New energy automobile automatic optimization calibration method
The automatic optimization calibration method for new energy vehicles addresses the challenge of adapting to diverse driving conditions and styles by using advanced algorithms and modules to dynamically update control parameters, enhancing adaptability and optimizing energy consumption.
Patent Information
- Application Number
- CN202411219445.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-12-27
AI Technical Summary
Traditional calibration methods for new energy vehicles fail to adapt to varying driving conditions and styles, leading to inefficient energy consumption due to fixed calibration parameters, and existing solutions do not adequately address the complexity of parameter optimization and environmental adaptability.
An automatic optimization calibration method using a parameter correction module, calibration parameter setting module, database module, and working condition calling module to dynamically update control parameters based on real-time driving conditions and styles, employing Bi-LSTM neural networks and improved SVM algorithms for recognition, and genetic algorithms for cluster optimization.
Enhances the adaptability of calibration parameters to changing environments, optimizing energy consumption and power performance by pre-processing actual vehicle data, performing cluster optimization, and iteratively updating control parameters for improved accuracy and efficiency.
Smart Images
Figure CN119200556A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of whole vehicle calibration, and in particular relates to an automatic optimization calibration method for a new energy vehicle. Background Art
[0002] The traditional economic car calibration method uses the standard driving condition of the car as the test condition, ignoring the impact of different driving styles on energy consumption, and the control parameters for the purpose of global optimization are not changed after calibration. However, the driving environment of the vehicle is not static. When the driver's style and driving conditions change, the fixed calibration parameters will cause the vehicle's energy consumption to deteriorate. It fails to fully consider the impact of differentiated driving conditions and driving styles during actual driving, making the calibration parameters poorly adaptable to the environment. At present, the market has higher and higher requirements for the reliability, comfort and economy of new energy vehicles, which requires major car companies and R&D institutions to accurately and quickly develop automotive electronic control system products. The traditional calibration method is to perform manual calibration based on the personal experience of engineers, and then correct the parameters based on the test results, and test again, and so on, until the appropriate control parameters are found. This calibration method takes a lot of time and manpower, and the control parameters selected by engineers based on experience are prone to miss the optimal control parameters, resulting in poor calibration results.
[0003] In existing research, Chinese patent CN110254417A discloses a hybrid vehicle control method based on dual recognition of actual working conditions and driving styles. In its technical solution, an LVQ neural network is used to perform offline training on different working conditions and different types of driving styles corresponding to different working conditions to obtain a working condition and driving style recognition model based on the LVQ neural network; the vehicle control parameters are optimized according to the working conditions reflecting the driving style under different types of driving conditions, the optimal control parameters are obtained, and a dynamic recognition strategy is set to make the engine work in a high-efficiency zone, thereby greatly improving fuel economy. This is a feasible solution, but the working condition has obvious timing characteristics. Taking the timing characteristics of the working condition into consideration can improve the accuracy of the working condition recognition model, and the complexity of the parameter optimization calibration process is not considered. As the calibration parameters increase, the workload will increase sharply. Chinese patent CN112859878A relates to a control parameter automatic calibration method for hybrid unmanned vehicles, which uses a simulated annealing algorithm as a control parameter optimization method, calibrates the current control parameters into the vehicle controller for testing, and then continuously iterates through the algorithm. If the test results meet the energy consumption requirements, the optimization is stopped, the current control parameters are recorded, and the calibration work is completed. However, this calibration method only introduces one case, and does not consider the problem of poor adaptability of calibration parameters under the influence of differentiated environments, and its target setting and constraint conditions are relatively simple, and other key factors are not considered. Summary of the invention
[0004] The purpose of the present invention is to provide an automatic optimization calibration method for new energy vehicles. The calibration method can optimize and update the total parameter library of control parameters through a parameter correction module, a calibration parameter setting module, a database module, and a working condition calling module in the automatic optimization calibration method, thereby enhancing the adaptability of the parameters to a changing environment and solving the problem of poor adaptability of the calibration parameters under the influence of differentiated driving conditions and driving styles during actual driving.
