Electric two-wheeler multi-modal extreme charging state safety regulation method based on edge computing

Abstract

The application provides an electric two-wheeled vehicle multi-modal extreme charging state safety regulation method based on edge computing, and relates to the technical field of safety regulation, which comprises the following steps: identifying the scene type of the electric two-wheeled vehicle extreme charging, collecting electrochemical parameters and thermal characteristic data, and collecting user behavior data. The electrochemical parameters and thermal characteristic data are preprocessed by an edge node to obtain standardized multi-modal extreme charging data, and the corresponding monitoring direction is set based on the scene type. A multi-modal fusion model is constructed, the battery state is evaluated by combining the monitoring direction and the standardized multi-modal extreme charging data to obtain the battery difference state, the influence of the user behavior data on the battery difference state is analyzed to obtain the charging abnormal risk control data, and the dynamic charging regulation strategy is generated in combination with the battery difference state. The application can dynamically adjust the charging regulation strategy according to the scene, the battery state and the user behavior, and can balance the charging speed and the battery life, thereby significantly improving the extreme charging safety.

Description

Technical Field

[0001] This invention relates to the field of safety control technology, and in particular to a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. Background Technology

[0002] With the rapid growth in demand for green urban travel, electric two-wheelers have become an important mode of transportation for short-distance commuting in my country due to their convenience, economy, and low carbon footprint. To improve user experience, the industry has widely adopted ultra-fast charging technology, which can replenish a large amount of energy to the battery in a short time. However, the ultra-fast charging process places great stress on the battery system, which can easily lead to electrochemical anomalies such as overheating, lithium plating, and sudden changes in internal resistance. In severe cases, it may even lead to thermal runaway or fire and explosion, significantly increasing safety risks.

[0003] Existing electric two-wheeler charging management systems mostly rely on fixed thresholds or simple rules for overcurrent and overvoltage protection, lacking the ability to finely perceive and dynamically control the real-time battery status. Especially in practical applications, charging scenarios are highly complex and diverse—for example, concentrated charging at night in high-density residential areas, high-frequency fast charging in commercial areas during the day, and continuous full-load operation of shared electric bicycle battery swapping stations. The grid load, ambient temperature, equipment aging, and user operating habits vary greatly under different scenarios. Traditional solutions often ignore these external factors and cannot adjust monitoring priorities and safety strategies for specific scenarios, leading to frequent false alarms or missed alarms.

[0004] Meanwhile, user behavior has a significant impact on the safety of the Extreme Charger. For example, frequent plugging and unplugging of the charger, using non-original or inferior charging equipment, and leaving the battery fully charged for extended periods can all accelerate battery degradation or induce momentary abnormalities.

[0005] Furthermore, existing systems largely rely on centralized cloud data processing, which suffers from high transmission latency, high bandwidth pressure, and privacy risks, making it difficult to meet the millisecond-level response and safety control requirements during ultra-fast charging. While edge computing has been applied in some industrial scenarios, a multimodal collaborative analysis architecture integrating electrochemistry, thermal management, and user behavior has not yet been formed in the field of ultra-fast charging for electric two-wheelers.

[0006] In summary, there is an urgent need for a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing to address the aforementioned shortcomings. Summary of the Invention

[0007] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing, which solves the shortcomings of existing technologies that cannot perform extreme charging safety control for different scenarios and different user behaviors.

[0008] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing, including:

[0009] Identify the scenario types for extreme charging of electric two-wheelers, collect electrochemical parameters and thermal characteristic data during the extreme charging process under the corresponding scenario types, and collect user behavior data.

[0010] Standardized multimodal extreme charge data are obtained by preprocessing electrochemical parameters and thermal characteristic data through edge nodes, and the corresponding monitoring direction is set based on the scene type.

[0011] A multimodal fusion model is constructed, and the battery status is evaluated by combining monitoring direction and standardized multimodal extreme charging data to obtain battery difference status. The impact of user behavior data on battery difference status is analyzed to obtain charging anomaly risk control data.

[0012] Dynamic charging control strategies are generated based on battery status differences and charging anomaly risk control data.

[0013] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The step of identifying the scene type includes:

[0014] Obtain the geographical location of the target charging area, and collect charging device density and device type as basic scenario data, as well as charging behavior data and load characteristic data.

[0015] Key features of the scenario are extracted from basic scenario data, charging behavior data, and load characteristic data, including scenario geographic attributes, device density level, charging frequency characteristics, charging duration characteristics, load peak characteristics, time period distribution characteristics, battery cycle characteristics, and vehicle type characteristics.

[0016] Collect historical charging data containing key features of the scene, label the features, input them into a lightweight CNN model, and train a scene classifier to obtain a scene classification model.

[0017] Input the key features of the scene to be identified into the scene classification model, and output the scene type corresponding to the charging area.

[0018] This invention provides a multimodal extreme charge state safety control method for electric two-wheeled vehicles based on edge computing, and the steps for obtaining electrochemical parameters and thermal characteristic data include:

[0019] Develop a scenario-specific data collection list based on the scenario type, and determine the equipment selection and deployment density.

[0020] Based on the scenario-specific data collection list, the data collection frequency and triggering mode are adapted, and the collected data is transmitted to the edge node through a preset transmission protocol.

[0021] The collected data is subjected to noise filtering and outlier removal, and different formats are converted into standardized JSON format. Scene identifiers, timestamps, and device IDs are added to obtain electrochemical parameters and thermal characteristic data.

[0022] This invention provides a method for safety control of multimodal extreme charging status of electric two-wheeled vehicles based on edge computing. The steps for obtaining standardized multimodal extreme charging data include:

[0023] Electrochemical parameters and thermal characteristic data are classified according to data type, and stage data before charging (standby) and after charging (power off) are removed to obtain preliminary processed data.

[0024] Based on the scene type, a denoising method is selected to remove noise from the preliminary processed data to obtain multimodal data, and scene-specific anomaly thresholds are set.

[0025] Temporal and dynamic features are extracted from multimodal data as feature parameters, mapped to the [0,1] interval to eliminate dimensional differences, and standardized multimodal extreme charging data is obtained by establishing the correspondence between feature parameters in the same charging process through timestamps and device IDs.

