A method for predicting the state of charge and the remaining range of an electric vehicle

Abstract

The application provides an electric vehicle power state and remaining mileage prediction method, which comprises the following steps: dividing the battery pack into regions and deploying sensors, collecting real-time power data of each region, constructing an anomaly detection model to identify abnormal regions and adjust local data, combining user driving behavior and environmental parameters, using a long short-term memory network to dynamically predict power consumption trends, calculating the remaining mileage according to real-time road conditions and temperature changes, and introducing an adaptive correction mechanism to dynamically adjust model parameters by comparing and analyzing historical prediction records, thereby improving prediction accuracy. The application can accurately monitor the power state of an electric vehicle and dynamically predict the remaining mileage, effectively solving the problem of large prediction deviation in complex driving environments, providing more reliable power and mileage information for users, and improving the use experience and safety of electric vehicles.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of electric vehicle power state and remaining range prediction method. BACKGROUND

[0002] New energy vehicle field as the important pillar of modern transportation and energy transformation, it has irreplaceable value to promote green travel and reduce carbon emissions, and its core technology directly affects vehicle performance and user experience.However, the current solution in power monitoring and prediction still has obvious deficiencies, many methods often have high precision in actual use and lead to large deviation between power display and actual situation, which further affects the user's judgment of remaining range, reduces the reliability and convenience of driving.

[0003] Under this background, the accuracy of power monitoring becomes a key problem to be solved.The first challenge is that when battery pack as a whole, the collection of power data often lacks detailed management, it is difficult to capture the abnormal changes in local area, which leads to errors in overall power estimation.The limitation of this overall monitoring further leads to another problem, that is, it cannot be adjusted individually according to the use habits and driving environment of different users, the difference between theoretical power consumption and actual consumption cannot be effectively corrected, which further makes the remaining range prediction often inaccurate.These two problems are interrelated, the former leads to inaccurate data basis, and the latter exacerbates the error due to lack of individual correction mechanism. SUMMARY

[0004] The present application proposes a kind of electric vehicle power state and remaining range prediction method, by refining battery pack management and combining user actual use scene to build individualized power correction mechanism, provide more reliable power and range information for user.

[0005] The technical solution of the present application is realized as follows:

[0006] A kind of electric vehicle power state and remaining range prediction method, the method comprises:

[0007] S1, by dividing battery pack into sub-regions, a multi-point power data acquisition framework is constructed, and the power state data of each region is collected in real time in each region to form a sub-regional power raw data set;The voltage and current data of each region are analyzed one by one, if the data of a certain region exceeds the preset threshold range, it is marked as an abnormal region, and abnormal region distribution information is generated;According to the abnormal region distribution information, a local power correction matrix is constructed, the power data of the abnormal region is adjusted by weighting, and the power estimation value of the overall battery pack is recalculated in combination with the data of non-abnormal region, to obtain the corrected overall power state data;

[0008] S2, combine the real-time acquired user historical driving records and driving environment parameters, including acceleration frequency, average vehicle speed and road slope, to construct a user behavior feature vector; according to the user behavior feature vector and the corrected overall power state data, a long short-term memory network model is used to dynamically predict the power consumption trend, and power consumption rate prediction data in a specific scene is generated;

[0009] S3, combined with the current driving environment adaptation parameters, including real-time road conditions and temperature changes, the remaining mileage prediction data is calculated through a preset mileage conversion formula: remaining mileage = (corrected overall power state data / power consumption rate prediction data)*environmental adjustment coefficient, wherein the environmental adjustment coefficient is obtained from a preset table based on temperature and road conditions, and finally the preliminary remaining mileage prediction data is obtained;

[0010] S4, for the preliminary remaining mileage prediction data, compared with the recent historical prediction records stored in the system, if the prediction deviation exceeds the preset threshold range, the adaptive correction mechanism is triggered, the parameters of the long short-term memory network model are adjusted, and the updated model parameters are generated; combined with the latest regional power original data set and user behavior feature vector, the power consumption rate is re-predicted, and the optimized power consumption rate prediction data is generated;

[0011] S5, for the optimized power consumption rate prediction data, continuously update the corrected overall power state data and driving environment adaptation parameters, through the loop iteration calculation of setting the maximum iteration number or convergence condition, dynamically adjust the remaining mileage prediction data, and finally determine the real-time displayed power state and mileage prediction data.

