Electric vehicle range prediction method and system based on mixed sequence model

Abstract

This invention discloses a method and system for predicting the driving range of electric vehicles based on a hybrid sequence model, belonging to the field of electric vehicle technology. The method involves collecting and preprocessing historical data related to vehicle operation, environment, and route; building and training a hybrid sequence prediction model that includes hierarchical feature extraction, cross-scale fusion, and multi-branch output; collecting and preprocessing real-time data of the same dimension and inputting it into the model to obtain energy consumption and driving range prediction results; and calculating a comprehensive temperature correction coefficient based on the driving environment and real-time battery temperature to correct the driving range prediction results and obtain the final result. The system includes corresponding data acquisition, preprocessing, model building, prediction, and correction units to execute this method. This invention utilizes multi-source data fusion to overcome the limitations of traditional models with long-term dependencies, and temperature correction makes the prediction results more consistent with actual operating conditions, improving prediction accuracy.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle technology and relates to a method and system for predicting the driving range of electric vehicles based on a hybrid sequence model. Background Technology

[0002] With increasing global focus on environmental protection and energy efficiency, electric vehicles (EVs) have rapidly developed as an alternative to traditional gasoline-powered vehicles. However, the widespread acceptance and use of EVs remain limited by their driving range, especially given the significant differences in performance under varying driving conditions, weather, and other external factors. Therefore, accurately predicting the driving range of EVs is crucial for enhancing user confidence and optimizing vehicle performance. The driving range of EVs is influenced by a variety of factors, including but not limited to battery capacity, driving habits, vehicle speed, road conditions, temperature, and the use of driver assistance systems (such as air conditioning). Traditionally, these factors have been estimated using simple linear models or empirical formulas, but these methods often overlook the complex interactions between these factors, leading to insufficient prediction accuracy.

[0003] Currently, to improve the accuracy of driving range prediction, the industry has adopted various advanced algorithms and technologies, such as machine learning and deep learning. For example, some studies have attempted to use recurrent neural networks (RNNs) to capture patterns in time-series data to improve prediction performance. Despite progress, RNNs still have certain limitations in handling long-term dependencies. Furthermore, traditional deep learning models may not be able to effectively extract long-distance dependencies between features when faced with complex input data.

[0004] Therefore, it is necessary to improve existing technologies to provide more reliable and accurate range prediction services, thereby promoting the popularization and development of electric vehicles. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting the driving range of electric vehicles based on a hybrid sequence model. By integrating multi-source data and neural network algorithms, the prediction accuracy is improved, thereby providing more reliable driving range information for electric vehicles.

[0006] To achieve the objectives of this invention, the technical solution provided by this invention is as follows: First aspect This invention provides a method for predicting the driving range of electric vehicles based on a hybrid sequence model, comprising the following steps: Step 1: Collect historical vehicle data; Step 2: Preprocess the vehicle history data to obtain the final structured feature dataset; Step 3: Based on the final structured feature dataset, build and train a prediction model based on the hybrid sequence model to obtain the trained prediction model based on the hybrid sequence model; The prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. Step 4: Collect real-time vehicle data; Step 5: Preprocess the real-time vehicle data to obtain the final structured feature real-time dataset; Step 6: Input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain the real-time energy consumption prediction results and the real-time driving range prediction results. Step 7: Collect real-time battery temperature and calculate the overall temperature correction factor. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

[0007] Second aspect This invention provides an electric vehicle range prediction system based on a hybrid sequence model, used to execute the electric vehicle range prediction method based on a hybrid sequence model, comprising the following units: Historical data acquisition unit, used to collect historical vehicle data; The historical data preprocessing unit is used to preprocess the vehicle historical data to obtain the final structured feature dataset; The model building unit is used to build and train a prediction model based on the hybrid sequence model based on the final structured feature dataset to obtain the trained prediction model based on the hybrid sequence model. The prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. Real-time data acquisition unit, used to collect real-time vehicle data; A real-time data preprocessing unit is used to preprocess the real-time vehicle data to obtain the final structured feature real-time dataset. The prediction unit is used to input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain real-time energy consumption prediction results and real-time driving range prediction results. The correction unit is used to collect the real-time battery temperature and calculate the comprehensive temperature correction coefficient. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: The hybrid sequence model built in this invention integrates multiple network structures such as LSTM, Transformer, and parameterless attention module. Through hierarchical feature extraction at the bottom, middle, and high levels, it can effectively capture short- and long-distance dependencies in time-series data, overcoming the limitations of traditional RNNs in handling long-term dependencies. At the same time, the cross-scale feature fusion unit achieves full fusion of high-dimensional time-series features of different dimensions, and the multi-branch parallel output can simultaneously and accurately predict energy consumption and driving range, improving the model's ability to process and predict complex data. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the electric vehicle range prediction method based on a hybrid sequence model provided in an embodiment of the present invention. Figure 2This is a structural diagram of the prediction model based on the hybrid sequence model in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0011] It should be noted that the acquisition of data and collection of information in this application are legal, compliant, or obtained with the consent of the subject of the data collection.

