Self-learning-based endurance mileage estimation method for pure electric vehicle
Through self-learning-based methods, a pure electric vehicle range estimation model is established using vehicle-mounted sensor data, neural network or support vector machine algorithm, which solves the problems of low estimation accuracy and insufficient adaptability in the prior art, and achieves higher precision and adaptability range estimation.
Patent Information
- Application Number
- CN202510049142.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-13
- Publication Date
- 2025-05-27
AI Technical Summary
The existing pure electric vehicles have low accuracy and cannot effectively adapt to complex and changeable actual driving conditions and vehicle usage changes.
A self-learning-based method is adopted to collect vehicle driving data through on-board sensors, and a range estimation model is established using neural networks or support vector machine algorithms, model training and optimization is performed, and the model is updated using incremental learning method.
It improves the accuracy and adaptability of range estimation, can more accurately consider various dynamic factors, adapt to different driving conditions and driving habits, and provide real-time and accurate range information.
Smart Images
Figure CN120039123A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of connectors, and in particular to a method for estimating the cruising range of a pure electric vehicle based on self-learning. Background Art
[0002] As the concept of environmental protection becomes more popular and the energy structure is adjusted, the share of pure electric vehicles in the market is gradually increasing. However, the uncertainty of driving range has always been one of the main problems that plague pure electric vehicle users. Accurately estimating the driving range of pure electric vehicles is of vital importance to improving user experience, enhancing the reliability of travel planning, and promoting the further development of pure electric vehicles.
[0003] At present, the traditional pure electric vehicle range estimation method has the following problems:
[0004] 1. Estimation method based on battery capacity: This method simply estimates the range based on the nominal capacity and current power of the battery. But in fact, the actual capacity of the battery will be affected by many factors, such as temperature, number of charge and discharge times, age, etc. In addition, the driving conditions of the vehicle (such as high-speed driving, frequent acceleration and deceleration, etc.) will also have a great impact on the power consumption of the battery, making the estimation method based on battery capacity less accurate.
[0005] 2. Estimation method based on driving conditions: This method takes into account the impact of driving conditions such as vehicle speed and acceleration on the range. However, this method can usually only estimate specific driving conditions and is not adaptable enough to complex and changeable actual driving conditions. Moreover, the driving habits of different users vary greatly, which will also lead to uncertainty in driving conditions, thus affecting the estimation accuracy.
[0006] 3. Estimation method based on machine learning: Although machine learning algorithms have improved the accuracy of mileage estimation to a certain extent, existing machine learning methods often require a large amount of labeled data for training, and the cost of obtaining high-quality labeled data is high. In addition, the technology of pure electric vehicles continues to develop, and new models and battery technologies continue to emerge. Existing machine learning models are difficult to quickly adapt to these changes, making it difficult to update and optimize the model. Summary of the invention
[0007] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a method for estimating the range of a pure electric vehicle based on self-learning.
[0008] In order to solve the problems in the above background technology, the present invention is implemented by the following technical solutions:
[0009] A method for estimating the range of a pure electric vehicle based on self-learning comprises the following steps:
[0010] S1. Obtain vehicle information, collect vehicle driving data through in-vehicle sensors, including vehicle speed, acceleration, battery power, battery temperature, driving mileage, and road conditions, and store the collected vehicle information in the in-vehicle storage device;
[0011] S2. Preprocess the collected data, and the preprocessing is data cleaning and normalization;
[0012] S3. Establish an estimated driving range model, select a self-learning algorithm and determine input variables and output variables, and initialize model parameters through the self-learning algorithm;
[0013] The self-learning algorithm includes a neural network algorithm or a support vector machine algorithm;
[0014] S4. Model division and training, divide the preprocessed data into a training set, a validation set, and a test set for model training, and continuously adjust model parameters to improve the estimation accuracy, obtain the estimated driving range value and display it to the user;
[0015] S5. Continuously collect new driving data, and update, evaluate, and optimize the estimated driving range model after preprocessing the newly collected data.
[0016] Furthermore, the vehicle speed information is obtained through a vehicle speed sensor; the acceleration information is obtained through an acceleration sensor; the battery power information is obtained through a battery power sensor; the battery temperature is obtained through a battery temperature sensor; the road condition information is determined by the vehicle position through a positioning system and combined with map data; the road condition information includes the road slope and the road congestion situation;
[0017] The frequency of collecting the vehicle driving data is 1 - 20 times / s.