[0012] The beneficial effects of the present application are: by dividing the battery pack into regions and deploying sensors, real-time acquisition of regional power data, construction of an anomaly detection model to identify abnormal regions and local data adjustment, combined with user driving behavior and environmental parameters, using long short-term memory network to dynamically predict power consumption trend, and calculating the remaining mileage according to real-time road conditions and temperature changes, while introducing an adaptive correction mechanism, through comparison and analysis with historical prediction records, dynamically adjusting the model parameters, improving the prediction accuracy. The present application can realize accurate monitoring of the power state of electric vehicles and dynamic prediction of the remaining mileage, effectively solving the problem of large prediction deviation of traditional methods in complex driving environment, providing more reliable power and mileage information for users, and improving the use experience and safety of electric vehicles. DETAILED DESCRIPTION

[0013] The technical solutions of the present application will be described clearly and completely below in connection with the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0014] A method for predicting the state of charge and the remaining range of an electric vehicle, specifically comprising:

[0015] In step S101, a multi-point power data acquisition framework is constructed by dividing the battery pack into regions, independent voltage sensors and current sensors are deployed in each region to collect real-time power state data of each region, forming a regional power raw data set, laying a data foundation for subsequent anomaly detection and overall power estimation.

[0016] According to the business attributes of regional division, the battery pack is managed by region, voltage sensors and current sensors are deployed in each region, and the power state of each region is monitored in real time. Voltage and current values are obtained from the sensors to get the partitioned initial data stream. Through the business attributes of data acquisition, the partitioned initial data stream is transmitted to the pre-established data storage unit, and the voltage and current values from different regions are integrated and processed in a data fusion manner to determine the integrity of the power state of each region and obtain a unified comprehensive data set. If the power state data of a certain region in the comprehensive data set exceeds the preset threshold range, the logic judgment module is called according to the business attributes of anomaly detection to compare and analyze the voltage and current values of the region to determine whether there is a potential anomaly. According to the business attributes of overall power, the power state data of all regions is extracted from the comprehensive data set, and the data is comprehensively processed by weighted average calculation according to the business attributes of state evaluation to obtain the overall power estimation value of the battery pack.

[0017] In step S102, for the regional power raw data set, a pre-established anomaly detection model is applied to analyze the voltage and current data of each region one by one. If the data of a certain region exceeds the preset threshold range, it is marked as an abnormal region, and abnormal region distribution information is generated for subsequent local data adjustment.

[0018] According to the pre-established abnormality detection rule, the voltage and current data of each region are obtained from the sub-regional power original data set, and the data points are compared one by one. If the voltage or current of a region exceeds the preset threshold range, it is marked as an abnormal region, and the preliminary abnormal region distribution information is obtained. Through secondary verification of the preliminary abnormal region distribution information, the adjacent region data of each abnormal region is obtained, and the mean comparison method is used to judge whether the voltage and current of the adjacent region deviate. If the adjacent region data also exceeds the preset threshold range, it is determined that the abnormal region marker is valid, and the confirmed abnormal region list is obtained. According to the confirmed abnormal region list, the time series data of the abnormal region is extracted from the original data set, the fluctuation amplitude of the time series data is calculated, and the standard deviation calculation tool is used to judge whether the fluctuation continuously exceeds the preset threshold range, and the fluctuation characteristic distribution of the abnormal region is obtained. Through classification processing of the fluctuation characteristic distribution of the abnormal region, the fluctuation characteristic category is obtained, and the data smoothing tool is used for local adjustment according to the category. It is determined whether the adjusted data point returns to the preset threshold range, and the final local data adjustment scheme is generated.

[0019] Step S103, according to the abnormal region distribution information, construct a local power correction matrix, adjust the power data of the abnormal region, combine the data of the non-abnormal region, and recalculate the power estimation value of the overall battery pack to obtain the corrected overall power state data, which provides an accurate basis for subsequent consumption prediction.

