehicle operating environments, data sources are diverse, such as battery BMS, vehicle VCU, in-vehicle navigation GPS, and online weather services, and data quality varies considerably (due to issues like signal interruption, sensor drift, and electromagnetic interference). Directly using this raw, unprocessed data for calculations may result in significant errors. The core objective of this step is to output a standardized data sequence that is absolutely time-aligned, absolutely valid in content, and absolutely uniform in format, providing the highest quality, clean data for subsequent behavior analysis, environmental coupling calculations, and model learning.

[0148] This step specifically includes: first, establishing a unified system time base and adding a synchronization timestamp to all collected data; then, setting data validity verification rules for each data source, including data range threshold verification and data change rate threshold verification; removing abnormal data points that fail the verification, and using interpolation algorithms to compensate for missing data caused by removal or collection loss; finally, outputting a standardized data sequence after time alignment and validity processing for use in subsequent steps.

[0149] For example, when a vehicle enters a long tunnel, an unprocessed system faces catastrophic data problems: the GPS signal is completely lost, leading to interruptions in altitude and location data; simultaneously, the slippery tunnel surface may cause the wheels to momentarily slip, causing the wheel speed sensor to upload a brief, erroneous spike signal far exceeding the actual vehicle speed. If this raw data is used directly, the range prediction model will collapse due to data loss or calculate an absurdly high energy consumption event due to the incorrect instantaneous speed. In contrast, the preprocessing steps in this embodiment first use interpolation or dead reckoning algorithms to reasonably complete the location and altitude data sequence within the tunnel; simultaneously, its rate of change threshold verification rules immediately identify and remove the speed spike that does not conform to physical laws. Ultimately, the upper-layer application obtains a smooth, complete, and reliable data stream, enabling accurate calculation of the vehicle's true energy consumption within the tunnel, ensuring the stability and accuracy of system decisions.

[0150] Optionally, this method also includes an online parameter self-learning and dynamic calibration mechanism. Since the various model parameters built into the system at the factory, such as the initial values ​​of user behavior weights, the coefficients of the SVR prediction model, and the strength factor of environmental coupling, are set based on laboratory data and statistical averages from a large number of users, they are effective for a standard user, but inevitably have biases for each unique individual. This mechanism endows the system with a learning capability. It no longer relies on offline, general firmware upgrades to optimize itself, but can silently and continuously collect data specific to the vehicle and the user during the vehicle's daily operation, and use this personalized big data to periodically fine-tune and calibrate its core parameters, making the system increasingly adaptable to the vehicle and the user.

[0151] This mechanism specifically includes the following: First, a regular parameter calibration cycle is set, and a change detection module is deployed to monitor significant changes in battery performance, user behavior patterns, or environmental conditions. When the calibration cycle is reached or a significant change is triggered, recent system operation data is automatically collected. Then, the collected data is used to recalculate the parameters of the Support Vector Regression (SVR) model, the weight coefficients in the calculation of the behavior influence coefficient, and the coupling coefficients in the calculation of the multiphysics coupling factor. Finally, the optimized new parameters are updated to each calculation module to achieve online self-learning and dynamic calibration of the model parameters.

[0152] For example, suppose a new car owner initially drives casually and charges irregularly. The system can only provide conservative warnings based on a general model. However, after six months, this owner develops a stable habit of using off-peak electricity for slow charging at home every Friday evening and fully charging before weekend long trips. At this point, the system's change detection module recognizes that the user's behavior pattern has shifted from random to regular. A calibration mechanism is then triggered, automatically collecting recent charging and driving data for relearning. As a result, the system dynamically increases the weight of charging time and location in the calculation of the behavior influence coefficient, significantly improving the final calculated coefficient value. Reflecting on the user experience, the system will no longer frequently issue low battery reminders during the week. Instead, it will more intelligently push a personalized, predictive message on Friday afternoons, based on the user's schedule, such as "Tonight is your usual charging time; please remember to recharge for your weekend trip," transforming from a general assistant to a more considerate one.

[0153] like Figure 7 The visualization verification results of the online self-learning and dynamic calibration mechanism for parameters are shown. The x-axis represents the monitoring time (0-360 days, covering the entire 1-year cycle), and the y-axis represents the charging safety warning deviation (unit: %SOC, i.e., the absolute deviation between the actual battery safety threshold and the model predicted threshold). From the curve characteristics, the red solid line represents the warning deviation without self-learning enabled. The deviation fluctuates wildly within the 3%-8%SOC range, with a standard deviation of 2.8%, consistent with the description that fixed parameters cannot adapt to dynamic battery changes, leading to a decrease in warning accuracy in later stages. The blue solid line represents the warning deviation with self-learning enabled. With each trigger point (regular calibration + significant change), the deviation gradually optimizes, decreasing from an initial 5%SOC to below 1%SOC in later stages, with a standard deviation of only 0.8%SOC. From the perspective of the triggering mechanism, the black dot is the trigger point for periodic calibration (the periodic parameter calibration cycle is every 90 days). After triggering, the SVR parameters are optimized, the sensitivity to abnormal data is improved, and the behavior weight is adjusted from 0.3 to 0.45, which is in line with user habits. The deviation is reduced from 4.2% SOC to 3.5% SOC on the 90th day, which is a good result.

