Electric vehicle battery charging and discharging management method

Abstract

The invention provides an electric vehicle battery charging and discharging management method, which comprises the following steps of: performing parameter fusion on a sensor data set to construct battery real-time state data; using the battery real-time state data and the environment data and the vehicle operation condition data corresponding to the battery real-time state data to construct vehicle charging and discharging state data, using the data to identify a battery state change trend and an abnormal state, and generating state change trend data and abnormal state diagnosis data; dynamically adjusting a battery charging and discharging strategy according to the vehicle charging and discharging state data and the state change trend data, and generating strategy adjustment data; and performing graded early warning according to the vehicle charging and discharging state data, the state change trend data, the abnormal state diagnosis data and the strategy adjustment data, and generating battery charging and discharging early warning data. Charging and discharging state data are constructed by fusing the real-time state, environment and working condition data of the battery, a charging and discharging strategy is dynamically adjusted, state trend recognition and abnormal early warning are carried out, and the service life of the battery is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicles, and particularly to a method for managing battery charging and discharging of an electric vehicle. Background Art

[0002] In the field of electric vehicles, the effectiveness of the battery charging and discharging management method plays a decisive role in vehicle performance, battery life, and usage cost. However, most of the existing battery charging and discharging management methods for electric vehicles adopt relatively fixed charging modes and cannot dynamically adjust the charging current and voltage according to the real-time state of the battery. During the charging and discharging process of the battery, its internal state changes constantly. Parameters such as battery power, temperature, and internal resistance will change with the charging and discharging time, usage environment, and battery aging degree. When the battery is in different usage stages and environmental conditions, if the fixed charging current and voltage are always used, a series of problems are likely to occur.

[0003] In some traditional charging and discharging management systems, although there are some simple protection mechanisms, such as overvoltage protection and overcurrent protection, these protection mechanisms can only work when the battery parameters exceed the set extreme thresholds and cannot optimize and adjust according to the real-time subtle changes of the battery throughout the charging and discharging process of the battery. For example, when the battery temperature gradually rises within the normal range, the existing system cannot perceive and adjust the charging and discharging strategy in advance until the temperature reaches the over-temperature protection threshold and takes measures, at which time the battery may have been damaged to a certain extent. Summary of the Invention

[0004] In view of this, the present invention proposes a method for managing battery charging and discharging of an electric vehicle, which solves the technical problem that the existing battery charging and discharging management method for an electric vehicle cannot dynamically adjust the charging current and voltage according to the real-time state of the battery, resulting in a short battery life.

[0005] A method for managing battery charging and discharging of an electric vehicle provided by the present invention includes the following steps:

[0006] Step S1: Obtain the sensor data set collected by each sensor during the charging of the electric vehicle, and perform parameter fusion on the sensor data set to generate real-time battery state data;

[0007] Step S2: Use the real-time battery state data, the corresponding environmental data, and the vehicle operating condition data of the real-time battery state data to construct vehicle charging and discharging state data;

[0008] Step S3: Based on the vehicle charging and discharging state data, identify the battery state change trend and abnormal state, and generate state change trend data and abnormal state diagnosis data;

[0009] Step S4: Dynamically adjust the battery charging and discharging strategy based on the vehicle charging and discharging state data and the state change trend data, and generate strategy adjustment data;

[0010] Step S5: Conduct hierarchical early warning based on the vehicle charging and discharging state data, the state change trend data, the abnormal state diagnosis data, and the strategy adjustment data, and generate battery charging and discharging early warning data.

[0011] Optionally, the specific steps of Step S1 are as follows:

[0012] Step S11: Clean the sensor data set to generate a first sensor data set;

[0013] Step S12: Normalize the first sensor data set to generate a second sensor data set;

[0014] Step S13: Extract features from the second sensor data set to generate multiple sensor features;

[0015] Step S14: Fuse the sensor features to generate fused data;

[0016] Step S15: Calculate the mean absolute error between the fused data and the battery state data corresponding to the sensor data set to generate error data;

[0017] Step S16: When the error data is less than the preset error threshold, use the fused data as the battery real-time state data.

[0018] Optionally, the specific steps of Step S3 are as follows:

[0019] Step S31: Predict the battery state change trend of the vehicle charging and discharging state data by using a preset state prediction model and a predictive control algorithm, generate first state change prediction data, and extract feature data;

[0020] Step S32: Use a deep learning algorithm to predict the battery state change trend of the vehicle charging and discharging state data and the feature data, and generate second state change prediction data;

[0021] Step S33: Conduct time dimension cross-validation based on the first state change prediction data and the second state change prediction data to determine the state change trend data;

[0022] Step S34: Judge the abnormal state of the vehicle charging and discharging state data and the preset hierarchical early warning threshold to generate first abnormal state data;

[0023] Step S35: Deeply analyze the first abnormal state data using an anomaly detection algorithm to generate second abnormal state data;

[0024] Step S36: Use intelligent dynamic adjustment strategy technology to perform multi-source data fusion assisted diagnosis on the first abnormal state data, the second abnormal state data, and the environmental data corresponding to the vehicle charging and discharging state data, and generate abnormal state diagnosis data.

[0025] Optionally, the preset state prediction model includes an electrochemical model, an equivalent circuit model, and an empirical model; the specific steps of step S31 are:

[0026] Step S311: Use an adaptive algorithm to dynamically correct the chemical reaction rate and equivalent circuit parameters in the preset state prediction model with the vehicle charging and discharging state data to generate a target state prediction model;

[0027] Step S312: Input the vehicle charging and discharging state data into the target state prediction model to predict the battery state change trend and generate initial prediction data;

[0028] Step S313: Use a traditional predictive control algorithm to formulate a charging and discharging strategy based on the initial prediction data and the vehicle charging and discharging state data to generate target prediction data;

[0029] Step S314: Use a reinforcement learning algorithm to optimize the target prediction data with the goal of maximizing battery life and minimizing energy consumption to generate first state change prediction data;

[0030] Step S315: Deeply extract features from the vehicle charging and discharging state data to generate feature data.

[0031] Optionally, the specific steps of step S32 are:

[0032] Step S321: Use an isolation forest algorithm to clean the vehicle charging and discharging state data to generate initial state data;

[0033] Step S322: Normalize the initial state data using a time series sliding window to generate target state data;

[0034] Step S323: Fuse the target state data and the feature data to generate state feature fusion data;

[0035] Step S324: Input the state feature fusion data into a preset trend prediction model to predict the battery state change trend and generate initial prediction data;

[0036] Step S325: Use a smoothing filter algorithm to remove noise and fluctuations in the initial prediction data and generate target prediction data;

[0037] Step S326: Calculate the confidence interval for the target prediction data to generate confidence interval data;

[0038] Step S327: Use the target prediction data and the confidence interval data to construct second state change prediction data.

[0039] Optionally, the specific steps of step S33 are as follows:

[0040] Step S331: Align the first state change prediction data and the second state change prediction data according to the timestamp and divide them into time windows to generate a first time window and a second time window;

[0041] Step S331: Calculate the prediction error and prediction accuracy of the first time window and the second time window respectively to generate a first prediction error, a second prediction error, a first prediction accuracy, and a second prediction accuracy;

[0042] Step S332: Evaluate the error consistency of the first prediction error and the second prediction error to generate error evaluation data;

[0043] Step S333: Evaluate the accuracy consistency of the first prediction accuracy and the second prediction accuracy to generate accuracy evaluation data;

[0044] Step S334: When both the error evaluation data and the accuracy evaluation data are within the preset evaluation threshold, use the first state change prediction data and the second state change prediction data to construct state change trend data.