[0020] According to the abnormal region and distribution information, a preliminary mapping table for identifying power abnormalities is constructed. The power data of each region is obtained from the battery state, and the preset threshold is compared according to the division of the abnormal region and the non-abnormal region to obtain the determination result of the specific position and influence range of the abnormal region. Through the determination result of the abnormal region, a correction matrix of local power is constructed, the power data of the abnormal region is adjusted by weighting, the adjusted local power value is obtained, and it is integrated with the power data of the non-abnormal region to determine a preliminary overall power distribution table. The overall power distribution table is used in combination with the running parameters of the battery state to perform secondary calibration on the overall power. If the calibrated power estimation value does not match the preset threshold range, the correction matrix is fine-tuned to obtain the final overall power state data. According to the overall power state data, a basic data set for consumption prediction is generated, key indicators are extracted for feature extraction, input parameters required for prediction are obtained, and it is determined whether the accuracy requirements are met to determine the final power estimation value.

[0021] Step S104, for the corrected overall power state data, combine the real-time acquired user historical driving records and driving environment parameters, including acceleration frequency, average speed and road slope, to construct a user behavior feature vector for subsequent personalized power consumption trend analysis.

[0022] The acceleration frequency, average speed and brake frequency data are obtained from the user historical driving records, and the road slope and environmental temperature data are extracted from the driving environment parameters. The data is denoised by a data cleaning tool to obtain the sorted driving behavior data set. According to the driving behavior data set, the acceleration frequency and the average speed, the brake frequency and the road slope are associated and mapped by a data fusion tool to determine the driving habit characteristic value of the user in different environments. Through the pre-established calculation rule, the driving habit characteristic value and the environmental temperature data are obtained. If the characteristic value exceeds the preset threshold value, the power state data is dynamically adjusted to determine the adjusted power consumption reference value.

[0023]

[0024] DH represents the driving habit characteristic value, T represents the total number of sampling time periods, represents the acceleration at time t, represents the brake force at time t, represents the steering angle at time t, and the formula calculates the comprehensive characteristic value of driving behavior.

[0025]

[0026] ET represents the environmental temperature influence factor, represents the environmental temperature data, and a, β and γ are preset temperature influence coefficients. The formula describes the influence relationship of environmental temperature on power consumption;

[0027]

[0028] δ represents the driving habit threshold value judgment result, DH represents the driving habit characteristic value, represents the preset driving habit threshold value. When the characteristic value exceeds the threshold value, power adjustment is needed;

[0029]

[0030] represents the adjusted power consumption reference value, represents the basic power consumption value, represents the flag whether adjustment is needed, represents the driving habit adjustment coefficient, DH represents the driving habit characteristic value, The temperature influence adjustment coefficient is represented by ET, and the environmental temperature influence factor is represented by ET. The formula realizes dynamic adjustment of the power consumption according to driving habits and environmental temperature.

[0031] For the power consumption reference value, time series tools are used to perform correlation analysis on the reference value and the driving mileage data to obtain user personalized power consumption trend characteristic values.

[0032] In step S105, according to the user behavior feature vector and the corrected overall power state data, a long short-term memory network model is used to dynamically predict the power consumption trend, and power consumption rate prediction data in a specific scene is generated, providing core parameters for subsequent remaining mileage calculation.

[0033] According to the user behavior feature vector and the corrected overall power state data, standardized processing is performed through a pre-established data processing module to obtain the unified feature data set, and during processing, a preset threshold is used to filter abnormal values to obtain the cleaned basic data set. The cleaned basic data set is used to segment the power consumption history record through a time series decomposition tool to obtain the consumption mode characteristics in different time periods. If the consumption fluctuation in a certain time period exceeds the preset threshold, the data segment is smoothed to determine the intermediate feature set. Through the intermediate feature set, the power consumption trend is dynamically modeled to obtain the consumption rate prediction value in the specific scene, and during processing, the characteristics weight of different scenes is adjusted to determine the prediction rate data of the matching user behavior. According to the prediction rate data, the prediction value is recalibrated through a scene adaptation tool combined with the user's current behavior environment information to obtain the final power consumption rate prediction result.

[0034] In step S106, for the power consumption rate prediction data, combined with the current driving environment adaptation parameters including real-time road conditions and temperature changes, the remaining mileage prediction data is calculated through a preset mileage conversion formula, which is: remaining mileage equals corrected overall power state data divided by power consumption rate prediction data multiplied by environmental adjustment coefficient, wherein the environmental adjustment coefficient is obtained from a preset table based on temperature and road conditions, the remaining mileage represents the distance that the vehicle can continue to travel, the corrected overall power state data represents the current available power of the battery pack, the power consumption rate prediction data represents the speed of power consumption per unit time, and the environmental adjustment coefficient represents the environmental influence factor on power consumption. Finally, the preliminary remaining mileage prediction data is obtained.