[0012] like Figure 1 As shown, this embodiment provides a method for predicting the driving range of electric vehicles based on a hybrid sequence model, including the following steps: Step 1: Collect historical vehicle data; The vehicle historical data includes: vehicle operation history data: vehicle speed, acceleration, motor power, battery voltage; driving environment history data: vehicle driving environment temperature, vehicle driving environment humidity; driving route history data: congestion coefficient, road type, altitude; Step 2: Preprocess the vehicle history data to obtain the final structured feature dataset; This step specifically includes the following steps: Step 2.1: Take energy consumption per 100 kilometers as the target variable, and vehicle speed, acceleration, motor power, battery voltage, vehicle driving environment temperature, vehicle driving environment humidity, congestion coefficient, road type, and altitude as feature variables. Step 2.2: Calculate the correlation between each feature variable and the target variable, and retain the feature variables whose correlation is greater than the first preset threshold to form an initial feature set; Step 2.3: Based on the initial feature set, calculate the correlation between each pair of feature variables, remove feature variables with a correlation greater than the second preset threshold, and form an intermediate feature set; Step 2.4: Normalize, imput missing values, and remove outliers sequentially on the intermediate feature set to obtain the final structured feature dataset.

[0013] It should be noted that the correlation is calculated using the Pearson correlation coefficient. Specifically, the Pearson correlation coefficient between the feature variable and the target variable and the two feature variables is calculated, and then the Pearson correlation coefficient is compared with the corresponding preset threshold.

[0014] This invention uses energy consumption per 100 kilometers as the target variable for feature correlation screening. It first retains features that are highly correlated with the target variable, and then removes variables with high redundancy among features. Combined with normalization, missing value imputation, and outlier removal, it effectively improves the quality and effectiveness of feature data and reduces the interference of noisy data and redundant features on the model's prediction accuracy.

[0015] Step 3: Based on the final structured feature dataset, build and train a prediction model based on the hybrid sequence model to obtain the trained prediction model based on the hybrid sequence model; like Figure 2 As shown, the prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The input layer is used to process the final structured feature dataset input and output a standardized three-dimensional tensor. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. The cross-scale feature fusion unit includes a first fusion network that processes the input high-dimensional globally enhanced temporal features and the second high-dimensional temporal features to output a first fused feature, and a second fusion network that processes the input first fused feature and the first high-dimensional temporal features to output a final fused feature; both the first fusion network and the second fusion network include upsampling, Conv, and Concat that are connected in sequence. The multi-branch parallel output unit includes a first branch network for outputting energy consumption prediction results and a second branch network for outputting driving range prediction results, which are connected in parallel; both the first branch network and the second branch network include a Conv and a prediction head connected in sequence.

[0016] It should be noted that the Transformer includes a multi-head self-attention mechanism and a feedforward neural network connected in sequence; the output of the multi-head self-attention mechanism is residually connected to its own input and then subjected to layer normalization; the output of the feedforward neural network is residually connected to its own input and then subjected to layer normalization.

[0017] Step 4: Collect real-time vehicle data; The real-time vehicle data includes: real-time vehicle operation data: vehicle speed, acceleration, motor power, and battery voltage; real-time driving environment data: ambient temperature and humidity; and real-time driving route data: congestion coefficient, road type, and altitude. Step 5: Preprocess the real-time vehicle data to obtain the final structured feature real-time dataset; It should be noted that the preprocessing process in this step refers to the preprocessing scheme used for historical data, and will not be elaborated here.