[0005] In order to achieve the above object, the present invention adopts the following technical solution: A new energy vehicle automatic optimization calibration method comprises the following steps: S1. Select characteristic parameters according to the characteristics of the working condition data, wherein the characteristic parameters include average vehicle speed, proportion of time when the speed is greater than 30 km / h, and maximum acceleration; S2. Local representative working conditions are obtained by clustering and optimizing the extracted characteristic parameters, and comprehensive cycle working conditions are obtained by combining the local representative working conditions. A real-time working condition recognition model and a driving style recognition model are respectively constructed according to the obtained working condition information; S3, obtaining n groups of vehicle control parameters according to the combination of different local representative working conditions and multiple driving style types, and then constructing a total parameter library of control parameters; S4. According to the designed automatic optimization calibration method, determine the minimum energy consumption of the whole vehicle as the goal and the power performance as the constraint condition, and establish an automatic optimization calibration model; the automatic optimization calibration method includes a parameter correction module, a calibration parameter setting module, a database module, and a working condition calling module. S5. Combine the test process with the optimization calculation, organize the optimization experiment according to the optimization goal, perform the optimization calculation online, automatically determine a set of optimal control parameters as the new parameters for the next test, and continuously iterate according to the optimization goal until the optimal solution is obtained or the algorithm requirements are met, and complete the automatic calibration of the total parameter library of control parameters one by one.
[0006] Furthermore, the operating condition data is obtained from actual vehicle testing of new energy buses. Before extracting characteristic parameters, the operating condition sample data needs to be divided into short-trip segments of specified time, and the principal component analysis method is used to extract the main features of the short trip to obtain several groups of information with high contribution.
[0007] Furthermore, in step S2, a three-dimensional scatter plot of the operating condition block is drawn according to the characteristic parameters, and a genetic algorithm is applied to optimize the initial cluster center to obtain the initial cluster center after genetic optimization, and then the local representative operating condition is obtained through the results of cluster analysis, and the local representative operating condition includes unimpeded operating condition in urban area, congested operating condition in urban area, unimpeded operating condition in suburbia, and expressway operating condition.
[0008] Furthermore, referring to foreign standard working conditions, the duration of typical urban working conditions is taken as about 1600 seconds. By using the proportion of time occupied by each representative working condition in the entire recorded data, its time occupied in the final constructed working condition can be determined to obtain the comprehensive cycle working condition.
[0009] Furthermore, the specific operation of the real-time working condition identification model includes the following steps: R1: Construct a Bi-LSTM neural network, and use the short-trip principal components obtained by principal component analysis as the input variables of the identification model. The short-trip principal components in each sequence are arranged in order to obtain the time series input layer. The number of neurons in the hidden layer is set according to the number of working condition categories. The loss function is calculated in the form of mean square error during the back propagation process, and the probability value of each working condition category is used as the output; R2. Take 90% of the short-stroke samples as training data, perform deep learning training on the neural network, and obtain a real-time working condition recognition model based on the Bi-LSTM neural network.
[0010] Furthermore, the specific operation arrangement of the driving style recognition model includes: Z1. Use SVM algorithm to identify driving style. Take driving style features as input information, reasonably set support vector machine kernel function, construct an optimal hyperplane, correctly separate input information, and obtain driving style classification model. Z2. By using Bayesian optimization to search for the best hyperparameter combination of SVM, first use the cross-validation method to preliminarily screen the hyperparameters to find the hyperparameter range with better performance, and then use Bayesian optimization to perform a more refined search within this range to establish a driving style recognition model for the improved SVM algorithm; Z3. Using the test set as input, the driving style recognition accuracy training of the improved SVM algorithm is carried out to obtain a driving style recognition model based on the improved SVM.
[0011] Furthermore, the driving style identified by the driving style recognition model is combined with the local representative working condition identified by the working condition recognition model, and according to the energy consumption and power requirements, a database of required optimization parameters is created and classified in a mode in which one driving style corresponds to all working condition types.
[0012] Furthermore, the parameter correction module is used to determine whether the control parameter test results meet the calibration targets. If the control parameter test results do not meet the requirements, the parameter correction module needs to continue to optimize and generate new control parameters until the control parameter test results meet the calibration targets and obtain qualified control parameters.
[0013] The calibration parameter setting module is used to store the generated new control parameters in the vehicle controller, and calibrate the optimal control parameters finally obtained into the vehicle controller; the database module is used to store the test data transmitted during the test, and provide the parameter correction module with the test data required to be called; the operating condition calling module is used to identify a new set of parameters output by the control parameter total library, and call the stored corresponding operating condition information to input it into the test bench equipment.
[0014] Furthermore, the control parameter total database outputs each set of control parameters into the calibration parameter setting module in sequence through an automatic optimization calibration process. After each set of parameter optimization and calibration is completed, the next set of parameters is adjusted to continue, and the optimal control parameter total database is obtained through experimental verification.