[0035] According to real-time traffic information and temperature change data, a record matching the traffic and the temperature is queried from a preset environment table to determine an environment adjustment coefficient value. By obtaining corrected state of charge data, combining the actual state of the battery pack power, using an internal calibration tool to check the current available power, an accurate corrected state of charge value is obtained. If the corrected state of charge value and the power consumption rate prediction data are both obtained, a mileage conversion formula is used for calculation, i.e. the remaining mileage is equal to the corrected state of charge value divided by the power consumption rate prediction data multiplied by the environment adjustment coefficient, to obtain a preliminary remaining mileage prediction result. According to the preliminary remaining mileage prediction result, combining the driving distance estimation logic, using a data comparison tool to perform secondary verification on the result, judging whether it conforms to the reasonable range of unit time consumption, determining the final remaining mileage prediction data.

[0036] In step S107, the preliminary remaining mileage prediction data is compared and analyzed with the recent historical prediction records stored in the system. If the prediction deviation exceeds the preset threshold range, an adaptive correction mechanism is triggered to adjust the parameters of the long short-term memory network model to generate updated model parameters for improving the accuracy of subsequent prediction.

[0037] The remaining mileage prediction data of the recent records is obtained from the storage library, and the recent records and the currently generated prediction data are compared item by item. By calculating the difference value of each group of data, the specific value of the prediction deviation is obtained. If the prediction deviation exceeds the preset threshold range, an adaptive correction mechanism is triggered to determine the parameter direction and amplitude that need to be adjusted by comparing the deviation trend in the historical records. According to the determined parameter direction and amplitude, the pre-established adjustment rule is used to update the parameters of the long short-term memory network to generate the updated parameter configuration.

[0038]

[0039] wherein represents the currently generated prediction mileage data, represents the actual mileage data of the recent records obtained from the storage library, represents the difference value between the two, i.e. the specific value of the prediction deviation.

[0040]

[0041] wherein represents the preset threshold range, and n represents the number of historical records, represents the prediction deviation value in the historical records, represents the standard deviation of the deviation, represents the adjustment coefficient;

[0042]

[0043] wherein represents the direction of parameter adjustment, m represents the number of records in the recent period of time, represents the prediction deviation;

[0044]

[0045] wherein represents the final adjustment amplitude, represents the current prediction deviation, represents the average value of historical deviation, and respectively represent the maximum and minimum values of historical deviation, represents the adjustment coefficient;

[0046]

[0047] wherein represents the updated parameter, represents the original parameter, represents the direction of parameter adjustment, represents the adjustment amplitude, represents the gradient of the parameter. Through this formula, the system can update the parameters of the long short-term memory network according to the determined parameter direction and amplitude, to generate an updated parameter configuration.

[0048] By comparing the average deviation of the last m records with the average deviation of the previous m records, the trend of the deviation is determined, so as to determine the direction of parameter adjustment. When the current prediction deviation exceeds this range, the adaptive correction mechanism will be triggered. By applying the updated parameter configuration to the prediction process, new remaining mileage prediction data is obtained, and compared with the historical records again to determine whether the prediction deviation has been reduced to within the threshold range.

[0049] Step S108, according to the updated model parameters, combine the latest regional power original data set and user behavior feature vector to re-predict the power consumption rate, generate optimized power consumption rate prediction data, and provide more accurate parameter support for subsequent remaining mileage prediction.

[0050] Based on the latest regional raw power consumption dataset, historical power consumption records for each region are obtained. Combined with user behavior feature vectors, the raw data is denoised using a data cleaning tool to obtain a refined basic power consumption dataset. For this refined dataset, a pre-established user behavior influence weight table is used to obtain the influence coefficient of each behavioral feature on power consumption. A weighted calculation tool is then used to determine the preliminary predicted power consumption rate for each region.

[0051]

[0052] in This represents the preliminary predicted rate of power consumption in region i. This represents the weighting coefficient of the influence of behavioral feature j. represents the value of behavioral feature j in region i, and n represents the total number of behavioral features.

[0053] A preliminary prediction value is obtained by weighted summation of various behavioral features. If the deviation between the preliminary prediction value of the power consumption rate and the historical true value exceeds a preset threshold, the preliminary prediction value is corrected using regression analysis tools to obtain an adjusted power consumption rate prediction dataset.