[0154] Example 5: Figure 8This paper presents a schematic diagram of a dynamic prediction system for charging safety warning thresholds throughout the entire vehicle lifecycle. The system includes: a data acquisition module for real-time acquisition of basic battery cycle data, electrochemical characteristic data, user charging behavior data, and multi-physics parameters of the vehicle's environment. The electrochemical characteristic data includes at least battery internal resistance and static voltage rebound. An aging stage determination module is used to construct a two-dimensional quantitative determination system based on the basic cycle data and electrochemical characteristic data. This system combines macroscopic battery cycle characteristics (including cumulative cycle count and cumulative capacity decay rate) with microscopic electrochemical states (including battery internal resistance and static voltage rebound), dividing the battery's entire lifecycle into the initial aging stage. The system comprises three stages: intermediate aging and non-aging; a model parameter adaptation and basic threshold generation module, used to dynamically adapt differentiated Support Vector Regression (SVR) model parameters (including different input feature weight combinations and model penalty parameters) and threshold correction coefficients based on the aging stage determination module output, generating a basic warning threshold based on battery aging status; a behavior influence coefficient calculation and threshold correction module, used to analyze user charging behavior data, extract multiple charging behavior features and quantify and classify them, multiply the quantized values ​​of each feature level with their weight coefficients and sum them to obtain the behavior influence coefficient, which is used to correct the basic warning threshold; and a coupling factor calculation and final threshold generation module, used to... The system employs a multi-physics parameter calculation method, which calculates multi-physics coupling factors (each coupling factor being a continuous function of the corresponding environmental parameter) reflecting changes in high-altitude heat dissipation efficiency, low-temperature electrolyte conductivity, and high-humidity insulation performance. These coupling factors are used to further refine the warning thresholds, which have already been corrected by the behavior influence coefficient, generating dynamically predicted charging safety warning thresholds. A data preprocessing submodule, integrated between the data acquisition module and subsequent modules, is used to establish a unified system time base for data synchronization timestamps. Abnormal data is removed through data range and rate-of-change threshold checks, and interpolation algorithms are used to compensate for missing data, outputting a standardized data sequence. An abnormal behavior filtering submodule, integrated with the behavior influence coefficient calculation and threshold... The value correction module is used to establish rules for identifying abnormal behavior and clean the original charging behavior data to eliminate the influence of non-habitual abnormal behavior. The stage switching trigger submodule, integrated into the aging stage determination module, is used to set trigger conditions such as cycle increment and capacity decay rate. When the conditions are met, the aging stage re-determination is automatically initiated. When the stage changes, the system parameters are updated and status information is pushed. The parameter self-learning calibration submodule is used to set the periodic calibration cycle and monitor significant changes in battery performance, user behavior patterns, or environmental conditions. When triggered, it collects recent operating data, re-optimizes SVR model parameters, behavior influence coefficient weights, and multiphysics coupling factors, and updates them to each module to achieve dynamic calibration.

[0155] This system ensures data reliability through multi-source data acquisition and fusion preprocessing. Combined with quantitative judgment of battery aging stages and adaptation to differentiated SVR models, it enables the warning threshold to dynamically match the battery's full life cycle state from initial decay to final decay. It introduces a user charging behavior influence coefficient to adapt the threshold to different usage habits. It uses multi-physics coupling factors to quantify the impact of extreme environments on battery performance, enhancing adaptability under complex operating conditions. At the same time, through abnormal behavior filtering, stage adaptive switching, and online self-learning calibration of parameters, it continuously optimizes the warning accuracy, effectively reducing false alarms and missed alarms throughout the entire life cycle, improving the reliability and adaptability of charging safety warnings, and providing stable protection for vehicle charging safety throughout the entire life cycle.

[0156] Example 6: Corresponding to the above examples, the present invention also proposes an electronic device.

[0157] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 200 includes a processor 201 and a memory 203. The processor 201 and the memory 203 are connected, for example, via a bus 202. Optionally, the electronic device 200 may further include a transceiver 204. It should be noted that in practical applications, the transceiver 204 is not limited to one unit, and the structure of this electronic device 200 does not constitute a limitation on the embodiments of the present invention.

[0158] Processor 201 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 201 may also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0159] Bus 202 may include a path for transmitting information between the aforementioned components. Bus 202 may be a PCI bus or an EISA bus, etc. Bus 202 may be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0160] The memory 203 stores a computer program corresponding to the dynamic prediction method for charging safety warning thresholds for the entire vehicle lifecycle according to the above embodiments of the present invention. This computer program is controlled and executed by the processor 201. The processor 201 executes the computer program stored in the memory 203 to implement the content shown in the aforementioned method embodiments.