[0026] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The step of setting the monitoring direction includes:

[0027] Based on the scenario type, identify the corresponding scenario objectives, break down the scenario-specific risks, and correlate them with user behavior risks to obtain risk breakdown data.

[0028] Based on the risk decomposition data, a unified basic monitoring dimension is established, and based on the differences in scenarios, the basic monitoring dimension is expanded to obtain scenario characteristic monitoring dimensions.

[0029] Based on the monitoring dimensions of scenario characteristics, electrochemical parameter monitoring indicators, thermal characteristic monitoring indicators, and user behavior monitoring indicators are set and integrated to obtain the total monitoring indicators.

[0030] Based on the scenario type and overall monitoring indicators, the monitoring direction is determined by setting the corresponding monitoring priority and early warning triggering mechanism for each scenario.

[0031] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The steps for constructing a multimodal fusion model include:

[0032] Electrochemical parameters are used as time-series modes, thermal feature data as spatial modes, and user behavior data as behavioral modes. The corresponding basic feature vectors and mode priorities are then determined.

[0033] The backbone network adopts a cross-modal attention mechanism, determines branch modules based on different modal characteristics, and determines the architecture configuration and sets a phased fusion mechanism according to the scenario type to obtain the basic architecture.

[0034] Based on the scenario type, historical scenario data is collected, and the infrastructure is trained in different scenarios to obtain a multimodal fusion model.

[0035] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The steps for evaluating the battery differential states include:

[0036] From standardized multimodal extreme charge data, core monitoring features relevant to the current scenario monitoring direction are selected.

[0037] Weights are assigned to core monitoring features based on monitoring priorities, and the multimodal fusion model outputs key state parameters in parallel according to scenario type.

[0038] Based on historical data of healthy batteries of the same type in the same scenario, a scenario-specific state benchmark is constructed. The scenario-specific state benchmark is updated as the battery usage cycle and environmental changes to obtain a scenario-based benchmark.

[0039] By comparing key state parameters with scenario-based benchmarks, multidimensional difference indicators are calculated, and adjustments are made based on scenario characteristics to obtain a comprehensive difference score.

[0040] Based on the comprehensive difference score, the battery difference status is divided into multiple levels and a correspondence with safety risks is established to obtain the battery difference status.

[0041] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The steps for obtaining charging anomaly risk control data include:

[0042] The system filters user behavior data into operational behaviors, device-related behaviors, and habitual behaviors, and converts non-numerical behaviors into quantitative features to obtain behavioral impact data.

[0043] By combining historical fault data, the behavioral impact data is divided into risky behaviors, and behavioral risk thresholds are set for different scenarios.

[0044] We extract battery difference indicators related to user behavior from standardized multimodal extreme charging data and establish a correlation model by combining statistical analysis and machine learning.

[0045] The impact of behavioral influence data on battery difference indicators is quantified based on the correlation model, generating charging anomaly risk control data that includes behavioral risk correlation, abnormal risk probability, and risk diffusion coefficient.

[0046] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing, the quantification steps of which include:

[0047] The formula for quantifying the impact of a single user behavior on battery performance is as follows:

[0048]

[0049] In the formula, It's about the degree of influence. It is the battery difference value at the time this behavior occurs. It is the baseline difference value. It is the behavioral risk coefficient.

[0050] The weighted summation method is used to calculate the overall impact when multiple user behaviors exist. The formula is as follows:

[0051]

[0052] In the formula, It is the overall impact. It is the frequency weight of the behavior.

[0053] This invention provides a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing. The steps for generating a dynamic charging control strategy include:

[0054] By integrating battery status differences and charging anomaly risk control data, decision indicators are extracted for both battery and risk control dimensions, and different priorities are set for different decision indicators for different scenarios.

[0055] Based on scenario type and priority, a framework for adapting strategies to different scenarios is built. Basic control parameters are calculated based on differences in battery status, and the control parameters are obtained by correcting them based on charging anomaly risk control data.

[0056] The control parameters are verified by setting constraints based on grid load, equipment compatibility, and extreme environments to determine whether the preset expectations have been met. If so, a dynamic charging control strategy is output; otherwise, the basic control parameters are recalculated.

[0057] This invention provides a multimodal extreme charging state safety control method for electric two-wheelers based on edge computing. It dynamically adjusts the charging strategy according to the characteristics of different scenarios and battery status, making the charging process more rational, avoiding unnecessary charging time and energy waste, improving charging efficiency, and shortening charging time. By accurately identifying scenario types, precisely assessing differences in battery status, and predicting abnormal charging risks, it can promptly detect potential safety hazards and generate dynamic charging control strategies to avoid safety accidents caused by overcharging, overheating, and other problems, ensuring the safety of electric two-wheelers during extreme charging. Utilizing edge computing and multimodal fusion technology, it achieves real-time monitoring and intelligent control of the extreme charging process of electric two-wheelers, improving the intelligence level of charging management and providing strong support for the operation and management of charging facilities. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0059] Figure 1 This is one of the flowcharts of the multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing provided in the embodiments of the present invention;

[0060] Figure 2 This is the second flowchart of the multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing provided in this embodiment of the invention.

[0061] Figure 3 This is the third flowchart of the multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing provided in this embodiment of the invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0063] The following is combined with Figures 1-3 This invention describes a multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing.

[0064] like Figure 1 As shown in the embodiment of the present invention, the multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing includes:

[0065] Identify the scenario types for extreme charging of electric two-wheelers, collect electrochemical parameters and thermal characteristic data during the extreme charging process under the corresponding scenario types, and collect user behavior data.

[0066] like Figure 2 As shown, the steps for identifying scene types include:

[0067] Obtain the geographical location of the target charging area, and collect charging device density and device type as basic scenario data, as well as charging behavior data and load characteristic data.

[0068] The data collected on charging behavior can include: extracting the average number of daily charging sessions per charging station, the duration of each charging session, and the distribution of charging times, such as the proportion of morning / evening peak hours, and the battery type, from the charging station terminal and the user's APP.