[0045] Optionally, the specific steps of step S4 are as follows:

[0046] Step S41: Evaluate the current charge and discharge strategy of the electric vehicle using the state change trend data and the abnormal state diagnosis data to generate evaluation data;

[0047] Step S42: When the evaluation data indicates that the battery power is insufficient to meet the driving demand, increase the charging time or charging power of the electric vehicle to generate first strategy adjustment data;

[0048] Step S43: When the evaluation data indicates that the battery power is excessive, reduce the charging power of the electric vehicle or stop charging in advance to generate second strategy adjustment data;

[0049] Step S44: When the evaluation data is in the low grid load period, increase the charging power of the electric vehicle to generate third policy adjustment data;

[0050] Step S45: When the evaluation data is in the high grid load period, reduce the charging power of the electric vehicle or discharge it to generate fourth policy adjustment data;

[0051] Step S46: When the battery temperature in the evaluation data is greater than the preset temperature threshold, reduce the charging power of the electric vehicle or suspend charging and start the heat dissipation device to generate fifth policy adjustment data.

[0052] Optionally, the specific steps of step S41 are as follows:

[0053] S411: Synchronize the time and unify the format of the state change trend data and the abnormal state diagnosis data to generate a comprehensive state data set;

[0054] S412: Use the comprehensive state data set to identify the battery power trend, charge and discharge power trend, and battery temperature change slope to generate trend feature data;

[0055] S413: Use the trend feature data and the abnormal state diagnosis data to perform abnormal state matching and policy conflict identification to generate a policy conflict list;

[0056] S414: Use the policy conflict list data and the trend feature data to perform priority sorting to generate evaluation data.

[0057] Optionally, the specific steps of step S5 are as follows:

[0058] Step S51: Align the timestamps and standardize the formats of the vehicle charge and discharge state data, the state change trend data, the abnormal state diagnosis data, and the policy adjustment data to generate an integrated data set;

[0059] Step S52: Extract the key risk features in the integrated data set to generate real-time deviation, trend severity data, and abnormal event trigger data;

[0060] Step S53: Match the real-time deviation, the trend severity data, and the abnormal event trigger data with the risk thresholds in the preset risk threshold library respectively to generate a risk feature vector;

[0061] Step S54: Match the risk feature vector with the risk level data in the preset classification rule library according to the principle of abnormal event priority to generate battery charge and discharge warning data.

[0062] Optionally, the specific steps of step S54 are as follows:

[0063] Step S541: Determine whether the risk feature vector conforms to the abnormal events in the preset classification rule library. If so, execute Step S542; if not, execute Step S543;

[0064] Step S542: Match the risk level corresponding to the risk feature vector in the preset classification rule library to generate primary battery charge and discharge warning data;

[0065] Step S543: Determine whether the power value corresponding to the risk feature vector is less than the preset power threshold. If so, execute Step S544; if not, execute Step S545;

[0066] Step S544: Use the preset secondary warning data to construct secondary battery charge and discharge warning data;

[0067] Step S545: Use the preset tertiary warning data to construct tertiary battery charge and discharge warning data.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] The present invention generates real-time battery state data through multi-sensor data fusion, breaking the limitation of traditional single data acquisition, making the battery state information more comprehensive and accurate, and laying a foundation for subsequent dynamic management. By incorporating environmental and vehicle operating condition data to construct vehicle charge and discharge state data, considering the influence of external factors such as temperature and load on the battery, the data is closer to the actual usage scenario and the analysis results are more reliable. By identifying the state change trend data and abnormal state diagnosis data, potential problems can be discovered in advance, avoiding failures caused by battery performance deterioration and reducing the risk of accidental battery damage. Dynamically adjusting the battery charge and discharge strategy based on the vehicle charge and discharge state data and state change trend data, changing the previous fixed parameter charging mode, adapting the current and voltage according to the real-time state of the battery, reducing battery loss and extending the service life. Comprehensive classification warning of various types of data helps users and managers understand the battery risks in a timely manner and take corresponding measures according to the warning level, ensuring battery safety and facilitating targeted maintenance. Description of the Drawings

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only the preferred embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0071] Figure 1 It is a schematic flowchart of a method for managing battery charge and discharge of an electric vehicle according to the present invention;

[0072] Figure 2 Schematic flow diagram of step S1 of a method for managing charging and discharging of an electric vehicle battery according to the present invention;

[0073] Figure 3 Schematic flow diagram of step S3 of a method for managing charging and discharging of an electric vehicle battery according to the present invention;

[0074] Figure 4 Schematic flow diagram of step S4 of a method for managing charging and discharging of an electric vehicle battery according to the present invention;

[0075] Figure 5 Schematic flow diagram of step S5 of a method for managing charging and discharging of an electric vehicle battery according to the present invention. Detailed implementation manners

[0076] An embodiment of the present invention provides a method for managing charging and discharging of an electric vehicle battery, which is used to solve the technical problem that the existing method for managing charging and discharging of an electric vehicle battery cannot dynamically adjust the charging current and voltage according to the real-time state of the battery, resulting in a short battery life.

[0077] To make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0078] Please refer to Figure 1 , a method for managing charging and discharging of an electric vehicle battery provided by the present invention includes:

[0079] Step S1, obtaining a sensor data set collected by each sensor when the electric vehicle is charging, performing parameter fusion on the sensor data set, and generating real-time battery state data;

[0080] Step S2, constructing vehicle charging and discharging state data by using the real-time battery state data, the environmental data corresponding to the real-time battery state data, and the vehicle operation condition data;

[0081] Step S3, identifying the battery state change trend and abnormal state based on the vehicle charging and discharging state data, and generating state change trend data and abnormal state diagnosis data;

[0082] Step S4, dynamically adjusting the battery charging and discharging strategy according to the vehicle charging and discharging state data and the state change trend data, and generating strategy adjustment data;

[0083] Step S5: Perform hierarchical early warning based on the vehicle charging and discharging status data, status change trend data, abnormal status diagnosis data, and policy adjustment data to generate battery charging and discharging early warning data.

[0084] In the embodiment of the present invention, a sensor data set collected by each sensor during the charging of an electric vehicle is obtained. Here, the sensors include a battery voltage sensor, a current sensor, a temperature sensor, an SOC (state of charge) sensor, an SOH (state of health) sensor, etc. The original data is collected in real time at a preset frequency (such as 100 Hz) through these sensors to obtain a sensor data set. The data format of the sensor data set is structured data (such as JSON or CSV), including fields such as a timestamp, a sensor ID, a measured value, and a unit. The sensor data set is sequentially passed through data cleaning (removing outliers and noise filtering), normalization (unifying the dimension), and feature extraction (statistical / time-domain / frequency-domain features) to generate a standardized feature vector. An adaptive weighted fusion algorithm is used to integrate the feature vectors, and by comparing the calculated mean absolute error (MAE) with a preset threshold (such as voltage MAE ≤ 0.02 V), the battery real-time status data (including more than 20 key parameters) is output after verification.

[0085] The environmental data includes the external temperature, humidity, grid load level (low valley / flat peak / high peak), and real-time electricity price. The vehicle operating condition data includes the current operating state of the vehicle (such as stationary charging, charging / discharging during driving), charging mode (fast charging / slow charging / wireless charging), load power (such as air conditioning, motor drive demand), historical driving mileage, driving habits (frequency of rapid acceleration / deceleration), etc., which reflects the actual working load of the battery. The three types of data, namely the battery real-time status data, the environmental data corresponding to the battery real-time status data, and the vehicle operating condition data, are aligned according to the timestamp to construct a multi-dimensional vector (such as [voltage, SOC, environmental temperature, grid load, vehicle speed, timestamp]) including battery parameters, environmental parameters, and condition parameters, forming the vehicle charging and discharging status data.

[0086] A multi-model prediction is performed on the vehicle charging and discharging status data by using an electrochemical model, an equivalent circuit model (dynamically modifying parameters), and a deep learning algorithm (such as a long short-term memory network). Through cross-validation in the time dimension (comparing the prediction error and accuracy, such as RMSE < 5% and R 2 > 0.95), the status change trend data (such as the SOC prediction curve for the next 30 minutes) is generated. In terms of abnormal diagnosis, it is a two-level detection. The first-level detection: Generate the first abnormal status data based on a preset threshold (such as voltage > 4.2 V or < 2.5 V); The second-level detection: Use algorithms such as the isolation forest to deeply analyze the data fluctuations, and perform multi-source fusion diagnosis in combination with environmental data (such as high temperature), and output the abnormal status diagnosis data including the fault type and severity.