[0018] Furthermore, the data cleaning and normalization are as follows:
[0019] Set large samples and small samples according to the driving data: The large sample consumes time T, the initial total power consumption within the large sample time T is E^0, and the initial driving mileage is S; the small sample consumes time t, the initial power consumption within the small sample time t is e^0, and the initial driving mileage value is s; the condition for selecting the driving data is: vehicle speed > 0;
[0020] The mileage that the vehicle can travel per degree of electricity = the driving mileage value within the sample period / (the total power consumption within the sample period + 0.001);
[0021] The total power consumption within the sample period = the total integrated power consumption within the sample period + the real-time power consumption - the average power within the sample period;
[0022] Mileage value within the sample period = Total integrated mileage within the sample period + Real-time mileage value - Average mileage within the sample period;
[0023] Mileage that can be traveled per 1% of SOC = (Mileage value that can be traveled per kWh of the vehicle within the T-time sample * 70% + Mileage value that can be traveled per kWh of the vehicle within the t-hour sample * 30%) * Total battery pack capacity / 100;
[0024] Endurance mileage = SOC * Mileage that can be traveled per 1% of SOC.
[0025] Furthermore, the input variables are vehicle speed, acceleration, battery power, battery temperature, mileage, road condition information, and driving habit information; the driving habit information includes the number of hard accelerations and the number of hard brakes;
[0026] The output variable is the estimated value of the endurance mileage of the battery electric vehicle.
[0027] Furthermore, the training set is used for the training of the endurance mileage estimation model;
[0028] The validation set is used to adjust the hyperparameters of the endurance mileage estimation model to minimize the error between the output value of the endurance mileage estimation model and the actual endurance mileage time and to monitor the model performance during training;
[0029] The test set is used to evaluate the generalization ability of the trained endurance mileage estimation model;
[0030] The preprocessed data is divided into a training set, a validation set, and a test set in the ratio of 70%, 15%, and 15%.
[0031] Furthermore, the hyperparameters include the number of layers of the neural network, the number of neurons in each layer, the learning rate, and the kernel function parameters and penalty coefficients of the support vector machine;
[0032] The adjustment of the hyperparameters is as follows: Use grid search, random search, or genetic algorithm to adjust the hyperparameters, and then select the hyperparameter combination with the best performance by evaluating the model performance under different hyperparameter combinations on the validation set.
[0033] Furthermore, the estimated value is displayed to the user on the vehicle dashboard, the center console screen, or the mobile phone APP, and the user selects the corresponding charging location and charging time based on the estimated value.
[0034] Furthermore, the endurance mileage estimation model is updated by using the incremental learning method to locally adjust and optimize the endurance mileage estimation model with new data without retraining the entire endurance mileage estimation model;
[0035] The evaluation and optimization of the driving range estimation model are as follows: regularly evaluate the performance of the updated driving range estimation model, and verify it using the test set or newly collected data; if the performance of the driving range estimation model deteriorates, adjust the hyperparameters, increase the amount of training data, or improve the algorithm to improve the accuracy and adaptability of the driving range estimation model.
[0036] Compared with the prior art, the present invention has the following beneficial technical effects:
[0037] 1. Improve the estimation accuracy: By continuously optimizing the model parameters through the self-learning algorithm, it can fully consider the influence of various dynamic factors on the driving range, so as to more accurately estimate the driving range of pure electric vehicles.
[0038] 2. Enhance the adaptability: It can adapt to different driving conditions, environmental conditions, driving habits, and vehicle usage situations, and has strong adaptability. As the vehicle is used and data accumulates, the model can continuously learn and optimize to better adapt to various changes.
[0039] 3. Real-time estimation: It can quickly estimate the driving range based on the real-time driving data of the vehicle, provide timely information for users, help users reasonably arrange their trips, and reduce the anxiety and inconvenience caused by insufficient driving range.
[0040] 4. Convenient for users to use: Users can understand the driving range of the vehicle at any time through the vehicle dashboard, center console screen, or mobile phone APP, which is convenient for travel planning; at the same time, the actual driving data (newly collected data) feedback by users can also help further optimize the model and improve the estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0042] As Figure 1 shown, a self-learning-based driving range estimation method for pure electric vehicles includes the following steps:
[0043] S1. Obtain vehicle information, collect vehicle driving data through on-vehicle sensors, including vehicle speed, acceleration, battery power, battery temperature, driving mileage, and road condition information, and store the collected vehicle information in the on-vehicle storage device;
[0044] S2. Preprocess the collected data, and the preprocessing is data cleaning and normalization;
[0045] S3. Establish a driving range estimation model, select a self-learning algorithm, determine the input variables and output variables, and initialize the model parameters through the self-learning algorithm;
[0046] The self-learning algorithm includes a neural network algorithm or a support vector machine algorithm;
[0047] The self-learning algorithm is selected as a neural network algorithm or a support vector machine algorithm;
[0048] Neural network algorithm: It has a powerful non-linear mapping ability and self-learning ability, and can automatically learn the complex relationship between input variables and output variables through a large amount of training data; A multi-layer feedforward neural network can be adopted, including an input layer, a hidden layer and an output layer.