[0069] The collection of load characteristic data may include: collecting the peak value of centralized charging load, load duration, and load fluctuation amplitude through the power grid load monitoring module, with a focus on recording the time period when the peak charging demand occurs on a single day.

[0070] Key features of the scenario are extracted from basic scenario data, charging behavior data, and load characteristic data, including scenario geographic attributes, device density level, charging frequency characteristics, charging duration characteristics, load peak characteristics, time period distribution characteristics, battery cycle characteristics, and vehicle type characteristics.

[0071] Collect historical charging data containing key features of the scene, label the features, input them into a lightweight CNN model, and train a scene classifier to obtain a scene classification model.

[0072] Input the key features of the scene to be identified into the scene classification model, and output the scene type corresponding to the charging area.

[0073] For example, the quantitative features of the scene to be identified are input into the scene classification model, and the similarity with the feature databases of community scenes and logistics station scenes is calculated. If the similarity is ≥90%, it is determined to be the corresponding scene. If the similarity is between 80% and 90%, a second feature supplementation collection is initiated and rematching is performed.

[0074] The steps to obtain electrochemical parameters and thermal characteristic data include:

[0075] Develop a scenario-specific data collection list based on the scenario type, and determine the equipment selection and deployment density.

[0076] The scenario-specific data collection list can include, when the scenario type is a community scenario: electrochemical parameters including battery voltage, current, individual cell voltage difference, SOC, and SOH; and thermal characteristic data including battery pack surface temperature, cell temperature difference, and ambient temperature and humidity.

[0077] For logistics terminal scenarios: additional electrochemical parameters include charging rate, internal resistance change rate, and polarization voltage. Additional thermal characteristic data include temperature rise rate and temperature field distribution gradient.

[0078] In community settings, every two charging stations share one infrared thermal imager. In logistics station settings, each charging station is independently equipped with a power sensor and a high-precision temperature sensor.

[0079] Electrochemical parameter sensor deployment: Current / voltage sensors are connected in series in the charging pile output circuit, and the BMS communication interface is connected to the edge node via a CAN bus. Additional internal resistance sensors are deployed at the positive and negative terminals of the battery pack at logistics stations.

[0080] Thermal signature sensor deployment: Temperature sensors are attached to the surface of the battery pack, and an infrared thermal imager is installed 1.5m above the charging station, with the lens aimed at the battery charging interface area. In community settings, the thermal imager covers 3-5 charging stations, while in logistics stations, each charging station provides independent coverage.

[0081] Sensor calibration: Before data acquisition, calibrate the current / voltage sensor using a standard power supply and calibrate the thermal imager using a blackbody furnace. Repeat the calibration once every 3 months.

[0082] Based on the scenario-specific data collection list, the data collection frequency and triggering mode are adapted, and the collected data is transmitted to the edge node through a preset transmission protocol.

[0083] The data acquisition frequency can be adapted as follows: 10Hz for electrochemical parameters and 5Hz for thermal characteristic data in community scenarios. For logistics station scenarios, due to the rapid dynamic changes caused by fast charging, the acquisition frequency for both types of data is increased to 20Hz.

[0084] Data collection can be triggered in the following ways: In community scenarios, continuous collection combined with timed storage is used. In logistics station scenarios, event-triggered collection combined with high-frequency storage is used, with 50Hz high-frequency collection initiated when the charging rate is greater than 2C.

[0085] Data is transmitted to edge nodes in real time via the MQTT protocol. In community scenarios, data from the last hour is cached, while in logistics station scenarios, high-frequency data from the last 30 minutes is cached to avoid data loss.

[0086] The collected data is subjected to noise filtering and outlier removal, and different formats are converted into standardized JSON format. Scene identifiers, timestamps, and device IDs are added to obtain electrochemical parameters and thermal characteristic data.

[0087] Noise filtering: Moving average filtering is used, for example, 10 data points in the community scene window and 5 data points in the logistics station window, to remove current / voltage fluctuation noise. Thermal imager data is filtered by median filtering to eliminate environmental interference.

[0088] Outlier removal: Set parameter thresholds, such as abnormal voltage range of <36V or >72V in community scenarios, and abnormal temperature rise rate of >5℃ / min in logistics stations. Data exceeding the threshold is marked as abnormal and removed.

[0089] Standardized multimodal extreme charge data are obtained by preprocessing electrochemical parameters and thermal characteristic data through edge nodes, and the corresponding monitoring direction is set based on the scene type.

[0090] The steps to obtain standardized multimodal polar-charge data include:

[0091] Electrochemical parameters and thermal characteristic data are categorized according to data type, and data from the standby phase before charging and the phase after power failure after charging are removed to obtain preliminary processed data. The categorization steps include: electrochemical parameters are temporarily stored in the edge node time-series database, and thermal characteristic image data are temporarily stored in the local cache directory, both associated with device ID, timestamp, and scene identifier.

[0092] Based on the scene type, a denoising method is selected to remove noise from the preliminary processed data to obtain multimodal data, and scene-specific anomaly thresholds are set.

[0093] Electrochemical parameter noise filtering can include: for centralized community charging scenarios, a moving average filter with a window size of 10 is used to smooth current / voltage fluctuation noise. For fast charging scenarios at logistics stations, due to the higher data acquisition frequency, a moving average filter with a window size of 5 is used to balance noise reduction and data real-time performance.

[0094] Noise filtering for thermal feature data may include: using 3×3 median filtering on thermal imaging image data to eliminate ambient light and shadow interference, using wavelet transform to extract low-frequency effective signals from temperature sensor data, and filtering high-frequency noise caused by electromagnetic interference.

[0095] In the logistics terminal scenario, Kalman filtering is additionally applied to the polarization voltage data to reduce the impact of dynamic noise under high-rate charging. The data error after filtering is ≤±0.3%.

[0096] The steps for setting abnormal thresholds for different scenarios may include: Community scenario: voltage abnormality range <36V or >72V, current abnormality range <0A or >30A, temperature abnormality >60℃.

[0097] Logistics station scenario: abnormal voltage range <36V or >84V, abnormal current range <0A or >50A, abnormal temperature rise rate >5℃ / min.