[0087] Integrate the status change trend data and abnormal status diagnosis data, identify key influencing factors such as power demand, grid load, and battery temperature, and generate evaluation data (such as "insufficient power", "grid peak"). Execute corresponding strategies for different evaluation results. For example, increase the charging time or power when the power is insufficient; increase the charging power during the grid valley period, and reduce the power or discharge during the grid peak period; suspend charging and start heat dissipation when the battery temperature is abnormal. Output strategy adjustment data including adjustment targets, parameters, and execution times (such as "Increase the charging power to 30kW and execute immediately").

[0088] Align the timestamps of the vehicle charge and discharge status data, status change trend data, abnormal status diagnosis data, and strategy adjustment data, and standardize the vehicle charge and discharge status, trend, diagnosis, and strategy data to form a unified format of integrated data set. Extract features such as real-time deviation (such as the voltage overlimit ratio), trend severity (such as the SOC decline rate), and abnormal event trigger (such as overcharging), and match them with the preset threshold library to generate a risk feature vector. Match the risk feature vector with the risk level data in the preset classification rule library according to the principle of abnormal event priority to generate battery charge and discharge early warning data.

[0089] The present invention generates real-time battery status data through multi-sensor data fusion, breaking through the limitations of traditional single data collection, making the battery status information more comprehensive and accurate, and laying a foundation for subsequent dynamic management. By incorporating environmental and vehicle operating condition data to construct vehicle charge and discharge status data, considering the influence of external factors such as temperature and load on the battery, the data is closer to the actual usage scenario, and the analysis results are more reliable. By identifying the status change trend data and abnormal status diagnosis data, potential problems can be discovered in advance, avoiding failures caused by battery performance deterioration, and reducing the risk of accidental battery damage. Dynamically adjust the battery charge and discharge strategy based on the vehicle charge and discharge status data and status change trend data, changing the previous fixed parameter charging mode, adapting the current and voltage according to the real-time battery status, reducing battery loss, and extending the service life. Comprehensively classify and warn various types of data to help users and managers understand the battery risks in a timely manner, and take corresponding measures according to the warning level, which not only ensures battery safety but also facilitates targeted maintenance. It solves the technical problem that the existing electric vehicle battery charge and discharge management method cannot dynamically adjust the charging current and voltage according to the real-time battery status, resulting in a short battery life.

[0090] Preferably, please refer to Figure 2 , the specific steps of step S1 are as follows:

[0091] Step S11: Clean the sensor data set to generate the first sensor data set;

[0092] Step S12: Normalize the first sensor data set to generate the second sensor data set;

[0093] Step S13: Extract features from the second sensor dataset to generate multiple sensor features;

[0094] Step S14: Perform feature fusion on the sensor features to generate fusion data;

[0095] Step S15: Calculate the mean absolute error between the fusion data and the battery state data corresponding to the sensor dataset to generate error data;

[0096] Step S16: When the error data is less than the preset error threshold, use the fusion data as the real-time battery state data.

[0097] In the embodiments of the present invention, during the data acquisition process of the sensor, it may be affected by factors such as environmental interference and equipment failure, resulting in noise data, missing data, or incorrect data. Data cleaning is to screen and process the original sensor dataset, remove outliers and duplicate values therein, fill in missing values, and correct incorrect data. Through this step, a cleaner and more accurate first sensor dataset is obtained.

[0098] The data collected by different sensors may have large differences in numerical range and dimension. For example, the data range collected by a temperature sensor may be between -40°C and 120°C, while the data collected by a voltage sensor may be between a few volts and several hundred volts. Data normalization is to map all the data in the first sensor dataset to the same numerical interval (usually [0, 1] or [-1, 1]) according to certain rules, eliminating the influence of dimension and numerical range differences. This can make subsequent data processing and analysis more fair and efficient, improve the accuracy and stability of data processing, and obtain the second sensor dataset.

[0099] Although the second sensor dataset has been cleaned and normalized, the data volume may be large and contain redundant information. Feature extraction is to extract the key information that can best represent the battery state from these data to form multiple sensor features. These features may include the battery voltage change rate, temperature gradient, charge and discharge current fluctuation amplitude, etc. Through feature extraction, the data structure is simplified, the key information is highlighted, the complexity of data processing is reduced, and at the same time, the data features valuable for battery state judgment are retained.

[0100] Each sensor feature reflects the battery state from different angles, but a single feature cannot comprehensively describe the real-time situation of the battery. Feature fusion is to integrate multiple sensor features and combine these scattered features into a comprehensive fusion data through methods such as weighted summation and principal component analysis. The fusion data can more comprehensively and accurately reflect the overall state of the battery, making the subsequent analysis of the battery state more comprehensive and in-depth.

[0101] To verify the accuracy of the fused data, the fused data is compared with the known battery state data corresponding to the sensor data set. The mean absolute error is a metric that measures the error between the predicted values (fused data) and the true values (known battery state data). It calculates the absolute error for each data point (the absolute value of the difference between the predicted value and the true value), and then takes the average of these absolute errors. By calculating the mean absolute error, error data is obtained, which is used to evaluate the reliability and accuracy of the fused data.

[0102] The preset error threshold is a standard set according to the accuracy requirements of the battery state data and the actual application scenario. When the error data calculated in step S15 is less than the preset error threshold, it indicates that the error between the fused data and the true battery state data is within an acceptable range, and the fused data can accurately reflect the real-time state of the battery. At this time, the fused data is used as the battery real-time state data. Conversely, if the error data is greater than the preset error threshold, it means that the accuracy of the fused data is insufficient, and the data processing process needs to be rechecked or relevant parameters need to be adjusted until the error meets the requirements.

[0103] Preferably, please refer to Figure 3 , the specific steps of step S3 are as follows:

[0104] Step S31: Use the preset state prediction model to adopt the predictive control algorithm to predict the battery state change trend of the vehicle charge and discharge state data, generate the first state change prediction data, and extract the feature data;

[0105] Step S32: Use the deep learning algorithm to predict the battery state change trend of the vehicle charge and discharge state data and the feature data, and generate the second state change prediction data;

[0106] Step S33: Perform time dimension cross-validation based on the first state change prediction data and the second state change prediction data to determine the state change trend data;

[0107] Step S34: Judge the abnormal state of the vehicle charge and discharge state data and the preset grading warning threshold to generate the first abnormal state data;

[0108] Step S35: Use the anomaly detection algorithm to deeply analyze the first abnormal state data to generate the second abnormal state data;

[0109] Step S36: Use the intelligent dynamic adjustment strategy technology to perform multi-source data fusion assisted diagnosis on the first abnormal state data, the second abnormal state data, and the environmental data corresponding to the vehicle charge and discharge state data to generate the abnormal state diagnosis data.

[0110] In the embodiments of the present invention, the preset state prediction model includes an electrochemical model (based on lithium-ion diffusion kinetics), an equivalent circuit model (such as a second-order RC network), and an empirical model (such as a regression model based on historical data). The predictive control algorithm belongs to the model-based rolling horizon optimization control method. Its core idea is to establish a dynamic model of the battery system (such as an equivalent circuit model, an electrochemical model, etc.), and use the current and historical vehicle charge and discharge state data (such as voltage, current, temperature, SOC / SOH, etc.) to perform rolling prediction on the state change trend of the battery in the next period of time (such as voltage fluctuation, capacity decay, temperature change, etc.), and consider system constraints (such as safety voltage / current thresholds, battery life constraints, etc.) in each step of optimization. An adaptive algorithm (such as the Extended Kalman Filter EKF) is used to dynamically adjust the model parameters (such as polarization resistance, diffusion coefficient) according to the real-time charge and discharge state data (voltage, current, temperature) to ensure that the model matches the current aging state of the battery. The input of the corrected model and the vehicle charge and discharge state data are used to predict the battery state (such as SOC, terminal voltage, temperature change curve) in the next 10 - 30 minutes, and generate the first state change prediction data. Time domain features (such as current change rate dI / dt), frequency domain features (such as AC internal resistance fluctuation frequency), and statistical features (such as temperature standard deviation) are extracted from the vehicle charge and discharge state data to form a feature vector of more than 50 dimensions, and feature data is obtained.