[0161] Among them, electronic devices 200 include, but are not limited to: mobile terminals such as laptops and tablets, as well as fixed terminals such as desktop computers. Figure 9 The electronic device 200 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0162] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dynamic prediction method for charging safety early warning thresholds throughout the entire vehicle lifecycle, characterized in that, Includes the following steps: S1. Real-time acquisition of basic battery cycle data, electrochemical characteristic data, user charging behavior data, and multi-physics parameters of the vehicle's environment; S2. Based on the aforementioned basic cycle data and electrochemical characteristic data, a quantitative judgment system is constructed to divide the entire life cycle of the battery into the initial decay stage, the intermediate decay stage, and the non-decay stage; the electrochemical characteristic data includes at least the battery internal resistance and the static voltage rebound. S3. Based on the determined aging stage, dynamically adapt the differentiated support vector regression (SVR) model parameters and threshold correction coefficients to generate a basic early warning threshold based on the aging state. S4. Analyze the user charging behavior data, calculate the behavior impact coefficient used to characterize the risk level of user charging habits, and correct the basic early warning threshold. S5. Based on the multi-physics parameters, calculate the multi-physics coupling factor for quantifying the impact of extreme environments on battery performance, and use the multi-physics coupling factor to further correct the warning threshold after the behavior influence coefficient correction, and finally generate a dynamically predicted charging safety warning threshold.

2. The method according to claim 1, characterized in that, The quantitative determination system described in step S2 is a two-dimensional determination model based on the macroscopic cycle characteristics and microscopic electrochemical state of the battery. The macro-cycle characteristics include the cumulative number of battery cycles and the cumulative capacity decay rate. The microscopic electrochemical state dimensions include the battery internal resistance and the voltage rebound after resting.

3. The method according to claim 2, characterized in that, The dual-dimensional judgment model uses the following judgment rules to divide the stages: Preliminary stage division is based on the comparison results of the battery's cumulative cycle count, cumulative capacity decay rate, and corresponding thresholds; Then, by combining the growth factor of the battery internal resistance relative to the initial internal resistance and the comparison results of the static voltage rebound with the preset threshold, the preliminary classification results are corrected and confirmed.

4. The method according to claim 1, characterized in that, The dynamically adapted and differentiated support vector regression (SVR) model parameters mentioned in step S3 include: Configure different combinations of input feature weights for different aging stages; Adjust the model's penalty parameters to change its sensitivity to outlier data; For the non-decay stage, the static voltage rebound is added as an input feature, and corresponding feature weights are assigned.

5. The method according to claim 1, characterized in that, The calculation of the behavioral influence coefficient in step S4 includes the following steps: Extract multiple charging behavior features from users' historical charging behavior data; Each charging behavior characteristic is quantified and classified. The final behavioral impact coefficient is obtained by multiplying the quantified value of each feature by its corresponding weight coefficient and then summing the results.

6. The method according to claim 1, characterized in that, The multiphysics coupling factor mentioned in step S5 includes: Altitude coupling factor, which reflects the change in heat dissipation efficiency at high altitudes; Low-temperature coupling factor, reflecting changes in electrolyte conductivity at low temperatures; Humidity coupling factor, which reflects the change in insulation performance under high humidity conditions.

7. The method according to claim 1, characterized in that, It also includes an abnormal behavior filtering mechanism: Establish rules for identifying abnormal behavior to distinguish between habitual behavior and short-term abnormal behavior; Before calculating the behavior impact coefficient, the original charging behavior data is cleaned based on the identification rules to exclude the influence of non-habitual abnormal behavior events.

8. The method according to claim 1, characterized in that, It also includes an adaptive triggering mechanism for phase switching: Set multiple re-determination trigger conditions, including loop count increment condition and capacity decay rate condition; When any triggering condition is met, the aging stage re-judgment process is automatically started; When changes occur during the determination phase, the system parameters are automatically updated and status change information is pushed out.

9. The method according to claim 1, characterized in that, It also includes a multi-source data fusion preprocessing step: Establish a unified system time benchmark and add a synchronization timestamp to all collected data; Set data validity verification rules for each data source, including data range threshold verification and data change rate threshold verification; Abnormal data points that fail the verification are removed, and data loss caused by removal or collection loss is compensated by interpolation algorithm; Output a normalized data sequence after time alignment and validity processing for use in subsequent steps.

10. The method according to claim 1, characterized in that, It also includes an online parameter self-learning and dynamic calibration mechanism: Set up regular parameter calibration cycles and deploy change detection modules to monitor significant changes in battery performance, user behavior patterns, or environmental conditions; When the calibration cycle is reached or a significant change is triggered, recent system operation data is automatically collected; Using the collected data, the parameters of the support vector regression (SVR) model, the weight coefficients in the calculation of the behavioral influence coefficient, and the coupling coefficients in the calculation of the multiphysics coupling factor were recalculated and optimized. The optimized new parameters are updated to each calculation module to achieve online self-learning and dynamic calibration of model parameters.

Images