[0054]

[0055] in This represents the percentage of prediction deviation for region i. This represents the preliminary predicted rate of power consumption in region i. Represents the historical actual consumption rate of region i, when Correction is required when the threshold is exceeded;

[0056]

[0057] Where CR_i' represents the predicted power consumption rate for region i after correction. HR_i represents the preliminary forecast value, and HR_i represents the historical actual value. This indicates that the correction coefficient ranges from 0 to 1 and is used to control the degree of correction.

[0058] when If the power consumption rate exceeds a preset threshold, correction is required. Based on the adjusted power consumption rate prediction dataset and combined with regional geographical environment data, the prediction dataset is further optimized using a data fusion tool to determine the final power consumption rate prediction result.

[0059]

[0060] in a final power consumption rate prediction result of the region i, a corrected prediction value, an influence value of the kth geographical environment factor of the region i, a weight coefficient of the kth geographical environment factor, and m represents the total number of geographical environment factors. The prediction result is secondarily optimized by fusing geographical environment data.

[0061] In step S109, the corrected overall power state data and the driving environment adaptation parameter are continuously updated for the optimized power consumption rate prediction data, and the remaining range prediction data is dynamically adjusted through loop iteration calculation by setting the maximum iteration number or the convergence condition, so as to finally determine the power state and the range prediction data displayed in real time.

[0062] According to the power consumption data and the driving environment data, corresponding adaptation parameters are obtained from a pre-established storage library, it is judged whether the adaptation parameters are within a preset threshold range by comparing the current power consumption rate with historical records, if the threshold range is exceeded, the adaptation parameters are corrected to obtain adjusted parameter values. The adjusted parameter values are used in combination with the power state data to determine a preliminary range prediction result through loop iteration calculation limited by the maximum iteration number and the convergence condition, and the remaining range prediction data is dynamically adjusted at each iteration. By comparing the preliminary range prediction result with the real-time displayed power state data, if the deviation exceeds a preset threshold, the latest driving environment data and power consumption data are reacquired, the adaptation parameters are updated, and a corrected range prediction value is obtained. According to the corrected range prediction value and the power state data, the final display content is integrated through data fusion in combination with real-time display requirements, it is judged whether the accuracy requirement is met, if not, the loop iteration calculation link is returned to obtain more accurate display data.

[0063] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of predicting the state of charge and the remaining distance of an electric vehicle, characterized by, The method comprises: S1, by dividing the battery pack into sub-regions, constructing a multi-point power data acquisition framework, collecting the power state data of each region in real time in each region to form a sub-regional power original data set; analyzing the voltage and current data of each region one by one, if the data of a certain region exceeds the preset threshold range, it is marked as an abnormal region, and the abnormal region distribution information is generated; according to the abnormal region distribution information, a local power correction matrix is constructed, the power data of the abnormal region is adjusted by weighting, and the power estimation value of the whole battery pack is recalculated combined with the data of the non-abnormal region, to obtain the corrected overall power state data; S2, combining the real-time obtained user historical driving record and driving environment parameters, a user behavior feature vector is constructed; according to the user behavior feature vector and the corrected overall power state data, a long short-term memory network model is used to dynamically predict the power consumption trend, and power consumption rate prediction data in a specific scene is generated; S3, combined with the current driving environment adaptation parameters, including real-time road conditions and temperature changes, the remaining mileage prediction data is calculated through a preset mileage conversion formula: remaining mileage=(corrected overall power state data / power consumption rate prediction data)*environmental adjustment coefficient, wherein the environmental adjustment coefficient is obtained from a preset table based on temperature and road conditions, and the preliminary remaining mileage prediction data is finally obtained; S4, for the preliminary remaining mileage prediction data, compared with the recent historical prediction record stored in the system, if the prediction deviation exceeds the preset threshold range, the adaptive correction mechanism is triggered, the parameters of the long short-term memory network model are adjusted, and the updated model parameters are generated; combined with the latest sub-regional power original data set and user behavior feature vector, the power consumption rate is predicted again, and the optimized power consumption rate prediction data is generated; S5, for the optimized power consumption rate prediction data, continuously update the corrected overall power state data and driving environment adaptation parameters, and dynamically adjust the remaining mileage prediction data through the loop iteration calculation of setting the maximum iteration number or convergence condition, and finally determine the real-time displayed power state and mileage prediction data.