[0098] Outlier detection and handling: Outliers are detected using the 3σ criterion. A single instance exceeding the threshold is marked as a suspected anomaly, while exceeding the threshold for three consecutive collection cycles is considered a valid anomaly, which is then directly removed and recorded in the anomaly log. Suspected anomaly data is retained for secondary verification using correlated data.

[0099] Temporal and dynamic features are extracted from multimodal data as feature parameters, mapped to the [0,1] interval to eliminate dimensional differences, and standardized multimodal extreme charging data is obtained by establishing the correspondence between feature parameters in the same charging process through timestamps and device IDs.

[0100] Timing features can include, for example, peak voltage, average current, SOC change rate, and internal resistance fluctuation amplitude. For logistics station scenarios, additional dynamic features such as charging rate curve and polarization voltage peak are extracted.

[0101] Thermal feature extraction: Extracting the surface temperature field distribution, maximum temperature difference, and temperature gradient direction of the battery pack from thermal imaging images. Extracting features such as temperature rise rate and steady-state temperature value from temperature sensor data.

[0102] like Figure 3 As shown, the steps for setting the monitoring direction include:

[0103] Based on the scenario type, identify the corresponding scenario objectives, break down the scenario-specific risks, and correlate them with user behavior risks to obtain risk breakdown data.

[0104] Disassembly-specific risks may include: Community scenario: Risks include battery overcharging, poor contact caused by improper plugging and unplugging, heat buildup caused by ambient temperature and humidity, and users charging beyond the allotted time.

[0105] Logistics terminal scenario: Risks include uncontrolled temperature rise caused by high-rate charging, sudden changes in internal resistance due to battery cycle aging, grid load impact caused by concentrated fast charging, and interface wear caused by frequent plugging and unplugging.

[0106] Related user behavior risks: In both scenarios, it is necessary to focus on analyzing common abnormal operation risks such as the use of non-standard chargers, unauthorized wiring, and forced modification of interfaces. Communities should pay extra attention to residents' misoperation, and logistics stations should pay extra attention to operators' illegal fast charging.

[0107] Based on the risk decomposition data, a unified basic monitoring dimension is established, and based on the differences in scenarios, the basic monitoring dimension is expanded to obtain scenario characteristic monitoring dimensions.

[0108] Basic monitoring dimensions can include battery electrochemical safety, thermal characteristics, and user operating procedures; scenario-specific dimensions can be expanded as follows:

[0109] Community-based centralized charging scenarios: Added features such as ventilation conditions in the charging area, distance from nearby combustibles, and compatibility status of different brand BMS protocols.

[0110] Fast charging scenarios at logistics stations: new factors include grid voltage fluctuations, peak total power in a single area, number of cycle charging cycles, and polarization voltage changes.

[0111] Based on the monitoring dimensions of scenario characteristics, electrochemical parameter monitoring indicators, thermal characteristic monitoring indicators, and user behavior monitoring indicators are set and integrated to obtain the total monitoring indicators.

[0112] Electrochemical parameter monitoring indicators: Community scenario: Set voltage threshold 36V-72V, current threshold 0A-30A, single cell voltage difference ≤200mV, and trigger overcharge warning when SOC≥95%.

[0113] Logistics station scenario: Set voltage threshold 36V-84V, current threshold 0A-50A, internal resistance change rate ≤5% / charge, polarization voltage peak ≤1.5V, and start dynamic monitoring when the charging rate is >3C.

[0114] Thermal characteristic monitoring indicators: Community scenario: Battery surface temperature ≤60℃, cell temperature difference ≤3℃, ambient humidity ≤85%RH, temperature rise rate ≤2℃ / min.

[0115] Logistics station scenario: Battery surface temperature ≤55℃, cell temperature difference ≤2℃, temperature rise rate ≤3℃ / min, thermal imaging monitoring temperature field gradient ≤1℃ / cm.

[0116] User behavior monitoring indicators: Community scenario: Single charging time ≤ 12 hours, plugging and unplugging frequency ≤ 3 times / charging cycle, non-standard charger identification accuracy ≥ 92%.

[0117] For logistics terminal scenarios: increase monitoring frequency when the continuous fast charging interval is ≥30 minutes, the plugging and unplugging operation standardization is ≥90%, and the battery cycle charging count is ≥500 times.

[0118] Based on the scenario type and overall monitoring indicators, the monitoring direction is determined by setting the corresponding monitoring priority and early warning triggering mechanism for each scenario.

[0119] The monitoring priority ranking is as follows: Community scenario: priority is thermal runaway risk > abnormal user operation > device compatibility > environment adaptation.

[0120] In logistics terminal scenarios, the priority order is: rate of temperature rise > grid load > battery durability > operating procedures.

[0121] Warning triggering mechanism settings: In community scenarios, the main approach is multi-level warnings plus audio-visual prompts; in logistics station scenarios, the main approach is real-time intervention plus operation and maintenance linkage. Warnings are activated within 1 second after an abnormal operation is triggered.

[0122] A multimodal fusion model is constructed, and the battery status is evaluated by combining monitoring direction and standardized multimodal extreme charging data to obtain battery difference status. The impact of user behavior data on battery difference status is analyzed to obtain charging anomaly risk control data.

[0123] The steps to construct a multimodal fusion model include:

[0124] Electrochemical parameters are used as time-series modes, thermal feature data as spatial modes, and user behavior data as behavioral modes. The corresponding basic feature vectors and mode priorities are then determined.

[0125] Time series modality: Extract the mean, peak value, volatility, and trend slope within the sliding window.

[0126] Spatial modes: extract the maximum temperature difference, the area ratio of high-temperature regions, and the direction of temperature gradient in the temperature field.

[0127] Behavioral modality: Convert operation logs into numerical features, such as the number of unauthorized plugging and unplugging incidents and the probability of using non-standard chargers.

[0128] Community-based centralized charging scenarios: Spatial modality > Temporal modality > Behavioral modality.

[0129] Fast charging scenario at logistics stations: Time-series modality > Spatial modality > Behavioral modality.

[0130] The backbone network adopts a cross-modal attention mechanism, determines branch modules based on different modal characteristics, and determines the architecture configuration and sets a phased fusion mechanism according to the scenario type to obtain the basic architecture.