[0018] Step 6: Input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain the real-time energy consumption prediction results and the real-time driving range prediction results. Step 7: Collect real-time battery temperature and calculate the overall temperature correction factor. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

[0019] This step specifically includes the following steps: Step 7.1: Calculate the first temperature correction coefficient for the vehicle's ambient temperature in the real-time driving environment data in segments. The second temperature correction coefficient of the battery's real-time temperature ; Among them, the first temperature correction coefficient The calculation formula is as follows: when At -20℃, ; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, = ; Among them, the second temperature correction coefficient The calculation formula is as follows: when At -20℃, =0.4; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, = ; in, The vehicle's ambient temperature is measured in real-time driving environment data. Real-time battery temperature; Step 7.2: Based on and Calculate the overall temperature correction factor KT The formula is as follows: ; in, The vehicle's operating environment temperature is used as the weighting factor. Weighted by real-time battery temperature; Step 7.3: Based on Calculate the final range prediction result based on the real-time range prediction results. The details are as follows: ; in, This is the result of real-time driving range prediction.

[0020] It should be noted that this invention introduces a dual correction mechanism based on ambient driving temperature and real-time battery temperature. By calculating the temperature correction sub-coefficient in segments and combining it with weights to obtain a comprehensive temperature correction coefficient, the model prediction results are corrected a second time. This fully considers the actual impact of temperature, a key factor, on battery performance and vehicle range, making the prediction results more consistent with the actual driving conditions of the vehicle under different temperature conditions.

[0021] It should be noted that this invention collects historical and real-time data from three major categories—vehicle operation, driving environment, and driving route—as the basis for prediction. This comprehensively covers key factors affecting the driving range of electric vehicles, such as vehicle speed, temperature, and congestion coefficient, avoiding the one-sidedness of prediction caused by a single data dimension and laying a data foundation for accurate prediction.

[0022] The following comparative experiment uses the same set of real-world multi-source data from electric vehicles. The data used includes vehicle operation data (vehicle speed, acceleration, motor power, battery voltage), driving environment data (ambient temperature, humidity), and driving route data (congestion coefficient, road type, altitude). Energy consumption per 100 kilometers is the target variable. The traditional RNN model and the hybrid sequence model of this invention are used for training and prediction, respectively. The comparison results are shown in Table 1 below:

[0023] As shown in Table 1, the average absolute error of the driving range using the traditional RNN model is 8.2 km, and the root mean square error is 10.5 km. The average absolute error of the driving range using the hybrid sequence model of this invention is reduced to 3.5 km, and the root mean square error is reduced to 4.2 km, significantly improving the prediction accuracy. In addition, the prediction accuracy of the driving range under complex road conditions reaches 89.5%, and the prediction error under extreme temperature conditions is reduced to 4.1 km. The overall prediction accuracy, stability, and adaptability are significantly better than those of the traditional RNN model.

[0024] In addition, the present invention provides an electric vehicle range prediction system based on a hybrid sequence model, used to execute the electric vehicle range prediction method based on a hybrid sequence model, comprising the following units: Historical data acquisition unit, used to collect historical vehicle data; The vehicle historical data includes: vehicle operation history data: vehicle speed, acceleration, motor power, battery voltage; driving environment history data: vehicle driving environment temperature, vehicle driving environment humidity; driving route history data: congestion coefficient, road type, altitude; The historical data preprocessing unit is used to preprocess the vehicle historical data to obtain the final structured feature dataset; The model building unit is used to build and train a prediction model based on the hybrid sequence model based on the final structured feature dataset to obtain the trained prediction model based on the hybrid sequence model. Real-time data acquisition unit, used to collect real-time vehicle data; The real-time vehicle data includes: real-time vehicle operation data: vehicle speed, acceleration, motor power, and battery voltage; real-time driving environment data: ambient temperature and humidity; and real-time driving route data: congestion coefficient, road type, and altitude. A real-time data preprocessing unit is used to preprocess the real-time vehicle data to obtain the final structured feature real-time dataset. The prediction unit is used to input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain real-time energy consumption prediction results and real-time driving range prediction results. The correction unit is used to collect the real-time battery temperature and calculate the comprehensive temperature correction coefficient. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

[0025] Specifically, the historical data preprocessing unit is used to perform the following steps: Step 2.1: Take energy consumption per 100 kilometers as the target variable, and vehicle speed, acceleration, motor power, battery voltage, vehicle driving environment temperature, vehicle driving environment humidity, congestion coefficient, road type, and altitude as feature variables. Step 2.2: Calculate the correlation between each feature variable and the target variable, and retain the feature variables whose correlation is greater than the first preset threshold to form an initial feature set; Step 2.3: Based on the initial feature set, calculate the correlation between each pair of feature variables, remove feature variables with a correlation greater than the second preset threshold, and form an intermediate feature set; Step 2.4: Normalize, imput missing values, and remove outliers sequentially on the intermediate feature set to obtain the final structured feature dataset.