2. The method of claim 1, wherein, In the step S1, by dividing the battery pack into sub-regions, constructing a multi-point power data acquisition framework, collecting the power state data of each region in real time in each region to form a sub-regional power original data set, comprising: According to the business attribute of regional division, the battery pack is managed by partition, and voltage sensors and current sensors are deployed in each region. The power state of each region is monitored in real time, the voltage value and current value are obtained from the sensors, and the partitioned initial data stream is obtained; Through the business attribute of data acquisition, the partitioned initial data stream is transmitted to the pre-established data storage unit, the voltage values and current values from different regions are integrated and processed in a data fusion manner, the integrity of the power state of each region is determined, and a unified comprehensive data set is obtained; If the power state data of a certain region in the comprehensive data set exceeds the preset threshold range, the logic judgment module is called for the service attribute of the anomaly detection, and the voltage and current values of the region are compared and analyzed to determine whether there is a potential anomaly. According to the service attribute of the overall power, the power state data of all regions is extracted from the comprehensive data set, and the data is comprehensively processed by a weighted average calculation method to obtain the overall power estimation value of the battery pack.

3. The method of claim 2, wherein, In step S1, the voltage and current data of each region are analyzed one by one, and if the data of a certain region exceeds the preset threshold range, it is marked as an abnormal region, and abnormal region distribution information is generated, including: According to the pre-established anomaly detection rule, the voltage and current data of each region are obtained from the regional power raw data set, and the data points are compared one by one, and if the voltage or current of a certain region exceeds the preset threshold range, it is marked as an abnormal region, and the preliminary abnormal region distribution information is obtained; By performing secondary verification on the preliminary abnormal region distribution information, the adjacent region data of each abnormal region is obtained, and the mean comparison method is used to determine whether the voltage and current of the adjacent region deviate, and if the adjacent region data also exceeds the preset threshold range, it is determined that the abnormal region marker is valid, and the confirmed abnormal region list is obtained; According to the confirmed abnormal region list, the time series data of the abnormal region is extracted from the raw data set, the fluctuation amplitude is calculated for the time series data, and the standard deviation calculation tool is used to determine whether the fluctuation continuously exceeds the preset threshold range, and the fluctuation characteristic distribution of the abnormal region is obtained; By classifying the fluctuation characteristic distribution of the abnormal region, the fluctuation characteristic category is obtained, and the data smoothing tool is used for local adjustment for the category to determine whether the adjusted data point returns to the preset threshold range, and the final local data adjustment scheme is generated.

4. The method of claim 3, wherein, In step S1, according to the abnormal region distribution information, a local power correction matrix is constructed, the power data of the abnormal region is weighted and adjusted, and the power estimation value of the overall battery pack is recalculated in combination with the data of the non-abnormal region, to obtain the corrected overall power state data, including: According to the abnormal region and distribution information, a preliminary mapping table for identifying power anomalies is constructed, the power data of each region is obtained from the battery state, and the preset threshold is compared for the division of the abnormal region and the non-abnormal region to obtain the determination result of the specific position and influence range of the abnormal region; According to the determination result of the abnormal region, a local power correction matrix is constructed, the power data of the abnormal region is weighted and adjusted, the adjusted local power value is obtained, and it is integrated with the power data of the non-abnormal region to determine a preliminary overall power distribution table; The overall power is secondarily calibrated by using the overall power distribution table and combining the running parameters of the battery configuration, and if the calibrated power estimation value does not match the preset threshold range, the correction matrix is fine-tuned to obtain final overall power state data; According to the overall power state data, a basic data set for consumption prediction is generated, feature extraction is performed for key indicators, input parameters required for prediction are obtained, whether the accuracy requirement is met is judged, and a final power estimation value is determined.