[0015] The beneficial effects of the present invention are as follows: compared with the existing conventional automobile calibration methods, the automatic optimization calibration method for new energy vehicles provided by the present invention can pre-process the overall data of the samples measured by the actual vehicle, perform cluster optimization to obtain various local representative working conditions, and use the actual vehicle test data to train the working condition recognition module and the driving style recognition module, combine the different working conditions obtained by recognition with the driving style, and obtain the vehicle total control parameter database, and then optimize and update the total parameter library of the control parameters through the parameter correction module, calibration parameter setting module, database module, and working condition calling module in the automatic optimization calibration method, thereby solving the problem of poor adaptability of the calibration parameters under the influence of differentiated driving conditions and driving styles during actual driving, realizing deep optimization of the calibration parameters, enhancing the adaptability of the parameters to a changing environment, and greatly optimizing the energy consumption of new energy vehicles. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a schematic diagram of the automatic optimization calibration method for new energy vehicles of the present invention; Figure 2 It is a parameter optimization calibration flow chart of the automatic optimization calibration method for new energy vehicles of the present invention; Figure 3 It is a comprehensive operating condition diagram of the automatic optimization calibration method of new energy vehicles of the present invention; Figure 4 It is a pedal opening diagram of the automatic optimization calibration method for new energy vehicles of the present invention; Figure 5 It is a parameter update command flow chart of the automatic optimization calibration method for new energy vehicles of the present invention; Figure 6 It is a Bi-LSTM neural network diagram of the automatic optimization calibration method of new energy vehicles of the present invention; Figure 7 It is an improved SVM diagram of the automatic optimization calibration method of new energy vehicles of the present invention; Figure 8It is a data processing principle diagram of the automatic optimization calibration method for new energy vehicles of the present invention; DETAILED DESCRIPTION
[0017] Specific embodiment 1: The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. It should be noted that: in the present invention, if there is no special explanation, all the implementation methods and preferred implementation methods mentioned in this article can be combined with each other to form a new technical solution. In the present invention, if there is no special explanation, all the technical features and preferred features mentioned in this article can be combined with each other to form a new technical solution. Unless otherwise specified, the professional and scientific terms used in this article have the same meaning as those familiar to those skilled in the art.
[0018] As the specification of the present invention Figure 1 , Instruction Manual Figure 2 As shown, the present invention provides a new energy vehicle automatic optimization calibration method, which specifically includes the following steps: Collecting actual vehicle driving condition data through the vehicle terminal device, the actual vehicle driving condition data may be speed, acceleration, GPS location information, etc.; The working condition data is preprocessed, divided into working condition segments according to the characteristics of the working condition data, and characteristic parameters are selected. In view of the large number of short trips, the principal component analysis method is used to extract data with a cumulative contribution rate greater than 80% for analysis to achieve the purpose of dimensionality reduction and eliminate redundant information; A three-dimensional scatter plot is drawn based on the characteristic parameters, and then the genetic algorithm is used to optimize the initial cluster center. According to the results of cluster optimization, the data is divided into local representative working conditions. In addition, referring to foreign standards, the time proportion and time length of several different representative working conditions in a certain city working condition are obtained by using the time proportion of each segment in the total duration. A Bi-LSTM neural network is constructed and trained with sample data to obtain an operating condition recognition model. A classification calculation is completed according to the time input sequence during the forward propagation process, which includes steps such as operating condition data feature extraction and operating condition type judgment. In the back propagation, the network weight coefficient and bias and other parameters are optimized by accumulating classification errors. When the time series input layer is extracted from the operating condition data during the forward propagation process, the historical operating condition data needs to be decomposed into multiple short-trip segments. The first few principal components with the highest contribution obtained by the principal component analysis method are used as the input variables of the identification model. Then, several numbers of neurons are set according to several operating condition categories. In the next step, the output discrete features are converted into probability distributions of corresponding categories.
[0019] The driving style recognition model is constructed by using the improved SVM recognition method and combining it with the driving style recognition method. The classification decision function constructed after nonlinear mapping is as follows: Where x is the driving style feature; y i is the driving style type; i is the sample number of the training set; j is the sample number of the test set; α * is the Lagrange multiplier; K is the kernel function; sgn is the sign function; At this time, the nonlinear classification optimization problem is as follows: Solving the above optimization problem can obtain the optimal classification hyperplane, and the relevant parameters of the support vector machine kernel function are set to ensure the classification accuracy. Then, it is necessary to select appropriate input information. The driving behavior-related information can use the pedal information as the main basis for judging the driving style. The pedal information corresponding to different types of driving styles can be obtained through simulation.