[0131] Timing branch: A bidirectional LSTM network is used, for example, layer 2 in a community scenario and layer 3 in a logistics station, to capture the dynamic changes of electrochemical parameters.

[0132] Spatial branch: Using MobileNetV3, the computing power is compressed by 50% in community scenarios, while logistics stations retain complete feature extraction capabilities to analyze the spatial temperature distribution of thermal imaging.

[0133] Behavioral branch: A text CNN with a fully connected layer is used to convert user operation features into behavioral risk score vectors.

[0134] Community scenario: Simplify the number of attention heads to support cross-modal attention mechanism (Transformer), reduce the complexity of dynamic prediction layers, keep the model parameter size within 5MB, and adapt to low computing power of edge nodes.

[0135] Logistics terminal scenario: Enhance the long short-term memory capability of temporal branches, add a high-rate feature adaptation layer in Transformer to capture parameter mutations when the charging rate is >2C, and allow the number of model parameters to be increased to 10MB to adapt to high computing power in edge nodes.

[0136] The phased fusion mechanism includes: Early fusion: Element-level concatenation of the electrochemical feature vector output from the time-series branch and the temperature field feature vector output from the spatial branch to form a 512-dimensional electro-thermal joint feature, preserving the correlation of the original data. The community scenario simplifies the concatenation logic and reduces computational load.

[0137] Mid-stage fusion: The electro-thermal features from the early fusion and the behavioral risk features output from the behavioral branch are input into the Transformer's cross-modal attention module to calculate the correlation weights between features.

[0138] Community scenario: Assign 40% attention weight to temperature field characteristics, 35% to electrochemical characteristics, and 25% to behavioral characteristics.

[0139] Logistics terminal scenario: Impart electrochemical dynamic characteristics, such as a 45% weight for the charging rate curve, a 35% weight for the temperature rise rate characteristic, and a 20% weight for the behavioral characteristics.

[0140] Late-stage fusion: Multi-task learning is performed on the global feature vector output by the Transformer, with parallel outputs including: battery state parameters, safety risk level, and probability of abnormal behavior.

[0141] Based on the scenario type, historical scenario data is collected, and the infrastructure is trained in different scenarios to obtain a multimodal fusion model.

[0142] Sample collection: ≥50,000 samples from community scenarios, covering 20+ battery brands and 30+ community environments; ≥30,000 samples from logistics stations, covering 10+ operating vehicle models and 5+ fast charging load scenarios.

[0143] Label definition: Label the sample with the true value of SOC, the degree of SOH decay, and the safety risk level, including manually labeled accidents, early warning cases, and abnormal behavior labels.

[0144] Community scenario: Employ a small batch + early stop strategy, with a batch size of 32 and 50 training rounds, using thermal runaway early warning accuracy as the core indicator, with a target of ≥95%.

[0145] Logistics terminal scenario: An incremental training strategy is adopted, and the model is updated once for every 1,000 new high-rate charging samples. The focus is on optimizing the SOH estimation accuracy under high rates, with a target error of ≤2%.

[0146] The steps for assessing the state of difference in batteries include:

[0147] From standardized multimodal extreme charge data, core monitoring features relevant to the current scenario monitoring direction are selected.

[0148] For example, in a community-based centralized charging scenario, the monitoring focus is on: thermal runaway risk > user operation standards > equipment compatibility. Features such as cell temperature difference, surface temperature, charging time, insertion and removal standards, and voltage stability are retained, while redundant data unrelated to high-rate dynamics are removed.

[0149] In fast charging scenarios at logistics stations, the monitoring focus is on: temperature rise rate > grid load adaptation > battery durability. Features such as charging rate curve, temperature rise rate, internal resistance change rate, polarization voltage peak, and regional total power are retained, and the temporal correlation of dynamic parameters is strengthened.

[0150] Weights are assigned to core monitoring features based on monitoring priorities, and the multimodal fusion model outputs key state parameters in parallel according to scenario type.

[0151] Community scenario: Thermal characteristics, such as temperature and temperature difference, account for 40%; electrochemical parameters, such as voltage and current, account for 30%; user behavior characteristics, such as operational standardization, account for 30%.

[0152] In logistics terminal scenarios: electrochemical dynamic characteristics, such as charging rate and internal resistance change, account for 45%; thermal characteristics, such as temperature rise rate, account for 35%; and load characteristics, such as regional power, account for 20%.

[0153] Key status parameters may include: basic parameters: state of charge (SOC), state of health (SOH), current charging rate, and real-time temperature.

[0154] Derivative parameters: Community scenario: cell temperature difference fluctuation rate, overcharge risk probability, and operational violation-related risks.

[0155] Logistics terminal scenario: temperature rise rate gradient, internal resistance abrupt change coefficient, polarization voltage recovery time.

[0156] Based on historical data of healthy batteries of the same type in the same scenario, a scenario-specific state benchmark is constructed. The scenario-specific state benchmark is updated as the battery usage cycle and environmental changes to obtain a scenario-based benchmark.

[0157] Scenario-specific benchmarks: Community scenario benchmarks include the SOC-voltage curve of a new battery at 25°C, the maximum temperature difference during normal charging, and the charge / discharge efficiency under standard operation.

[0158] Logistics station scenario benchmark: including the temperature rise rate of new batteries at 3C rate, internal resistance change within 100 cycles, and peak polarization voltage.

[0159] For every additional 100 cycles, the SOH baseline value is reduced by 3% (community) / 5%. For logistics stations, losses are faster due to high rates.

[0160] For every 25℃±5℃ deviation of the ambient temperature, the temperature-related benchmark is dynamically adjusted. For example, when the temperature is -10℃, the benchmark for the rate of temperature rise is allowed to be increased by 0.5℃ / min.

[0161] By comparing key state parameters with scenario-based benchmarks, multidimensional difference indicators are calculated, and adjustments are made based on scenario characteristics to obtain a comprehensive difference score.

[0162] The steps for calculating the multidimensional difference index include: Electrochemical difference: Voltage deviation rate = (Measured voltage - Reference voltage) / Reference voltage × 100%.

[0163] Internal resistance difference rate = (current internal resistance - reference internal resistance) / reference internal resistance × 100% (key indicator for logistics stations).