[0049] Support vector machine algorithm: It performs well in small-sample and non-linear problems and can effectively perform pattern recognition and regression analysis. By finding an optimal hyperplane, it classifies data of different categories or performs regression prediction on continuous variables.
[0050] S4. Model division and training: Divide the preprocessed data into a training set, a validation set and a test set for model training, and continuously adjust the model parameters to improve the estimation accuracy, obtain the estimated value of the cruising range and display it to the user;
[0051] S5. Continuously collect new driving data, and update, evaluate and optimize the cruising range estimation model after preprocessing the newly collected data.
[0052] The vehicle speed information is obtained through a vehicle speed sensor; the acceleration information is obtained through an acceleration sensor; the battery power information is obtained through a battery power sensor; the battery temperature is obtained through a battery temperature sensor; the road condition information is determined by the vehicle position through a positioning system and combined with map data; the road condition information includes the road gradient and the road congestion situation;
[0053] The frequency of collecting vehicle driving data is 1 - 20 times / s.
[0054] Data cleaning and normalization are as follows:
[0055] Set large samples and small samples according to the driving data: The large sample consumes time T, the initial total power consumption within the large sample time T is E^0, and the initial driving mileage is S; the small sample consumes time t, the initial power consumption within the small sample time t is e^0, and the initial driving mileage value is s; The condition for selecting driving data is: vehicle speed > 0;
[0056] The mileage value that the vehicle can travel per degree of electricity = the driving mileage value within the sample period / (the total power consumption within the sample period + 0.001);
[0057] The total power consumption within the sample period = the total integrated power within the sample period + the real-time power consumption - the average power within the sample period;
[0058] Driving mileage value within the sample period = Total integrated mileage within the sample period + Real-time driving mileage value - Average mileage within the sample period;
[0059] Mileage that can be driven per 1% of SOC = (Mileage that can be driven per kWh of the vehicle within the T-time sample * 70% + Mileage that can be driven per kWh of the vehicle within the t-hour sample * 30%) * Total battery pack capacity / 100;
[0060] Endurance mileage = SOC * Mileage that can be driven per 1% of SOC.
[0061] The input variables are vehicle speed, acceleration, battery charge, battery temperature, driving mileage, road condition information, and driving habit information; driving habit information includes the number of hard accelerations and hard brakes;
[0062] The output variable is the estimated value of the endurance mileage of the pure electric vehicle.
[0063] The training set is used for training the endurance mileage estimation model;
[0064] The validation set is used to adjust the hyperparameters of the endurance mileage estimation model to minimize the error between the output value of the endurance mileage estimation model and the actual endurance mileage time and to monitor the model performance during training;
[0065] The test set is used to evaluate the generalization ability of the trained endurance mileage estimation model;
[0066] The preprocessed data is divided into a training set, a validation set, and a test set in the ratio of 70%, 15%, and 15%.
[0067] Use the training set to train the endurance mileage estimation model. By continuously adjusting the parameters of the endurance mileage estimation model, minimize the error between the output value of the endurance mileage estimation model and the actual endurance mileage value;
[0068] For the neural network algorithm, the backpropagation algorithm can be used for training. Calculate the error between the output value and the actual value of the endurance mileage estimation model, and then adjust the weights and biases of the network by backpropagating the error signal to reduce the error;
[0069] For the support vector machine algorithm, an optimization algorithm (such as the sequential minimal optimization algorithm) can be used to solve the parameters of the optimal hyperplane to minimize the error of the endurance mileage estimation model on the training set.
[0070] Initialize the model parameters. For the neural network algorithm, it is necessary to initialize the weights and biases of the network. Using the method of random initialization, initialize the weights to small random values and the biases to zero or small constants;
[0071] For the support vector machine algorithm, determine the kernel function type and related parameters. Common kernel functions include linear kernel function, polynomial kernel function, radial basis kernel function, etc. Select the appropriate kernel function and parameters through methods such as cross-validation.
[0072] Hyperparameters include the number of layers of the neural network, the number of neurons in each layer, the learning rate, as well as the kernel function parameters and penalty coefficients of the support vector machine;
[0073] The adjustment of hyperparameters is as follows: Use grid search, random search or genetic algorithm to adjust hyperparameters, and then evaluate the model performance under different hyperparameter combinations on the validation set to select the hyperparameter combination with the best performance.
[0074] The estimated value is displayed to the user on the vehicle dashboard, the center console screen or the mobile phone APP. Based on the estimated value, the user selects the corresponding charging location and charging time.