[0111] The Isolation Forest algorithm is used to detect and remove outliers (such as sensor instantaneous fault data) in the charge and discharge state data. Through time series sliding window normalization (window size 5 minutes), parameters such as voltage and current are mapped to the [0, 1] interval to meet the input requirements of the deep learning model. The preprocessed charge and discharge state data and the feature data generated in S31 are concatenated to form a fusion dataset containing time series information and deep features. An LSTM network (3 hidden layers, 128 neurons in each layer) is used to train the fusion dataset to predict the battery state change in the next 30 minutes and generate the second state change prediction data (including prediction values and confidence intervals). Utilize the non-linear fitting ability of deep learning to capture the battery state change law under complex working conditions (such as frequent start-stop, extreme temperature) that are difficult to describe by traditional models. Combine the feature data of S31 to make up for the defect that the physical model is sensitive to noise and improve the prediction robustness.

[0112] Align the first state change prediction data and the second state change prediction data according to the timestamp (accuracy 1 second) and divide them into 15-minute time windows (such as the current moment t to t + 15 min). Calculate the root mean square error (RMSE) and the coefficient of determination (R 2 ) of each window, and the formulas are as follows:

[0113]

[0114] Among them, n is the number of samples within a single time window (the length of the time window is related to the data acquisition frequency). If the time window is 15 minutes and the acquisition frequency is 1 Hz (1 sample per second), then n = 15 × 60 = 900. y i is the true value of the battery state at the i-th time point (obtained through on-site measurement by sensors or off-line calibration). For example: the battery terminal voltage (unit: V), SOC (state of charge, percentage), temperature (unit: °C), etc. Data source: the measured parameters in the "vehicle charging and discharging state data" constructed in step S2 (such as the real-time acquisition value of the voltage sensor). is the predicted value of the battery state at the i-th time point (output from the traditional model in step S31 or the deep learning model in step S32). For example: if predicting the SOC change in the next 10 minutes, is the predicted result of the model for the SOC at the i-th minute. is the average value of the true values of the battery state, serving as a reference value to reflect the central tendency of the real data (such as the average level of SOC within a certain time window). is the total squared error between the predicted value and the true value. is the total squared error between the true value and the average value (i.e., when not using any model and only using the average value for prediction, reflecting the degree of fluctuation of the data itself).

[0115] When the RMSE of both types of models is < 8% and R 2 > 0.9, take the weighted average (weights are dynamically allocated according to historical accuracy) as the final state change trend data; otherwise, trigger online fine-tuning of the model. By comparing the prediction results of the traditional model and the deep learning model, select the trend data with the highest reliability and reduce the prediction bias of a single model (such as the inaccuracy of the physical model under extreme working conditions and the overfitting of the deep learning model).

[0116] Preset hierarchical warning thresholds: including safety thresholds (such as voltage > 4.2 V / < 2.5 V, temperature > 55 °C), performance thresholds (such as SOC change rate > 15% / h, internal resistance growth rate > 5% / month). Directly compare the vehicle charging and discharging state data with the threshold library. If the parameters exceed the limits, mark them as abnormal and generate the first abnormal state data (only including the boolean value of "whether the limit is exceeded", such as "voltage overlimit: yes"). It can quickly capture obvious abnormalities (such as overcharging, overheating) and provide an immediate trigger signal for emergency protection (such as cutting off the charging circuit) to ensure that the battery safety boundary is not breached.

[0117] Adopt algorithms such as Isolation Forest and One-Class Support Vector Machine to analyze the original signals (such as voltage waveforms and current curves) associated with the first abnormal state data. Identify "non-threshold abnormalities" (such as voltage within the safe range but the fluctuation amplitude > 10%, which may indicate poor internal contact), and distinguish between "real faults" and "sensor noise" (such as judging whether abnormal points appear continuously through DBSCAN clustering), generating the second abnormal state data (including the type of abnormality, such as "contact fault", "sensor drift"). Avoid false alarms / missing alarms caused by single-threshold judgment (the false alarm rate of the traditional method is about 20%, and this step is reduced to less than 5%), for example, identify early faults such as micro-short circuits inside the battery (voltage not exceeding the limit but abnormal fluctuations).

[0118] Integrate the first abnormal state data (whether exceeding the limit), the second abnormal state data (type of abnormality), environmental data (external temperature, grid frequency), and operating condition data (such as instantaneous large current caused by rapid acceleration). Use Bayesian networks or rule engines to establish an "abnormal feature - cause - impact" association model. For example:

[0119] If "voltage exceeds the limit + external temperature > 45°C", it is determined as "high temperature causes increased polarization", rather than a battery body fault;

[0120] If "abnormal current fluctuation + SOH < 80%", it is determined as "battery aging causes internal resistance fluctuation".

[0121] Through the "abnormal feature - cause - impact" association model, use intelligent dynamic adjustment strategy technology to perform multi-source data fusion-assisted diagnosis on the first abnormal state data, the second abnormal state data, and the environmental data corresponding to the vehicle charging and discharging state data, generating abnormal state diagnosis data, including the cause of the abnormality (such as "excessive charging current → temperature rise → voltage exceeds the limit"), the degree of impact (mild / moderate / severe), and recommended measures (such as "reduce the charging power by 30%"). Combining environmental and operating condition data, avoid isolated analysis of the battery state, and improve the comprehensiveness and accuracy of abnormal diagnosis. Provide a dual basis of "fault location + risk level" for strategy adjustment. For example, for the "sensor drift" abnormality, only trigger a prompt warning instead of directly shutting down the machine, reducing the impact on the user experience.

[0122] Preferably, the specific steps of step S31 are as follows:

[0123] Step S311: Use an adaptive algorithm to dynamically correct the chemical reaction rate and equivalent circuit parameters in the preset state prediction model with the vehicle charging and discharging state data, generating a target state prediction model;

[0124] Step S312: Input the vehicle charging and discharging state data into the target state prediction model to predict the battery state change trend, generating initial prediction data;

[0125] Step S313: Use the traditional predictive control algorithm to formulate the charging and discharging strategy based on the initial prediction data and the vehicle charging and discharging state data, and generate the target prediction data;

[0126] Step S314: Use the reinforcement learning algorithm to optimize the target prediction data with the goal of maximizing battery life and minimizing energy consumption, and generate the first state change prediction data;

[0127] Step S315: Perform deep feature extraction on the vehicle charging and discharging state data to generate feature data.

[0128] In the embodiment of the present invention, the preset state prediction model includes an electrochemical model, an equivalent circuit model, and an empirical model. The electrochemical model is based on the intercalation / deintercalation kinetics of lithium ions in the positive and negative electrode materials, and includes the Nernst equation, Fick's diffusion law, etc., to describe the electrochemical reaction rate inside the battery (such as the solid-phase diffusion coefficient, charge transfer resistance). The equivalent circuit model uses an RC network to simulate the polarization characteristics of the battery (such as a second-order RC model, including parameters such as ohmic internal resistance, polarization resistance, and polarization capacitance). The empirical model is a regression model fitted based on historical data (such as polynomial fitting, Gaussian process regression) to describe the mapping relationship between the input (current, temperature) and the output (voltage, SOC).

[0129] Adopt the Extended Kalman Filter (EKF) or the Unscented Kalman Filter (UKF), use the vehicle charging and discharging state data (voltage U, current I, temperature T, SOC) collected in real time as the observed values, and iteratively correct the model parameters. Update the model parameters every 50 ms to ensure that the model matches the current aging state of the battery in real time (such as the increase in internal resistance caused by the thickening of the SEI film, the decrease in capacity caused by the attenuation of the active material), and generate the target state prediction model. Solve the time-varying characteristics of the battery parameters (such as the internal resistance fluctuation caused by temperature change, the capacity attenuation caused by the increase in the number of cycles), and reduce the model prediction error by more than 40% (compared with the fixed-parameter model).