5. The method of claim 1, wherein, In the step S2, the user behavior feature vector is constructed by combining the real-time acquired user historical driving records and driving environment parameters, including: From the user historical driving records, acceleration frequency, average speed and braking frequency data are obtained, and from the driving environment parameters, road slope and environmental temperature data are extracted. The data is denoised by a data cleaning tool to obtain a sorted driving behavior data set. The acceleration frequency is associated and mapped with the average speed and the braking frequency with the road slope by using a data fusion tool to determine the driving habit characteristic value of the user in different environments, ， DH represents the driving habit characteristic value, T represents the total number of sampling time periods, represents the acceleration at time t, represents the braking force at time t, represents the steering angle at time t, and the formula calculates the comprehensive characteristic value of the driving behavior; Through the pre-established calculation rule, the driving habit characteristic value and the environmental temperature data are obtained. If the characteristic value exceeds the preset threshold, the power state data is dynamically adjusted, and the adjusted power consumption reference value is determined, ， ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an environmental temperature influence factor, ET represents an ， δ represents the driving habit threshold value judgment result, DH represents the driving habit characteristic value, represents the preset driving habit threshold value, when the characteristic value exceeds the threshold value, the power adjustment needs to be performed; ， represents the adjusted power consumption reference value, represents the basic power consumption value, represents a flag indicating whether adjustment is needed, represents a driving habit adjustment coefficient, DH represents a driving habit characteristic value, represents a temperature influence adjustment coefficient, ET represents an environmental temperature influence factor, and the formula realizes dynamic adjustment of power consumption according to driving habits and environmental temperature; For the power consumption reference value, the reference value and the driving mileage data are associated and analyzed by using a time series tool to obtain a user personalized power consumption trend characteristic value.

6. The method of claim 5, wherein, In the step S2, the long short-term memory network model is used to dynamically predict the power consumption trend according to the user behavior feature vector and the corrected overall power state data, to generate power consumption rate prediction data in a specific scenario, including: According to the user behavior feature vector and the corrected overall power state data, the data is standardized by using a pre-established data processing module to obtain the unified feature data set, and during processing, the abnormal values are filtered by using a preset threshold to obtain the cleaned basic data set; The cleaned basic data set is used to segment the power consumption history record by using a time series decomposition tool to obtain the consumption mode characteristics in different time periods. If the consumption fluctuation of a certain time period exceeds the preset threshold, the data segment is smoothed to determine the intermediate feature set; Through the intermediate feature set, the power consumption trend is dynamically modeled to obtain the consumption rate prediction value in the specific scenario, and during processing, the characteristics weight of different scenarios is adjusted to determine the prediction rate data matching the user behavior; According to the prediction rate data, the prediction value is secondarily calibrated by using a scenario adaptation tool in combination with the user's current behavior environment information to obtain the final power consumption rate prediction result.

7. The method of claim 1, wherein, The step S3 specifically includes: According to real-time road condition information and temperature change data, records matching the road condition and the temperature are queried from a preset environment table to determine an environment adjustment coefficient value; By acquiring the corrected electric quantity state data, combining the actual condition of the electric quantity of the battery pack, and adopting an internal calibration tool to check the current available electric quantity, an accurate corrected electric quantity state value is obtained; If the corrected electric quantity state value and the electric quantity consumption rate prediction data are both acquired, a mileage conversion formula is used for calculation, that is, remaining mileage=(corrected overall electric quantity state data / electric quantity consumption rate prediction data)*environmental adjustment coefficient, wherein the environmental adjustment coefficient is obtained from a preset table based on temperature and road conditions, the remaining mileage represents the distance that the vehicle can continue to travel, the corrected overall electric quantity state data represents the current available electric quantity of the battery pack, the electric quantity consumption rate prediction data represents the speed of electric quantity consumption per unit time, and the environmental adjustment coefficient represents an environmental influence factor on electric quantity consumption, to obtain a preliminary remaining mileage prediction result; According to the preliminary remaining mileage prediction result, combined with driving distance estimation logic, a data comparison tool is used to perform secondary checking on the result to determine whether it meets a reasonable range of consumption per unit time, and to determine final remaining mileage prediction data.