[0020] The improved SVM algorithm mainly searches for the best hyperparameter combination of SVM by using Bayesian optimization. The specific operation process is as follows: first, define the hyperparameter search space, select the hyperparameters of the optimized Gaussian kernel function, define the value range, and integrate the Gaussian kernel function with the hyperparameter combination that needs to be optimized. Select the cross-validation accuracy, divide the data set into a training set and a test set, use the working condition sample data to train the model, and predict and calculate the accuracy in the test set. Then, use Gaussian process regression to establish an optimization model for hyperparameter search. According to the previous search results and the evaluation of the objective function, infer the next set of optimal hyperparameter combinations. When the number of iterations or the convergence threshold of the objective function is reached, the optimal hyperparameter combination is obtained and the result is output.
[0021] Combining the identification results of the working condition identification model with the driving style identification model, multiple groups of different working conditions can be obtained, including aggressive urban congestion conditions, aggressive urban unobstructed conditions, aggressive expressway conditions, aggressive suburban conditions, cautious urban congestion conditions, cautious urban unobstructed conditions, cautious expressway conditions, cautious suburban conditions, standard urban congestion conditions, standard urban unobstructed conditions, standard suburban conditions, standard expressway conditions, etc. According to multiple groups of working condition categories, a total control parameter library is established by considering the control parameters to be optimized. More local representative working conditions can be selected according to different cities.
[0022] The parameter optimization and calibration module is designed based on the total parameter library of the obtained control parameters, which specifically includes a parameter correction module, which is used to replace the manual continuous adjustment of the control parameters to make them quickly approach the optimal control parameters. The principle is as shown in the attached manual. Figure 2As shown. At the beginning, it is necessary to manually set the initial value and value range of the control parameters that need to be optimized, energy consumption indicators, dynamic constraints and other parameter optimization conditions. The parameter correction module sets the first set of test control parameters as the set initial values. The initial values can be determined by the calibration engineer based on calibration experience, which can provide an approximate optimization area for the optimization algorithm and improve the efficiency of parameter calibration. The parameters that have passed the bench test are judged to see whether they meet the above-set parameter optimization conditions. If not, the parameter correction module adjusts the control parameters to generate a new set of control parameters for the calibration parameter setting module. The control parameters that meet the conditions are saved with the energy consumption values and compared with the subsequent parameter values that also meet the conditions. The optimal parameters are selected for update. Finally, the parameters are compared with the calibration end conditions. If they are not met, the control parameters continue to be generated. If they are met, the optimization ends.
[0023] Calibration parameter setting module: used for online calibration of control parameters. It replaces manual automatic parameter calibration. First, it detects whether there is a control parameter update instruction. If so, it updates the control parameters to the memory of the vehicle controller. As the optimization proceeds, the vehicle controller is always ready to update the optimal control parameters. Database module: used to classify and store the test data collected during the test. The test data includes test results and control parameters. All data can be called arbitrarily by the parameter correction module. The database module can save all test results and allow new test results to overwrite the original results. Working condition calling module: Every time the total parameter library outputs a set of parameters, the working condition calling module will identify and classify them, and automatically load the corresponding working conditions that meet this set of parameters to the test bench. Every time the optimization calibration of a set of parameters is completed, the working condition calling module will identify the new parameters and automatically load the corresponding working conditions; Finally, through the calibration parameter setting module, each working condition in the total parameter library of the control parameters is optimized and calibrated step by step corresponding to the driving style, and the parameters completed by the optimization calibration are updated to the total parameter library in real time. By automatically optimizing the calibration process, the calibration time and calibration effect are greatly saved, and the optimal control parameter library is obtained, thereby greatly developing the energy-saving potential of new energy vehicles in the calibration stage, and fully increasing the adaptability of the control parameters to differentiated driving conditions and driving styles, thereby achieving the purpose of energy saving and emission reduction. The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical content disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still belongs to the scope of the technical solution of the present invention.
Claims
1. A new energy vehicle automatic optimization calibration method, characterized in that: The following steps are involved: S1. Select characteristic parameters according to the characteristics of the working condition data, wherein the characteristic parameters include average vehicle speed, proportion of time when the speed is greater than 30 km / h, and maximum acceleration; S2. Local representative working conditions are obtained by clustering and optimizing the extracted characteristic parameters, and comprehensive cycle working conditions are obtained by combining the local representative working conditions. A real-time working condition recognition model and a driving style recognition model are respectively constructed according to the obtained working condition information; S3, obtaining n groups of vehicle control parameters according to the combination of different local representative working conditions and multiple driving style types, and then constructing a total parameter library of control parameters; S4. According to the designed automatic optimization calibration method, determine the minimum energy consumption of the whole vehicle as the goal and the power performance as the constraint condition, and establish an automatic optimization calibration model; the automatic optimization calibration method includes a parameter correction module, a calibration parameter setting module, a database module, and a working condition calling module. S5. Combine the test process with the optimization calculation, organize the optimization experiment according to the optimization goal, perform the optimization calculation online, automatically determine a set of optimal control parameters as the new parameters for the next test, and continuously iterate according to the optimization goal until the optimal solution is obtained or the algorithm requirements are met, and complete the automatic calibration of the total parameter library of control parameters one by one.