[0130] Input the vehicle charging and discharging state data (including more than 50-dimensional parameters such as voltage, current, temperature, SOC, ambient temperature, grid load, etc.) constructed in Step S2 into the target state prediction model. Among them, the electrochemical model predicts the SOC, terminal voltage, and temperature change curves (time step 1 second) in the next 10-30 minutes by solving partial differential equations (such as the solid-phase diffusion equation, electrolyte concentration distribution equation). The equivalent circuit model calculates the transient response of the RC network based on the current input and outputs the dynamic changes of components such as polarization voltage and ohmic voltage. The empirical model directly maps the input to the output through a fitting function to provide fast prediction (computing delay < 10 ms). Generate the initial prediction data including the predicted values of the physical quantities of the battery state (such as "the SOC will increase from 40% to 75% and the terminal voltage will increase from 3.6V to 4.1V in the next 15 minutes").

[0131] The specific process of formulating the charging and discharging strategy based on the initial prediction data and the vehicle charging and discharging state data using the traditional predictive control algorithm is as follows:

[0132] 1. Adopt model predictive control (MPC), and based on the target state prediction model, construct an optimization problem: where u t is the charging and discharging current / voltage control quantity, y t is the predicted state, is the target state (such as the SOC charging target of 80%), and ρ is the control quantity penalty coefficient.

[0133] 2. The constraint conditions are: safety constraints: voltage 2.5V ≤ U ≤ 4.2V, current -I max ≤ I ≤ I max (negative value for discharging), temperature -20°C ≤ T ≤ 60°C; performance constraints: charging efficiency ≥ 90%, capacity retention rate SOH ≥ 80% (long-term constraint).

[0134] 3. Generate target prediction data, that is, the charging and discharging strategy sequence for the next 10 minutes (such as "charging current 100A in the first minute, 95A in the second minute, gradually decreasing to 50A in the tenth minute").

[0135] Use the reinforcement learning algorithm to optimize the target prediction data with the goal of the longest battery life and the lowest energy consumption. Specifically, use the deep Q network (DQN) or the policy gradient algorithm (such as PPO), and through interactive training with the target state prediction model, generate the first state change prediction data (optimized charging and discharging strategy, such as "automatically reduce the initial charging current in a low-temperature environment to avoid the thickening of the SEI film"). Break through the limitation of the single-target optimization of traditional MPC and achieve the multi-target balance of "the longest battery life + the lowest energy consumption" (the measured cycle life is increased by 25% and the energy consumption is reduced by 15%).

[0136] Standardize (z-score) and normalize (map to [0, 1]) the vehicle charging and discharging state data to generate feature data with 50+ dimensions (such as "[0.8, 0.3, 0.6,...]", corresponding to the normalized values of different features). Among them, the feature data includes time-domain features, frequency-domain features, and statistical features. Provide richer input information for the deep learning model in step S32, and make up for the insufficient description of the physical model for non-linear features (such as noise, abnormal fluctuations). Improve the subsequent trend prediction accuracy (such as when the LSTM model combines with the feature data, the SOC prediction error is reduced from 10% to 6%), especially suitable for the weak feature recognition in the early stage of battery aging (such as the slight increase in internal resistance when SOH drops by 1%).

[0137] Preferably, the specific steps of step S32 are:

[0138] Step S321: Use the Isolation Forest algorithm to clean the vehicle charge and discharge state data to generate initial state data;

[0139] Step S322: Perform time-series sliding window normalization on the initial state data to generate target state data;

[0140] Step S323: Fuse the target state data and the feature data to generate state feature fusion data;

[0141] Step S324: Input the state feature fusion data into a preset trend prediction model to predict the battery state change trend and generate initial prediction data;

[0142] Step S325: Use a smoothing filter algorithm to remove the noise and fluctuations in the initial prediction data to generate target prediction data;

[0143] Step S326: Calculate the confidence interval for the target prediction data to generate confidence interval data;

[0144] Step S327: Use the target prediction data and the confidence interval data to construct the second state change prediction data.

[0145] In the embodiment of the present invention, the Isolation Forest is an unsupervised anomaly detection algorithm based on a binary tree structure. It randomly partitions the data space to construct multiple "isolation trees", and the path length of each data point in the tree reflects its "isolation degree". The shorter the path, the more likely the data point is an outlier. The cleaning process of using the Isolation Forest algorithm to clean the vehicle charge and discharge state data is as follows: taking the vehicle charge and discharge state data (time series including parameters such as voltage, current, temperature, and SOC) as input; calculating the average path length of each data point in all isolation trees, and setting a threshold (such as the average path length being lower than 30% of the global average) to determine as an anomaly; removing the abnormal data and retaining the normal data to generate the initial state data. It can quickly identify and remove instantaneous sensor failures (such as the voltage suddenly becoming 0V) and invalid data caused by communication anomalies, reducing the interference of noise on subsequent predictions.

[0146] Define a time window with a fixed size (such as 5 minutes), slide point by point in chronological order, and perform normalization processing on the initial state data within the window. Use Min-Max Scaling to map the data within each window to the [0, 1] interval to generate target state data, ensuring the consistency of the data in the time dimension and eliminating the dimensional differences of different parameters (such as voltage and temperature).

[0147] The target state data (charge and discharge state data that has been cleaned and normalized) is concatenated with the feature data generated in step S315 (such as current change rate, voltage frequency domain features) to form state feature fusion data that includes original time series information and depth features (such as a vector sequence of 50+ dimensions), which is used as the input to the preset trend prediction model.

[0148] The preset trend prediction model is a neural network model that uses a long short-term memory network (LSTM) or a gated recurrent unit (GRU) to construct 3 hidden layers (with 128 neurons in each layer). The model is trained using historical state feature fusion data (such as data from the past 1 hour), with the battery state (such as SOC, voltage) in the next 10 - 30 minutes as the label; the current state feature fusion data is input into the trained model to output initial prediction data (such as "the SOC will increase from 50% to 75% in the next 15 minutes").

[0149] Savitzky-Golay filtering or median filtering is used to smooth the initial prediction data. The filtering window size is set (such as 3 time steps), and high-frequency noise and local fluctuations are eliminated by polynomial fitting or sorting to take the median, generating smoother target prediction data (such as a continuous SOC change curve) and reducing the random jitter of the predicted values.

[0150] The Bootstrap resampling method or the Delta method is used to sample the target prediction data multiple times (such as 1000 times); the statistics (such as mean, standard deviation) of each sampling result are calculated, and the confidence interval boundaries are determined according to the set confidence level (such as 95%); confidence interval data is generated (such as "the predicted value of SOC in the next 15 minutes is 75%, and the confidence interval is [73%, 77%]"). The target prediction data (the smoothed predicted value) is combined with the confidence interval data; second state change prediction data that includes the predicted value, confidence interval, and timestamp is generated, which is used as the input for subsequent cross-validation (S33).

[0151] Preferably, the specific steps of step S33 are as follows:

[0152] Step S331: Align the first state change prediction data and the second state change prediction data according to the timestamp and divide them into time windows to generate the first time window and the second time window;

[0153] Step S332: Calculate the prediction errors and prediction accuracies of the first time window and the second time window respectively to generate the first prediction error, the second prediction error, the first prediction accuracy, and the second prediction accuracy;

[0154] Step S333: Conduct an error consistency evaluation on the first prediction error and the second prediction error to generate error evaluation data;

[0155] Step S334: Perform accuracy consistency evaluation on the first prediction accuracy and the second prediction accuracy to generate accuracy evaluation data;

[0156] Step S335: When both the error evaluation data and the accuracy evaluation data are within the preset evaluation thresholds, use the first state change prediction data and the second state change prediction data to construct state change trend data.