[0026] The prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The input layer is used to process the final structured feature dataset input and output a standardized three-dimensional tensor. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. The cross-scale feature fusion unit includes a first fusion network that processes the input high-dimensional globally enhanced temporal features and the second high-dimensional temporal features to output a first fused feature, and a second fusion network that processes the input first fused feature and the first high-dimensional temporal features to output a final fused feature; both the first fusion network and the second fusion network include upsampling, Conv, and Concat that are connected in sequence. The multi-branch parallel output unit includes a first branch network for outputting energy consumption prediction results and a second branch network for outputting driving range prediction results, which are connected in parallel; both the first branch network and the second branch network include a Conv and a prediction head connected in sequence.

[0027] The Transformer includes a multi-head self-attention mechanism and a feedforward neural network connected in sequence; the output of the multi-head self-attention mechanism is residually connected to its own input and then subjected to layer normalization; the output of the feedforward neural network is residually connected to its own input and then subjected to layer normalization.

[0028] The correction unit is configured to perform the following steps: Step 7.1: Calculate the first temperature correction coefficient for the vehicle's ambient temperature in the real-time driving environment data in segments. The second temperature correction coefficient of the battery's real-time temperature ; Among them, the first temperature correction coefficient The calculation formula is as follows: when At -20℃, =0.4; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, = ; Among them, the second temperature correction coefficient The calculation formula is as follows: when At -20℃, =0.4; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, = ; in, The vehicle's ambient temperature is measured in real-time driving environment data. Real-time battery temperature; Step 7.2: Based on and Calculate the overall temperature correction factor KT The formula is as follows: ; in, The vehicle's operating environment temperature is used as the weighting factor. Weighted by real-time battery temperature; Step 7.3: Based on Calculate the final range prediction result based on the real-time range prediction results. The details are as follows: ; in, This is the result of real-time driving range prediction.

[0029] Finally, it should be noted that the above embodiments are merely illustrative and explanatory of the present invention, and are not intended to limit the present invention to the scope of the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many more variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting the driving range of electric vehicles based on a hybrid sequence model, characterized in that, Includes the following steps: Step 1: Collect historical vehicle data; Step 2: Preprocess the vehicle history data to obtain the final structured feature dataset; Step 3: Based on the final structured feature dataset, build and train a prediction model based on the hybrid sequence model to obtain the trained prediction model based on the hybrid sequence model. The prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. Step 4: Collect real-time vehicle data; Step 5: Preprocess the real-time vehicle data to obtain the final structured feature real-time dataset; Step 6: Input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain the real-time energy consumption prediction results and the real-time driving range prediction results. Step 7: Collect real-time battery temperature and calculate the overall temperature correction factor. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

2. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 1, characterized in that, In step 1, the vehicle historical data includes: vehicle operation historical data: vehicle speed, acceleration, motor power, battery voltage; driving environment historical data: vehicle driving environment temperature, vehicle driving environment humidity; driving route historical data: congestion coefficient, road type, altitude.

3. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2.1: Take energy consumption per 100 kilometers as the target variable, and vehicle speed, acceleration, motor power, battery voltage, vehicle driving environment temperature, vehicle driving environment humidity, congestion coefficient, road type, and altitude as feature variables. Step 2.2: Calculate the correlation between each feature variable and the target variable, and retain the feature variables whose correlation is greater than the first preset threshold to form an initial feature set; Step 2.3: Based on the initial feature set, calculate the correlation between each pair of feature variables, remove feature variables with a correlation greater than the second preset threshold, and form an intermediate feature set; Step 2.4: Normalize, imput missing values, and remove outliers sequentially on the intermediate feature set to obtain the final structured feature dataset.

4. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 3, characterized in that, In step 3, the input layer is used to process the final structured feature dataset and output a standardized three-dimensional tensor. The cross-scale feature fusion unit includes a first fusion network that processes the input high-dimensional globally enhanced temporal features and the second high-dimensional temporal features to output a first fused feature, and a second fusion network that processes the input first fused feature and the first high-dimensional temporal features to output a final fused feature; both the first fusion network and the second fusion network include upsampling, Conv, and Concat that are connected in sequence. The multi-branch parallel output unit includes a first branch network for outputting energy consumption prediction results and a second branch network for outputting driving range prediction results, which are connected in parallel; both the first branch network and the second branch network include a Conv and a prediction head connected in sequence.

5. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 4, characterized in that, The Transformer comprises a multi-head self-attention mechanism and a feedforward neural network connected in sequence; the output of the multi-head self-attention mechanism is residually connected to its own input and then subjected to layer normalization; the output of the feedforward neural network is residually connected to its own input and then subjected to layer normalization.

6. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 5, characterized in that, In step 4, the real-time vehicle data includes: real-time vehicle operation data: vehicle speed, acceleration, motor power, and battery voltage; real-time driving environment data: vehicle driving environment temperature and vehicle driving environment humidity; and real-time driving route data: congestion coefficient, road type, and altitude.

7. The method for predicting the driving range of an electric vehicle based on a hybrid sequence model according to claim 6, characterized in that, Step 7 specifically includes the following steps: Step 7.1: Calculate the first temperature correction coefficient for the vehicle's ambient temperature in the real-time driving environment data in segments. The second temperature correction coefficient of the battery's real-time temperature ; Among them, the first temperature correction coefficient The calculation formula is as follows: when At -20℃, ; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, ; Among them, the second temperature correction coefficient The calculation formula is as follows: when At -20℃, ; When -20℃≤ At <0℃, = ; When 0℃≤ At ≤25℃, = ; When 25℃ At ≤55℃, = ; When 55℃ hour, ; in, The vehicle's ambient temperature is measured in real-time driving environment data. Real-time battery temperature; Step 7.2: Based on and Calculate the overall temperature correction factor KT The formula is as follows: ； in, The vehicle's operating environment temperature is used as the weighting factor. Weighted by real-time battery temperature; Step 7.3: Based on Calculate the final range prediction result based on the real-time range prediction results. The details are as follows: ； in, This is the result of real-time driving range prediction.

8. An electric vehicle range prediction system based on a hybrid sequence model, used to execute the electric vehicle range prediction method based on a hybrid sequence model as described in any one of claims 1-7, characterized in that, Includes the following units: Historical data acquisition unit, used to collect historical vehicle data; The historical data preprocessing unit is used to preprocess the vehicle historical data to obtain the final structured feature dataset; The model building unit is used to build and train a prediction model based on the final structured feature dataset to obtain the trained prediction model based on the hybrid sequence model. The prediction model based on the hybrid sequence model includes an input layer, a backbone feature extraction unit, a cross-scale feature fusion unit, and a multi-branch parallel output unit connected in sequence. The backbone feature extraction unit comprises, in sequence, a low-level temporal feature extraction network that processes the input standardized 3D tensor and outputs a first high-dimensional temporal feature; a mid-level feature extraction network that processes the input first high-dimensional temporal feature and outputs a second high-dimensional temporal feature that is higher than the first high-dimensional temporal feature; and a high-level feature extraction and enhancement network that processes the input second high-dimensional temporal feature and outputs a high-dimensional globally enhanced temporal feature that is higher than the second high-dimensional temporal feature. The low-level and mid-level feature extraction networks have the same structure, each comprising, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and a temporal downsampling layer. The high-level feature extraction and enhancement network comprises, in sequence, an LSTM stacked layer, a Transformer, a parameterless attention module, a temporal nonlocal block, and an attention pyramid. Real-time data acquisition unit, used to collect real-time vehicle data; A real-time data preprocessing unit is used to preprocess the real-time vehicle data to obtain the final structured feature real-time dataset. The prediction unit is used to input the final structured feature real-time dataset into the trained prediction model based on the hybrid sequence model to obtain real-time energy consumption prediction results and real-time driving range prediction results. The correction unit is used to collect the real-time battery temperature and calculate the comprehensive temperature correction coefficient. KT; based on KT The real-time driving range prediction results are corrected to obtain the final driving range prediction results.

9. The electric vehicle range prediction system based on a hybrid sequence model according to claim 8, characterized in that, The vehicle historical data includes: vehicle operation historical data: vehicle speed, acceleration, motor power, battery voltage; driving environment historical data: vehicle driving environment temperature, vehicle driving environment humidity; driving route historical data: congestion coefficient, road type, altitude.

10. The electric vehicle driving range prediction system based on a hybrid sequence model according to claim 9, characterized in that, The real-time vehicle data includes: real-time vehicle operation data: vehicle speed, acceleration, motor power, and battery voltage; real-time driving environment data: vehicle driving environment temperature and humidity; and real-time driving route data: congestion coefficient, road type, and altitude.

Images