8. The method of claim 1, wherein, In step S4, the preliminary remaining mileage prediction data is compared and analyzed with the recent historical prediction records stored in the system. If the prediction deviation exceeds the preset threshold range, an adaptive correction mechanism is triggered to adjust the parameters of the long short-term memory network model to generate updated model parameters, including: The remaining mileage prediction data of the recent records is obtained from the storage library, and the recent records and the currently generated prediction data are compared item by item. By calculating the difference value of each group of data, the specific value of the prediction deviation is obtained, ， wherein represents the currently generated predicted mileage data, represents the recent recorded actual mileage data obtained from the repository, represents the difference value between the two, i.e. the concrete value of the prediction bias; If the prediction deviation exceeds the preset threshold range, an adaptive correction mechanism is triggered. By comparing the deviation trend in the historical records, the direction and amplitude of the parameters that need to be adjusted are determined. The pre-established adjustment rule is used to update the parameters of the long short-term memory network to generate the updated parameter configuration, ， wherein represents a preset threshold range, n represents the number of historical records, represents a predicted deviation value in the historical record, represents a standard deviation of the deviation, represents an adjustment coefficient; ， wherein denotes the direction of the parameter adjustment, m denotes the number of records in the most recent time period, denotes the prediction bias; ， wherein represents the final adjustment amplitude, represents the current prediction deviation, represents the average of the historical deviations, and respectively represent the maximum and minimum of the historical deviations, represents the adjustment coefficient; ， wherein denotes the updated parameter, denotes the original parameter, denotes the direction of the parameter adjustment, denotes the magnitude of the adjustment, denotes the gradient of the parameter, The system can update the parameters of the long short-term memory network according to the determined parameter direction and amplitude to generate the updated parameter configuration. By comparing the average deviation of the last m records with the average deviation of the previous m records, the change trend of the deviation is determined to determine the direction of parameter adjustment. When the current prediction deviation exceeds this range, the adaptive correction mechanism is triggered. By applying the updated parameter configuration to the prediction process, new remaining mileage prediction data is obtained, and the prediction deviation is compared again with the historical records to determine whether the prediction deviation has been reduced to within the threshold range.

9. The method of claim 1, wherein, In S4, the latest regional electric quantity original data set and user behavior feature vector are combined to re-predict the electric quantity consumption rate to generate optimized electric quantity consumption rate prediction data, including: According to the latest regional electric quantity original data set, the electric quantity consumption history record of each region is obtained, and the user behavior feature vector is combined to perform denoising processing on the original data by using a data cleaning tool to obtain the sorted electric quantity consumption basic data set; According to the collated power consumption basic data set, a user behavior influence weight table established in advance is used to obtain an influence coefficient of each behavior feature on power consumption, and a weighting calculation tool is used to determine a preliminary prediction value of the power consumption rate of each region, ， wherein represents the preliminary prediction of the power consumption rate of region i, represents the influence weight coefficient of behavior feature j, represents the value of behavior feature j in region i, and n represents the total number of behavior features; A preliminary prediction value is obtained by weighted summation of each behavior feature, and if the deviation of the preliminary prediction value of the power consumption rate from the historical true value exceeds a preset threshold, the preliminary prediction value is corrected by a regression analysis tool to obtain an adjusted power consumption rate prediction data set, ， wherein represents the percentage of prediction bias for region i, represents the preliminary forecast of the power consumption rate for region i, represents the historical real consumption rate for region i, when a correction is needed when it exceeds a preset threshold; ， where CR_i' represents the corrected electric quantity consumption rate prediction value of region i, represents the preliminary prediction value, and HR_i represents the historical real value, represents the correction coefficient, the value range of which is 0 to 1, and is used to control the degree of correction; According to the adjusted power consumption rate prediction data set, in combination with regional geographic environment data, the prediction data set is secondarily optimized by a data fusion tool to determine a final power consumption rate prediction result, ， wherein represents the final power consumption rate prediction result of the region i, represents the corrected prediction value, represents the influence value of the kth geographical environmental factor of the region i, represents the weight coefficient of the kth geographical environmental factor, and m represents the total number of geographical environmental factors.

10. The method of claim 1, wherein, The step S5 specifically includes: According to the power consumption data and the driving environment data, corresponding adaptive parameters are obtained from a pre-established storage library, and by comparing the current power consumption rate with historical records, it is determined whether the adaptive parameters are within a preset threshold range, and if the threshold range is exceeded, the adaptive parameters are corrected to obtain adjusted parameter values; Using the adjusted parameter values, in combination with the power state data, a preliminary mileage prediction result is determined by a cyclic iteration calculation method, limited by the maximum iteration number and the convergence condition, and the remaining mileage prediction data is dynamically adjusted at each iteration; By comparing the preliminary mileage prediction result with the real-time displayed power state data, if the deviation exceeds a preset threshold, the latest driving environment data and power consumption data are reacquired, the adaptive parameters are updated, and a corrected mileage prediction value is obtained; According to the corrected mileage prediction value and the power state data, in combination with real-time display requirements, the final display content is integrated by a data fusion method, and it is determined whether the accuracy requirement is met, and if not, the loop iteration calculation link is returned to obtain more accurate display data.

Images