[0157] In an embodiment of the present invention, the first state change prediction data (from step S31, based on a traditional model) and the second state change prediction data (from step S32, based on a deep learning model) are aligned according to timestamps to ensure that the two sets of data are in one-to-one correspondence in the time series. A fixed time window length (such as 15 minutes) is set, and starting from the current moment, windows are sequentially divided with a sliding step (such as 1 minute) to generate multiple first time windows and second time windows. For example, the first window covers the data from the current moment t to t + 15 minutes, the next window covers t + 1 to t + 16 minutes, and so on.

[0158] For each time window, the root mean square error (RMSE) is used to evaluate the deviation between the predicted value and the true value; the coefficient of determination (R 2 ) is used to measure the ability of the model to explain the data fluctuations. Generate a first prediction error (traditional model), a second prediction error (deep learning model), a first prediction accuracy (traditional model), and a second prediction accuracy (deep learning model) for each time window.

[0159] Calculate the difference between the first prediction error and the second prediction error within the same time window, and compare the difference with a preset error consistency threshold (such as RMSE difference ≤ 0.05V). If the difference is within the threshold, mark it as "error consistent" and generate the error evaluation data for the corresponding window (such as "Window 1: Error consistent, √"); otherwise, mark it as "error inconsistent".

[0160] Calculate the difference between the first prediction accuracy and the second prediction accuracy within the same time window, and compare the difference with a preset accuracy consistency threshold (such as R 2 difference ≤ 0.03). If the difference is within the threshold, mark it as "accuracy consistent" and generate the accuracy evaluation data for the corresponding window (such as "Window 2: Accuracy consistent, √"); otherwise, mark it as "accuracy inconsistent".

[0161] When both the error evaluation data and the accuracy evaluation data meet the preset thresholds (that is, all windows are "error consistent" and "accuracy consistent"), perform weighted averaging on the prediction data of the two types of models (the weights are dynamically assigned according to historical accuracy, such as the weight of the traditional model is 0.4 and the weight of the deep learning model is 0.6) to generate the final state change trend data (such as the SOC prediction curve for the next 30 minutes, including the predicted value and the confidence interval).

[0162] Preferably, refer to Figure 4 , the specific steps of step S4 are as follows:

[0163] Step S41: Evaluate the current charging and discharging strategy of the electric vehicle by using the state change trend data and abnormal state diagnosis data to generate evaluation data;

[0164] Step S42: When the evaluation data indicates that the battery power is insufficient to meet the driving demand, increase the charging time or charging power of the electric vehicle to generate the first strategy adjustment data;

[0165] Step S43: When the evaluation data indicates that the battery power is excessive, reduce the charging power of the electric vehicle or stop charging in advance to generate the second strategy adjustment data;

[0166] Step S44: When the evaluation data is in the low grid load period, increase the charging power of the electric vehicle to generate the third strategy adjustment data;

[0167] Step S45: When the evaluation data is in the high grid load period, reduce the charging power of the electric vehicle or discharge to generate the fourth strategy adjustment data;

[0168] Step S46: When the battery temperature in the evaluation data is greater than the preset temperature threshold, reduce the charging power of the electric vehicle or pause charging and start the heat dissipation device to generate the fifth strategy adjustment data.

[0169] In the embodiment of the present invention, the state change trend data (such as the SOC change trend and voltage change trend of the battery in a future period of time) generated in step S33 and the abnormal state diagnosis data (such as whether there is an abnormality in the battery, the cause of the abnormality, and the degree of influence) generated in step S36 are used as inputs. Set multiple evaluation indicators, such as the matching degree between the battery power and the driving demand, the grid load period at the current time, the battery temperature, etc. According to these indicators and preset rules, evaluate the current charging and discharging strategy of the electric vehicle. For example, judge whether the battery power meets the driving demand according to the remaining cruising range of the vehicle and the preset driving demand; query the grid load information according to the current time to determine whether it is in the low or high grid load period. Generate evaluation data, which contains the evaluation conclusion of the current charging and discharging strategy, such as "the battery power is insufficient to meet the driving demand", "the battery power is excessive", "in the low grid load period", "in the high grid load period", "the battery temperature is greater than the preset temperature threshold", etc.

[0170] When the evaluation data shows that "the battery power is insufficient to meet the driving demand", increase the charging time of the electric vehicle. For example, extend the originally planned charging duration from 1 hour to 1.5 hours; or increase the charging power, such as increasing the charging power from 5 kW to 7 kW. Thus, the first strategy adjustment data is obtained, including the adjusted charging time or charging power information.

[0171] When the evaluation data shows that "the battery power is excessive", reduce the charging power of the electric vehicle. For example, reduce the charging power from 7 kW to 5 kW; or stop charging in advance to terminate the current charging process. Thus, the second strategy adjustment data is obtained, including the adjusted charging power or the instruction to stop charging.

[0172] When the evaluation data shows that "it is in the low grid load period", increase the charging power of the electric vehicle and make full use of the low-price electric energy during the low grid load period for fast charging. For example, increase the charging power from 5 kW to 10 kW. Thus, the third strategy adjustment data is generated, including the adjusted charging power information.

[0173] When the evaluation data shows that "it is in the high grid load period", reduce the charging power of the electric vehicle, such as reducing the charging power from 10 kW to 3 kW; or when the battery power permits, let the electric vehicle discharge and feed the electric energy back to the grid. Thus, the fourth strategy adjustment data is obtained, including the adjusted charging power or the discharge instruction.

[0174] When the evaluation data shows that "the battery temperature is greater than the preset temperature threshold", reduce the charging power of the electric vehicle, such as reducing the charging power from 8 kW to 4 kW; or pause charging, stop the current charging process, and start the heat dissipation device (such as a fan, a coolant circulation system, etc.) to dissipate heat from the battery. Thus, the fifth strategy adjustment data is obtained, including the adjusted charging power, the instruction to pause charging, and the instruction to start the heat dissipation device.

[0175] Preferably, the specific steps of step S41 are as follows:

[0176] S411. Synchronize the time and unify the format of the state change trend data and the abnormal state diagnosis data to generate a comprehensive state data set;

[0177] S412. Use the comprehensive state data set to identify the battery power trend, the charge and discharge power trend, and the battery temperature change slope to generate trend feature data;

[0178] S413. Use the trend feature data and the abnormal state diagnosis data to perform abnormal state matching and strategy conflict identification to generate a strategy conflict list;

[0179] S414. Use the strategy conflict list data and the trend feature data to perform priority sorting to generate evaluation data.

[0180] In the embodiments of the present invention, the state change trend data and the abnormal state diagnosis data may be collected at different time scales and time points. It is necessary to find a unified time reference to align the two sets of data according to the timestamps. For example, if the state change trend data is collected once per minute and the abnormal state diagnosis data is collected once every two minutes, it is necessary to interpolate the abnormal state diagnosis data to the time scale of every minute so that the two correspond to each other in time. Ensure that the two sets of data have the same data format. For example, the battery power in the state change trend data may be represented by a percentage, while the battery power in the abnormal state diagnosis data may be represented by ampere-hour (Ah), and they need to be converted to a unified format. Integrate the converted and aligned data together to generate a comprehensive state dataset.

[0181] For the battery power data in the comprehensive state dataset, methods such as polynomial fitting and moving average can be used to identify the change trend of the battery power over time. For example, through quadratic polynomial fitting, it can be obtained whether the battery power shows a linear increase, decrease, or a more complex change trend. Similarly, analyze the charge and discharge power data to judge whether the charge and discharge power is stable, gradually increasing, or gradually decreasing, etc. Calculate the change rate of the battery temperature data, that is, the temperature change slope. The slope can be calculated by dividing the temperature difference between adjacent time points by the time interval to understand the change speed of the battery temperature. Organize the identified trend feature information into trend feature data.