2. The automatic optimization calibration method for new energy vehicles according to claim 1 is characterized in that: The operating condition data is obtained from actual vehicle testing of new energy buses. Before extracting characteristic parameters, the operating condition sample data needs to be divided into short-trip segments of a specified time, and the principal component analysis method is used to extract the main features of the short trip to obtain several groups of information with high contribution.
3. The automatic optimization calibration method for new energy vehicles according to claim 1 is characterized in that: In step S2, a three-dimensional scatter plot of the operating condition block is drawn according to the characteristic parameters, and the initial cluster center is optimized by a genetic algorithm to obtain the initial cluster center after genetic optimization, and then the local representative operating conditions are obtained through the results of cluster analysis. The local representative operating conditions include unimpeded urban conditions, congested urban conditions, unimpeded suburban conditions, and expressway conditions.
4. The automatic optimization calibration method for new energy vehicles according to claim 3 is characterized in that: The duration of the typical urban working condition is taken as about 1600 seconds. The time proportion of each representative working condition in the entire recorded data can be used to determine its time in the final constructed working condition and obtain the comprehensive cycle working condition.
5. The automatic optimization calibration method for new energy vehicles according to claim 1 is characterized in that: The specific operation of the real-time working condition identification model includes the following steps: R1: Construct a Bi-LSTM neural network, and use the short-trip principal components obtained by principal component analysis as the input variables of the identification model. The short-trip principal components in each sequence are arranged in order to obtain the time series input layer. The number of neurons in the hidden layer is set according to the number of working condition categories. The loss function is calculated in the form of mean square error during the back propagation process, and the probability value of each working condition category is used as the output; R2. Take 90% of the short-stroke samples as training data, perform deep learning training on the neural network, and obtain a real-time working condition recognition model based on the Bi-LSTM neural network.
6. The automatic optimization calibration method for new energy vehicles according to claim 5 is characterized in that: The specific operation arrangement of the driving style recognition model includes: Z1. Use SVM algorithm to identify driving style. Take driving style features as input information, reasonably set support vector machine kernel function, construct an optimal hyperplane, correctly separate input information, and obtain driving style classification model. Z2. By using Bayesian optimization to search for the best hyperparameter combination of SVM, first use the cross-validation method to preliminarily screen the hyperparameters to find the hyperparameter range with better performance, and then use Bayesian optimization to perform a more refined search within this range to establish a driving style recognition model for the improved SVM algorithm; Z3. Using the test set as input, the driving style recognition accuracy training of the improved SVM algorithm is carried out to obtain a driving style recognition model based on the improved SVM.
7. The automatic optimization calibration method for new energy vehicles according to claim 6 is characterized in that: The driving style identified by the driving style recognition model is combined with the local representative working condition identified by the working condition recognition model. According to the energy consumption and power requirements, a database of required optimization parameters is created and classified in a mode in which one driving style corresponds to all working condition types.
8. The automatic optimization calibration method for new energy vehicles according to claim 1 is characterized in that: The parameter correction module is used to determine whether the control parameter test results meet the calibration targets. If the control parameter test results do not meet the requirements, the parameter correction module needs to continue to optimize and generate new control parameters until the control parameter test results meet the calibration targets and obtain qualified control parameters.
9. The automatic optimization calibration method for new energy vehicles according to claim 8, characterized in that: The calibration parameter setting module is used to store the generated new control parameters in the vehicle controller, and calibrate the optimal control parameters finally obtained into the vehicle controller; the database module is used to store the test data transmitted during the test, and provide the parameter correction module with the test data required to be called; the operating condition calling module is used to identify a new set of parameters output by the control parameter total library, and call the stored corresponding operating condition information to input it into the test bench equipment.
10. The automatic optimization calibration method for new energy vehicles according to claim 9, characterized in that: The control parameter total database outputs each group of control parameters into the calibration parameter setting module in sequence through the automatic optimization calibration process. When each group of parameter optimization calibration is completed, the next group of parameters is adjusted to continue, and the optimal control parameter total database is obtained through experimental verification.