[0164] Difference in thermal characteristics: Temperature deviation = Measured temperature - Reference temperature.

[0165] Temperature rise rate difference = measured temperature rise rate - reference temperature rise rate (community scenario ≤ 2℃ / min, logistics station ≤ 3℃ / min).

[0166] Differences in health status: SOH attenuation rate = baseline SOH - measured SOH.

[0167] Failure risk difference = measured risk probability - baseline safety probability (baseline safety probability ≥ 95%).

[0168] Adjust the difference index according to the characteristics of the scenario: Community scenario: Increase the tolerance for voltage deviation of old batteries by 2% to accommodate normal fluctuations caused by aging.

[0169] Logistics terminal scenario: Strictly limit the difference in temperature rise rate during high-rate (3C) charging (correction factor 1.2, i.e., actual difference = calculated value × 1.2).

[0170] Based on the comprehensive difference score, the battery difference status is divided into multiple levels and a correspondence with safety risks is established to obtain the battery difference status.

[0171] Slight differences (0-20 points): The difference of a single indicator is ≤5%, and there are no thermal characteristics / safety risk indicators exceeding the standard, such as a cell temperature difference of 2.5℃ in community scenarios and a change rate of internal resistance of 3% in logistics stations.

[0172] Moderate difference (21-50 points): any core indicator differs by 5%-15%, or multiple indicators differ by 5%-10%, such as SOH attenuation of 12% in community scenarios and temperature rise rate difference of 1.5℃ / min in logistics stations.

[0173] Severe discrepancies (51-100 points): Any core indicator difference >15%, or thermal safety / operational risk indicators exceed the standard, such as a temperature deviation of 15℃ in a community setting or a sudden change in internal resistance of 20% in a logistics station.

[0174] The corresponding relationship can be: slight difference → no risk, continuous monitoring.

[0175] Moderate difference → Level 1 / Level 2 warning, such as voltage fluctuation warning triggered in community scenarios, and internal resistance abnormality warning triggered in logistics stations.

[0176] Significant discrepancies → Level 3 warning, such as signs of impending thermal runaway or short-circuit risk due to operational violations.

[0177] The steps to obtain charging anomaly risk control data include:

[0178] The system filters user behavior data into operational behaviors, device-related behaviors, and habitual behaviors, and converts non-numerical behaviors into quantitative features to obtain behavioral impact data.

[0179] Operational behaviors: frequency of plugging and unplugging the charging gun, standardization of plugging and unplugging, and charging time settings.

[0180] Equipment-related behaviors: whether non-standard chargers are used, whether unauthorized wiring is installed, and signs of modification to the charging interface.

[0181] Habitual behaviors: deviation from charging time period, frequency of continuous charging.

[0182] For insertion and removal compliance and confidence level of modification traces, a quantitative value of 0-100 is directly retained.

[0183] The use of non-standard chargers and other Boolean behaviors are converted into risk frequency and risk percentage.

[0184] For continuous charging behavior, calculate the continuous charging time divided by the battery's rated capacity to obtain the behavior intensity.

[0185] By combining historical fault data, the behavioral impact data is divided into risky behaviors, and behavioral risk thresholds are set for different scenarios.

[0186] Risky behaviors can be categorized as follows: Low-risk behaviors include standardized plugging and unplugging, charging time within the recommended range (e.g., ≤8 hours in communities, ≤2 hours in logistics stations), and the use of original equipment. These behaviors are characterized by a risk frequency of 0 and a standardization score of ≥80.

[0187] Medium-risk behaviors: Occasional improper plugging and unplugging, such as 1-2 times / week, charging time exceeding the standard but ≤12 hours (community), slight modification of the interface. These behaviors are characterized by a risk ratio of 5%-20% and a compliance score of 60-80.

[0188] High-risk behaviors: Frequent unauthorized plugging and unplugging, such as ≥3 times / week, using non-standard chargers, illegally wiring, or severely modifying interfaces. Characterized by a risk frequency of ≥5 times / month and a compliance score of <60.

[0189] In community settings, there is a higher tolerance for charging exceeding the allotted time. The medium-risk threshold is relaxed to ≤15 hours because residents' charging habits are more dispersed.

[0190] In logistics hub scenarios, the risk level of continuous fast charging is strictly controlled. If a battery is fast charged ≥3 times within 24 hours, it is considered to be of medium risk, because frequent charging and discharging at high rates will cause greater damage to the battery.

[0191] We extract battery difference indicators related to user behavior from standardized multimodal extreme charging data and establish a correlation model by combining statistical analysis and machine learning.

[0192] Battery variation indicators may include those related to insertion / removal behavior: differences in interface contact resistance and voltage fluctuation deviations.

[0193] Related to the use of non-standard equipment: charging current deviation rate, cell temperature difference.

[0194] Related to overcharge: SOC overcharge deviation, SOH attenuation.

[0195] Statistical analysis: Calculate the co-occurrence probability of behaviors with different risk levels and battery differences. For example, the probability of severe battery differences occurring under high-risk behaviors: ≥30% for high-risk behaviors → severe differences in community scenarios, and ≥50% for logistics stations.

[0196] Machine learning: Use a random forest model to learn the mapping relationship between user behavior features and battery difference indicators, and output the importance of features, such as the influence weight of non-standard chargers on current deviation ≥40%.

[0197] The impact of behavioral influence data on battery difference indicators is quantified based on the correlation model, generating charging anomaly risk control data that includes behavioral risk correlation, abnormal risk probability, and risk diffusion coefficient.

[0198] The steps involved in quantification include:

[0199] The formula for quantifying the impact of a single user behavior on battery performance is as follows:

[0200]

[0201] In the formula, It's about the degree of influence. It is the battery difference value at the time this behavior occurs. It is the baseline difference value. It is the behavioral risk coefficient.

[0202] The weighted summation method is used to calculate the overall impact when multiple user behaviors exist. The formula is as follows:

[0203]

[0204] In the formula, It is the overall impact. It is the frequency weight of the behavior.

[0205] Behavioral risk correlation: the correlation coefficient between user behavior and battery status (0-1, ≥0.7 is considered a strong correlation).