[0182] Compare the trend feature data with the abnormal state diagnosis data to find the features related to the abnormal state in the trend feature data. For example, if the abnormal state diagnosis data shows that there is a risk of overcharging the battery, check whether the upward trend of the battery power in the trend feature data is too fast, etc. According to the existing charge and discharge strategy rules, check whether there are situations of conflict with the strategy in the trend feature data and the abnormal state diagnosis data. For example, the current strategy requires reducing the charging power when the battery power reaches 80%, but the trend feature data shows that the battery power is rising rapidly and is about to exceed 80%, while the charging power shows no trend of decreasing, which constitutes a strategy conflict. Organize all the identified strategy conflict situations into a strategy conflict list.

[0183] Assign corresponding priorities to each conflict item in the strategy conflict list and each feature in the trend feature data. The assignment of priorities can be determined according to the degree of influence on battery safety, performance, and user needs. For example, the priority of the conflict item of battery overheating may be higher than that of the conflict item of slightly excessive battery power. Sort the strategy conflict list data and the trend feature data from high to low according to the priority, and use the sorting result as the evaluation data.

[0184] Preferably, please refer to Figure 5, the specific steps of step S5 are as follows:

[0185] Step S51: Align the timestamps and standardize the formats of the vehicle charge and discharge status data, status change trend data, abnormal status diagnosis data, and policy adjustment data to generate an integrated dataset;

[0186] Step S52: Extract the key risk features from the integrated dataset to generate real-time deviation, trend severity data, and abnormal event trigger data;

[0187] Step S53: Match the real-time deviation, trend severity data, and abnormal event trigger data with the risk thresholds in the preset risk threshold library respectively to generate risk feature vectors;

[0188] Step S54: Match the risk feature vectors with the risk level data in the preset classification rule library according to the principle of abnormal event priority to generate battery charge and discharge warning data.

[0189] In the embodiment of the present invention, the vehicle charge and discharge status data, status change trend data, abnormal status diagnosis data, and policy adjustment data may come from different data sources or acquisition systems, and their timestamps may be different. By aligning these data in chronological order, it is ensured that the data at the same time point can be accurately associated. For example, using the timestamp as an index, the data at the same time point in different datasets is merged. The data formats of different data sources may be different, such as data types, units, etc. The formats of these data are unified. For example, the representation of battery power is unified as a percentage, and the unit of voltage is unified as volts. After timestamp alignment and format standardization, all the data is integrated together to generate an integrated dataset.

[0190] Compare the real-time data in the integrated dataset with the reference value under normal conditions to calculate the real-time deviation. For example, calculate the percentage difference between the current battery voltage and the rated voltage, or the percentage difference between the current charging current and the recommended charging current. Analyze the status change trend data to evaluate the severity of the battery status change trend. For example, by calculating indicators such as the rate of battery power decrease and the slope of temperature rise, determine whether the trend is developing in a dangerous direction. According to the abnormal status diagnosis data, determine whether a preset abnormal event has been triggered. For example, when the battery temperature exceeds a certain threshold or the battery voltage shows abnormal fluctuations, it is determined that an abnormal event has been triggered. Organize the calculated real-time deviation, trend severity data, and abnormal event trigger data into key risk feature data.

[0191] A risk threshold library is established in advance, which contains thresholds corresponding to different risk characteristics. For example, the threshold of real-time deviation, the threshold of trend severity, etc. The real-time deviation, trend severity data and abnormal event trigger data are respectively compared with the risk thresholds in the preset risk threshold library. If a certain characteristic value exceeds the corresponding threshold, it is marked that the characteristic has a risk. The matching results of all characteristics are sorted into a risk characteristic vector, and each element in the vector represents whether a characteristic has a risk.

[0192] A preset classification rule library is established, which contains combinations of risk characteristics corresponding to different risk levels. For example, when multiple key risk characteristics have risks at the same time, it is determined as a high risk level; when only a few characteristics have risks, it is determined as a low risk level. According to the principle of abnormal event priority, the risk characteristic vector is matched with the risk level data in the preset classification rule library. The principle of abnormal event priority means that if there is an abnormal event trigger, the risk level corresponding to the abnormal event will be considered first. According to the matching result, the risk level of battery charging and discharging is determined, and battery charging and discharging warning data is generated. The warning data contains information such as the risk level and the triggered risk characteristics.

[0193] Preferably, the specific steps of step S54 are as follows:

[0194] Step S541: Judge whether the risk characteristic vector conforms to the abnormal event in the preset classification rule library. If so, execute step S542; if not, execute step S543;

[0195] Step S542: Match the risk level corresponding to the risk characteristic vector in the preset classification rule library to generate primary battery charging and discharging warning data;

[0196] Step S543: Judge whether the battery level value corresponding to the risk characteristic vector is less than the preset battery level threshold. If so, execute step S544; if not, execute step S545;

[0197] Step S544: Use the preset secondary warning data to construct secondary battery charging and discharging warning data;

[0198] Step S545: Use the preset tertiary warning data to construct tertiary battery charging and discharging warning data.

[0199] In an embodiment of the present invention, first, obtain the risk feature vector generated in step S53, which contains information on key risk features such as real-time deviation, trend severity data, and abnormal event trigger data. Then, compare the risk feature vector with the abnormal events defined in the preset classification rule library. The preset classification rule library details the risk feature combinations corresponding to various abnormal events. For example, abnormal events may include overcharging of the battery (voltage exceeds a certain threshold and the charging current is not zero), overheating of the battery (temperature is higher than the safe range), etc., and each abnormal event has its specific risk feature description. Make a judgment based on the comparison result. If the risk feature vector conforms to the feature combination of an abnormal event in the preset classification rule library, it is determined to conform to the abnormal event, and step S542 is executed; otherwise, it is determined not to conform to the abnormal event, and step S543 is executed.

[0200] When it is determined in step S541 that the risk feature vector conforms to the abnormal event in the preset classification rule library, search for the risk level corresponding to the risk feature vector in the preset classification rule library. The preset classification rule library not only defines abnormal events but also clearly divides the risk levels corresponding to each abnormal event. After finding the corresponding risk level, generate first-level battery charge and discharge warning data according to this level. The first-level battery charge and discharge warning data usually includes risk level information (first level), a specific description of the abnormal event triggering the warning (such as "the battery is overcharged, the voltage reaches 4.3V, exceeding the safety threshold of 4.2V"), and possible countermeasure suggestions (such as "immediately stop charging and check the battery status"), etc.

[0201] After it is determined in step S541 that the risk feature vector does not conform to the abnormal event in the preset classification rule library, extract the corresponding power value information in the risk feature vector. Compare the extracted power value with the preset power threshold. The preset power threshold is a power boundary preset according to factors such as the capacity of the battery and the driving requirements of the vehicle. For example, the preset power threshold may be set to 20% of the battery capacity. Make a judgment based on the comparison result. If the power value is less than the preset power threshold, it is determined that the power is insufficient, and step S544 is executed; otherwise, it is determined that the power is sufficient, and step S545 is executed.

[0202] When it is determined in step S543 that the power value corresponding to the risk feature vector is less than the preset power threshold, use the preset secondary warning data to construct secondary battery charge and discharge warning data. The preset secondary warning data contains relevant information for the situation of insufficient power, such as risk level information (second level), a specific description of insufficient power (such as "the battery power remains only 15%, lower than the preset power threshold of 20%"), and corresponding suggestions (such as "quickly find a charging facility to charge"), etc. Organize and combine these information to generate complete secondary battery charge and discharge warning data.

[0203] When it is determined in step S543 that the power value corresponding to the risk feature vector is not less than the preset power threshold, the preset three-level warning data is used to construct the three-level battery charge and discharge warning data. The preset three-level warning data indicates that the battery is in a relatively safe state and the risk level is low. The three-level warning data may include content such as risk level information (level three), a brief description of the current state of the battery (such as "the battery power is sufficient, and the current power is 60%"), and general tips (such as "continue to use normally and pay attention to the battery status"). These information are sorted and combined to generate the complete three-level battery charge and discharge warning data.