[0075] The driving range estimation model is updated as follows: Adopt the incremental learning method to locally adjust and optimize the driving range estimation model using new data without retraining the entire model;
[0076] For the neural network algorithm, the weights and biases of the network can be gradually adjusted according to new data through online learning; for the support vector machine algorithm, new data points can be added to the original driving range estimation model for re-optimization.
[0077] The evaluation and optimization of the driving range estimation model are as follows: Regularly evaluate the performance of the updated driving range estimation model, and use the test set or newly collected data for verification; if the performance of the driving range estimation model deteriorates, adjust the hyperparameters, increase the amount of training data or improve the algorithm to improve the accuracy and adaptability of the model.
Claims
1. A method for estimating the range of a pure electric vehicle based on self-learning, characterized in that: The following steps are involved: S1, obtaining vehicle information, collecting vehicle driving data through vehicle-mounted sensors, including vehicle speed, acceleration, battery power, battery temperature, mileage, and road condition information, and storing the collected vehicle information in a vehicle-mounted storage device; S2, preprocessing the collected data, wherein the preprocessing includes data cleaning and normalization; S3, establishing a cruising range estimation model, selecting a self-learning algorithm and determining input variables and output variables, and initializing model parameters through the self-learning algorithm; The self-learning algorithm includes a neural network algorithm or a support vector machine algorithm; S4, model division and training, divide the preprocessed data into training set, validation set and test set for model training, and continuously adjust the model parameters to improve the estimation accuracy, obtain the estimated mileage value and display it to the user; S5, continuously collects new driving data, and pre-processes the newly collected data to update, evaluate and optimize the range estimation model.
2. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 1, characterized in that: The vehicle speed information is obtained through a vehicle speed sensor; the acceleration information is obtained through an acceleration sensor; the battery power information is obtained through a battery power sensor; the battery temperature is obtained through a battery temperature sensor; the road condition information is obtained by determining the vehicle position through a positioning system and combining it with map data; the road condition information includes road slope and road congestion; The frequency of collecting the vehicle driving data is 1 to 20 times / s.
3. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 2, characterized in that: The data cleaning and normalization are as follows: Set large samples and small samples according to driving data: the consumption time of the large sample is T, the initial value of the total power consumed in the large sample T time is E^0, and the initial mileage is S; the consumption time of the small sample is t, the initial value of the power consumed in the small sample t time is e^0, and the initial mileage is s; the driving data selection condition is: vehicle speed>0; The mileage that the vehicle can travel per kWh = the mileage value in the sample period / (total power consumed in the sample period + 0.001); Total power consumption during the sample period = total integrated power consumption during the sample period + real-time power consumption - average power consumption during the sample period; Mileage value within the sample period = total mileage points within the sample period + real-time mileage value - average mileage value within the sample period; The mileage that can be driven for every 1% SOC = (the mileage value that can be driven per kWh of the sample vehicle at time T * 70% + the mileage value that can be driven per kWh of the sample vehicle at hour t * 30%) * total battery pack capacity / 100; Cruising range = SOC * mileage per 1% SOC.
4. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 1, characterized in that: The input variables are vehicle speed, acceleration, battery power, battery temperature, mileage, road condition information and driving habit information; the driving habit information includes the number of sudden accelerations and sudden braking times; The output variable is an estimated value of the cruising range of the pure electric vehicle.
5. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 1, characterized in that: The training set is used for training a range estimation model; The validation set is used to adjust the hyperparameters of the range estimation model to minimize the error between the output value of the range estimation model and the actual range time and to monitor the model performance during the training process; The test set is used to evaluate the generalization ability of the trained range estimation model; The preprocessed data is divided into training set, validation set and test set in the ratio of 70%, 15% and 15%.
6. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 5, characterized in that: The hyperparameters include the number of layers of the neural network, the number of neurons in each layer, the learning rate, and the kernel function parameters and penalty coefficient of the support vector machine; The hyperparameters are adjusted by using grid search, random search or genetic algorithm to adjust the hyperparameters, and then evaluating the model performance under different hyperparameter combinations on a validation set to select the hyperparameter combination with the best performance.
7. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 1, characterized in that: Displaying the estimated value to the user means displaying the estimated value on the vehicle dashboard, central control screen or mobile phone APP, and the user selects the corresponding charging location and charging time based on the estimated value.
8. The method for estimating the range of a pure electric vehicle based on self-learning according to claim 1, characterized in that: The cruising range estimation model is updated by using an incremental learning method to locally adjust and optimize the cruising range estimation model using new data without retraining the entire cruising range estimation model; The evaluation and optimization of the range estimation model is to regularly evaluate the performance of the updated range estimation model and verify it using a test set or newly collected data; if the performance of the range estimation model deteriorates, adjust the hyperparameters, increase the amount of training data, or improve the algorithm to improve the accuracy and adaptability of the range estimation model.