[0206] Abnormal risk probability: The predicted probability that this combination of behaviors will cause the battery to enter a severely different state (0-100%, ≥50% triggers an alert).

[0207] Risk diffusion coefficient: The potential impact of a single user's high-risk behavior on surrounding charging stations / batteries, such as the fire risk diffusion range of illegally installed electrical wires.

[0208] Dynamic charging control strategies are generated based on battery status differences and charging anomaly risk control data.

[0209] The steps for generating a dynamic charging control strategy include:

[0210] By integrating battery status differences and charging anomaly risk control data, decision indicators are extracted for both battery and risk control dimensions, and different priorities are set for different decision indicators for different scenarios.

[0211] Battery dimension: Difference level, such as slight / moderate / severe, key abnormal indicators, such as temperature rise rate difference, internal resistance mutation coefficient.

[0212] Risk control dimensions: behavioral risk level, overall impact, and probability of abnormal risks.

[0213] Based on scenario type and priority, a framework for adapting strategies to different scenarios is built. Basic control parameters are calculated based on differences in battery status, and the control parameters are obtained by correcting them based on charging anomaly risk control data.

[0214] The steps to build a framework for adapting strategies to different scenarios may include: Community scenario: with the goal of safety protection and universal compatibility, while taking into account the convenience of residents' charging and avoiding excessive regulation.

[0215] Logistics terminal scenario: With the goal of ensuring efficiency and battery durability, the high-frequency fast charging requirement is met under the premise of safety.

[0216] Basic strategy framework construction: Community scenario: includes four major modules: dynamic power adjustment, charging time limit, operation specification guidance, and abnormal shutdown.

[0217] Logistics station scenario: includes five major modules: charging rate adaptation, temperature rise collaborative control, load balancing, V2G interaction, and equipment protection.

[0218] Scene-specific module activation: Community scene activation operation guide module, such as voice prompts for standardized plugging and unplugging.

[0219] In logistics terminal scenarios, the load balancing + V2G interaction module is activated, such as peak-shaving fast charging and reverse discharge when the power grid is overloaded.

[0220] The steps for calculating basic control parameters may include: minor differences: maintain basic charging parameters and optimize only details, such as keeping the charging power at 80% of the rated value in community scenarios and keeping the rate at 2C in logistics stations.

[0221] Moderate differences: Adjust core parameters, such as reducing power to 60% of the rated value in community scenarios and reducing the rate of change to 1.5C in logistics stations, and simultaneously start air cooling.

[0222] Significant differences: Emergency intervention, such as suspending charging, pushing battery detection reminders in community scenarios, and coordinating with logistics stations to replace batteries.

[0223] Correction steps may include: Low-risk behavior: Do not adjust parameters, only record the behavior trajectory.

[0224] Medium-risk behaviors: Additional constraints, such as reducing the charging time limit in community scenarios from 8 hours to 6 hours, and extending the continuous fast charging interval at logistics stations from 30 minutes to 45 minutes.

[0225] High-risk behaviors: Strong control measures, such as locking charging piles to prevent the user from charging, restricting access to non-standard equipment in community settings, and reducing the charging rate of the vehicle to 1C at logistics stations.

[0226] The control parameters are verified by setting constraints based on grid load, equipment compatibility, and extreme environments to determine whether the preset expectations have been met. If so, a dynamic charging control strategy is output; otherwise, the basic control parameters are recalculated.

[0227] Power grid load constraint verification: Edge nodes acquire regional power grid load data in real time. If the strategy parameters exceed the power grid capacity, such as the total peak power of the logistics station being greater than 100kW, the system will make dynamic adjustments, such as proportionally reducing the power of each charging pile to prioritize charging of operating vehicles.

[0228] Equipment compatibility constraint verification: Verify whether the strategy parameters are compatible with the charging pile and battery BMS protocol. If they are incompatible, trigger protocol conversion or parameter fine-tuning.

[0229] Extreme environment constraint verification: Low temperature: Correct the charging rate and extend the pre-charge time.

[0230] High temperature: Reduce power by an additional 20% and activate the condensation protection mechanism.

[0231] The control parameters, such as charging power, rate, duration limit, and warning method, are converted into instructions that can be executed by the charging pile. The edge node completes the distribution within 100ms, the execution delay in the community scenario is ≤200ms, and the execution delay in the logistics station is ≤100ms.

[0232] Edge nodes collect strategy execution feedback data every 2 seconds (community) / 1 second, such as changes in battery temperature and user operation adjustments. If the battery condition difference is alleviated, parameters are gradually restored, such as increasing power by 10% every 5 minutes. If the risk control risk does not decrease, management is strengthened, such as directly suspending charging.

[0233] This embodiment provides a multimodal extreme charging state safety control method for electric two-wheelers based on edge computing. It utilizes a lightweight CNN model, integrating multi-dimensional features such as geographical location, device density, and charging behavior to construct a scene classification model, achieving accurate identification of different extreme charging scenarios. Data classification, scene-specific denoising, feature extraction, and association mapping are completed at edge nodes, unifying heterogeneous data into a standardized format and reducing transmission latency and computing power consumption. A fusion model of backbone network + branch modules is built, assigning feature weights according to scenarios and combining dynamically updated scenario-based battery health benchmarks to achieve accurate quantification of battery state differences. User behavior is transformed into quantitative features, and a correlation model between behavior and battery differences is established through statistical analysis and machine learning. Impact is calculated based on a dedicated formula to generate multi-dimensional risk control data. Battery state differences and charging anomaly risk control data are integrated, prioritizing different scenarios and dynamically optimizing control parameters by combining grid load, device compatibility, and extreme environmental constraints. This significantly improves scenario adaptability and enhances charging safety.