[0204] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for managing charging and discharging of an electric vehicle battery, characterized in that, Including the following steps: Step S1: Obtain the sensor data set collected by each sensor during the charging of the electric vehicle, perform parameter fusion on the sensor data set, and generate real-time battery state data; Step S2: Use the real-time battery state data, the corresponding environmental data and vehicle operating condition data of the real-time battery state data to construct vehicle charge and discharge state data; Step S3: Based on the vehicle charge and discharge state data, identify the battery state change trend and abnormal state, and generate state change trend data and abnormal state diagnosis data; Step S4: Dynamically adjust the battery charge and discharge strategy according to the vehicle charge and discharge state data and the state change trend data, and generate strategy adjustment data; Step S5: Perform hierarchical warning according to the vehicle charge and discharge state data, the state change trend data, the abnormal state diagnosis data and the strategy adjustment data, and generate battery charge and discharge warning data.

2. The method for managing charging and discharging of an electric vehicle battery according to claim 1, wherein The specific steps of step S1 are as follows: Step S11: Clean the sensor data set to generate a first sensor data set; Step S12: Normalize the first sensor data set to generate a second sensor data set; Step S13: Extract features from the second sensor data set to generate multiple sensor features; Step S14: Perform feature fusion on the sensor features to generate fusion data; Step S15: Calculate the mean absolute error between the fusion data and the battery state data corresponding to the sensor data set to generate error data; Step S16: When the error data is less than the preset error threshold, use the fusion data as the real-time battery state data.

3. The method for managing charging and discharging of an electric vehicle battery according to claim 1, wherein The specific steps of step S3 are as follows: Step S31: Use a preset state prediction model to predict the battery state change trend of the vehicle charge and discharge state data by using a predictive control algorithm, generate first state change prediction data, and extract feature data; Step S32: Use a deep learning algorithm to predict the battery state change trend of the vehicle charge and discharge state data and the feature data, and generate second state change prediction data; Step S33: Perform time dimension cross-validation according to the first state change prediction data and the second state change prediction data to determine the state change trend data; Step S34: Judge the abnormal state of the vehicle charge and discharge state data and the preset hierarchical warning threshold to generate first abnormal state data; Step S35: Use an anomaly detection algorithm to deeply analyze the first abnormal state data to generate second abnormal state data; Step S36: Use an intelligent dynamic adjustment strategy technology to perform multi-source data fusion assisted diagnosis on the first abnormal state data, the second abnormal state data and the environmental data corresponding to the vehicle charge and discharge state data, and generate abnormal state diagnosis data.

4. A method for managing battery charging and discharging of an electric vehicle according to claim 3, characterized in that, The preset state prediction model includes an electrochemical model, an equivalent circuit model and an empirical model; the specific steps of step S31 are as follows: Step S311: Use an adaptive algorithm to dynamically correct the chemical reaction rate and equivalent circuit parameters in the preset state prediction model with the vehicle charge and discharge state data, and generate a target state prediction model; Step S312: Input the vehicle charge and discharge state data into the target state prediction model to predict the battery state change trend, and generate initial prediction data; Step S313: Use a traditional predictive control algorithm to formulate a charge and discharge strategy based on the initial prediction data and the vehicle charge and discharge state data, and generate target prediction data; Step S314: Use a reinforcement learning algorithm to optimize the target prediction data with the goal of maximizing battery life and minimizing energy consumption, and generate first state change prediction data; Step S315: Perform deep feature extraction on the vehicle charge and discharge state data to generate feature data.

5. A method for managing charging and discharging of an electric vehicle battery according to claim 3, characterized in that, The specific steps of step S32 are as follows: Step S321: Use an isolation forest algorithm to clean the vehicle charge and discharge state data to generate initial state data; Step S322: Normalize the initial state data using a time series sliding window to generate target state data; Step S323: Fuse the target state data and the feature data to generate state feature fusion data; Step S324: Input the state feature fusion data into a preset trend prediction model to predict the battery state change trend, and generate initial prediction data; Step S325: Use a smoothing filter algorithm to remove noise and fluctuations in the initial prediction data to generate target prediction data; Step S326: Calculate the confidence interval of the target prediction data to generate confidence interval data; Step S327: Use the target prediction data and the confidence interval data to construct second state change prediction data.

6. The method for managing charging and discharging of an electric vehicle battery according to claim 3, wherein, The specific steps of step S33 are as follows: Step S331: Align the first state change prediction data and the second state change prediction data according to the timestamp and divide them into time windows to generate a first time window and a second time window; Step S332: Calculate the prediction error and prediction accuracy of the first time window and the second time window respectively to generate a first prediction error, a second prediction error, a first prediction accuracy, and a second prediction accuracy; Step S333: Evaluate the error consistency of the first prediction error and the second prediction error to generate error evaluation data; Step S334: Evaluate the accuracy consistency of the first prediction accuracy and the second prediction accuracy to generate accuracy evaluation data; Step S335: When both the error evaluation data and the accuracy evaluation data are within the preset evaluation threshold, use the first state change prediction data and the second state change prediction data to construct state change trend data.

7. A method for managing charging and discharging of an electric vehicle battery according to claim 1, characterized in that, The specific steps of step S4 are as follows: Step S41: Evaluate the current charge and discharge strategy of the electric vehicle using the state change trend data and the abnormal state diagnosis data to generate evaluation data; Step S42: When the evaluation data indicates that the battery power is insufficient to meet the driving requirements, increase the charging time or charging power of the electric vehicle to generate first policy adjustment data; Step S43: When the evaluation data indicates that the battery power is excessive, reduce the charging power of the electric vehicle or stop charging in advance to generate second policy adjustment data; Step S44: When the evaluation data is during the low grid load period, increase the charging power of the electric vehicle to generate third policy adjustment data; Step S45: When the evaluation data is during the high grid load period, reduce the charging power of the electric vehicle or discharge to generate fourth policy adjustment data; Step S46: When the battery temperature in the evaluation data is greater than the preset temperature threshold, reduce the charging power of the electric vehicle or pause charging and start the cooling device to generate fifth policy adjustment data.

8. A method for managing charging and discharging of an electric vehicle battery according to claim 7, characterized in that, The specific steps of step S41 are as follows: S411: Synchronize the time and unify the format of the state change trend data and the abnormal state diagnosis data to generate a comprehensive state data set; S412: Use the comprehensive state data set to identify the battery power trend, charge and discharge power trend, and battery temperature change slope to generate trend feature data; S413: Use the trend feature data and the abnormal state diagnosis data to perform abnormal state matching and policy conflict identification to generate a policy conflict list; S414: Use the policy conflict list data and the trend feature data to perform priority sorting to generate evaluation data.

9. A method for managing charging and discharging of an electric vehicle battery according to claim 1, characterized in that The specific steps of step S5 are as follows: Step S51: Align the timestamps and standardize the formats of the vehicle charge and discharge state data, the state change trend data, the abnormal state diagnosis data, and the policy adjustment data to generate an integrated data set; Step S52: Extract the key risk features in the integrated data set to generate real-time deviation, trend severity data, and abnormal event trigger data; Step S53: Respectively match the real-time deviation, the trend severity data, and the abnormal event trigger data with the risk thresholds in the preset risk threshold library to generate a risk feature vector; Step S54: Match the risk feature vector with the risk level data in the preset classification rule library according to the principle of abnormal event priority to generate battery charge and discharge warning data.

10. A method for managing battery charging and discharging of an electric vehicle according to claim 9, characterized in that, The specific steps of step S54 are as follows: Step S541: Judge whether the risk feature vector conforms to the abnormal event in the preset classification rule library. If so, execute step S542; if not, execute step S543; Step S542: Match the corresponding risk level of the risk feature vector in the preset classification rule library to generate first-level battery charge and discharge warning data; Step S543: Judge whether the battery power value corresponding to the risk feature vector is less than the preset battery power threshold. If so, execute step S544; if not, execute step S545; Step S544: Use the preset second-level warning data to construct second-level battery charge and discharge warning data; Step S545: Use the preset third-level warning data to construct third-level battery charge and discharge warning data.

Images