[0234] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0235] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multimodal extreme charging state safety control method for electric two-wheeled vehicles based on edge computing, characterized in that, include: Identify the scenario types for extreme charging of electric two-wheeled vehicles, collect electrochemical parameters and thermal characteristic data during the extreme charging process under the corresponding scenario types, and collect user behavior data; The steps for identifying the scene type include: Obtain the geographical location of the target charging area, collect charging device density and device type as basic scene data, and collect charging behavior data and load characteristic data; The scene's geographical attributes, device density level, charging frequency characteristics, charging duration characteristics, load peak characteristics, time period distribution characteristics, battery cycle characteristics, and vehicle type characteristics are extracted from the basic scene data, the charging behavior data, and the load characteristic data as key scene features. Collect historical charging data containing key features of the scene, label the features, input the data into a lightweight CNN model, and train a scene classifier to obtain a scene classification model. Input the key features of the scene to be identified into the scene classification model, and output the scene type corresponding to the charging area; The steps for obtaining the electrochemical parameters and the thermal characteristic data include: Develop a scenario-specific data collection list based on the scenario type, and determine the equipment selection and deployment density; Based on the scenario-specific data collection list, the data collection frequency and triggering data collection mode are adapted, and the collected data is transmitted to the edge node through a preset transmission protocol; The collected data is subjected to noise filtering and outlier removal, and different formats are converted into a standardized JSON format. Scene identifiers, timestamps, and device IDs are added to obtain the electrochemical parameters and thermal characteristic data. Standardized multimodal extreme charge data is obtained by preprocessing the electrochemical parameters and thermal characteristic data through edge nodes, and the corresponding monitoring direction is set based on the scene type; The steps to obtain the standardized multimodal extreme-charge data include: The electrochemical parameters and thermal characteristic data are classified according to data type, and the standby data before charging and the data after power failure after charging are removed to obtain preliminary processed data. According to the scene type, a denoising method is selected to remove noise from the preliminary processed data to obtain multimodal data, and scene-specific anomaly thresholds are set. Temporal and dynamic features are extracted from the multimodal data as feature parameters, mapped to the [0,1] interval, the difference in units is eliminated, and the correspondence between the feature parameters in the same charging process is established through the timestamp and the device ID to obtain the standardized multimodal extreme charging data; A multimodal fusion model is constructed, and the battery status is evaluated by combining the monitoring direction and the standardized multimodal extreme charging data to obtain the battery difference status. The impact of the user behavior data on the battery difference status is analyzed to obtain charging anomaly risk control data. A dynamic charging control strategy is generated based on the battery's differential state and the charging anomaly risk control data.

2. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 1, characterized in that, The steps for setting the monitoring direction include: Based on the scenario type, identify the corresponding scenario objectives, break down the scenario-specific risks, and correlate them with user behavior risks to obtain risk breakdown data; Based on the risk decomposition data, a unified basic monitoring dimension is established, and based on scenario differences, the basic monitoring dimension is expanded to obtain scenario characteristic monitoring dimensions. Based on the scenario characteristic monitoring dimensions, electrochemical parameter monitoring indicators, thermal characteristic monitoring indicators, and user behavior monitoring indicators are set and integrated to obtain the total monitoring indicators. Based on the scenario type and the total monitoring indicators, the monitoring direction is obtained by setting the monitoring priority and early warning triggering mechanism corresponding to the scenario.

3. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 1, characterized in that, The steps for constructing the multimodal fusion model include: The electrochemical parameters are used as time-series modes, the thermal feature data as spatial modes, and the user behavior data as behavioral modes, and the corresponding basic feature vectors and mode priorities are determined. A cross-modal attention mechanism is adopted as the backbone network, branch modules are determined according to different modal characteristics, and the architecture configuration and a phased fusion mechanism are set according to the scenario type to obtain the basic architecture. Based on the aforementioned scenario type, historical scenario data is collected, and the infrastructure is trained for different scenarios to obtain the multimodal fusion model.

4. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 2, characterized in that, The steps for assessing the battery state differences include: From the standardized multimodal extreme charging data, core monitoring features related to the current scene monitoring direction are selected; Weights are assigned to the core monitoring features based on the monitoring priority, and the multimodal fusion model outputs key state parameters in parallel according to the scene type. Based on historical data of healthy batteries of the same type in the same scenario, a scenario-specific state benchmark is constructed. The scenario-specific state benchmark is updated as the battery usage cycle and environmental changes to obtain a scenario-based benchmark. By comparing the key state parameters with the scenario-based benchmark, a multidimensional difference index is calculated, and a comprehensive difference score is obtained by adjusting the score based on scenario characteristics. Based on the comprehensive difference score, the battery difference status is divided into multiple levels and a correspondence with safety risks is established to obtain the battery difference status.

5. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 1, characterized in that, The steps for obtaining the charging anomaly risk control data include: From the user behavior data, we filter out operation-related behaviors, device-related behaviors, and habit-related behaviors, and convert non-numerical behaviors into quantitative features to obtain behavior impact data; By combining historical fault data, the behavioral impact data is divided into risky behaviors, and behavioral risk thresholds are set for different scenarios; From the standardized multimodal extreme charging data, battery difference indicators related to user behavior are extracted, and a correlation model is established by combining statistical analysis and machine learning. The impact of the behavioral influence data on the battery difference index is quantified according to the association model, and the charging anomaly risk control data including behavioral risk correlation degree, abnormal risk probability and risk diffusion coefficient is generated.

6. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 5, characterized in that, The steps involved in quantification include: The formula for quantifying the impact of a single user behavior on battery performance is as follows: In the formula, It's about the degree of influence. It is the battery difference value at the time this behavior occurs. It is the baseline difference value. It is the behavioral risk coefficient; The weighted summation method is used to calculate the overall impact when multiple user behaviors exist. The formula is as follows: In the formula, It is the overall impact. It is the frequency weight of the behavior.

7. The method for multimodal extreme charging state safety control of electric two-wheeled vehicles based on edge computing according to claim 1, characterized in that, The steps for generating the dynamic charging control strategy include: By integrating the battery difference status with the charging anomaly risk control data, decision indicators for battery and risk control dimensions are extracted, and different priorities for decision indicators are set according to different scenarios. Based on the scenario type and the priority, a framework for adapting strategies for different scenarios is built. Basic control parameters are calculated based on the battery difference state, and control parameters are obtained by correcting them based on the charging anomaly risk control data. The control parameters are verified by setting constraints based on grid load, equipment compatibility, and extreme environments to determine whether the preset expectations have been met. If so, the dynamic charging control strategy is output; otherwise, the basic control parameters are recalculated.

Images