Method and device for determining vehicle control parameter, vehicle and storage medium

Abstract

The application discloses a kind of determination method of vehicle control parameter and its device, vehicle and storage medium. Determination method includes: obtaining the current time data and current space data of vehicle;Current time data and current space data of vehicle are input into the preset model obtained by preset algorithm training from the historical path data and historical driving style data of vehicle, and the vehicle control parameter of vehicle in future driving path is determined.The preset model is obtained by prediction algorithm training from historical path data and historical driving style data in the application, and current time data and current space data of vehicle are input into the preset model, the future driving path of vehicle and the vehicle control parameter of vehicle in future driving path can be more accurately and simply predicted, so as to effectively improve vehicle performance, driving experience and safety factor, improve driving experience and safety.

Description

Technical Field

[0001] This invention relates to the field of vehicle technology, and in particular to a method and apparatus for determining vehicle control parameters, a vehicle, and a storage medium. Background Technology

[0002] Currently, there are two common methods for calibrating vehicle control parameters:

[0003] (1) Adaptive calibration method for vehicle control parameters. Specifically, the vehicle controller constructs a calibration equation between the analog signal parameters and the vehicle control parameters through a Kalman filter prediction model, and calibrates the vehicle control parameters based on the observed driving data, vehicle end mass and slope value.

[0004] (2) A vehicle control parameter calibration method based on multi-objective particle swarm optimization algorithm. Specifically, firstly, the preset control model of the vehicle control system to be calibrated is obtained, the target output conditions and boundary conditions of each subsystem are determined to obtain the range of values ​​of the parameters to be calibrated in the preset control model, and then the global optimal solution of the calibration parameters is obtained according to the multi-objective particle swarm optimization algorithm and boundary conditions. If the target output conditions are met, it is the output of the optimal vehicle control parameters.

[0005] Among them, the Kalman filter prediction model only performs well for linear models, but the real model is generally nonlinear and the input signal cannot have too many dimensions. Therefore, the vehicle control parameters calibrated by the adaptive calibration method of vehicle control parameters are not accurate enough.

[0006] In addition, although the multi-objective particle swarm optimization algorithm can obtain better vehicle control parameters, the algorithm has drawbacks such as high computational complexity and poor convergence. Summary of the Invention

[0007] This invention provides a method and apparatus for determining vehicle control parameters, as well as a vehicle and a storage medium.

[0008] This invention provides a method for determining vehicle control parameters. The method includes: acquiring the vehicle's current time data and current spatial data; inputting the vehicle's current time data and current spatial data into a preset model trained by a preset algorithm using the vehicle's historical path data and historical driving style data, to determine the vehicle control parameters for the vehicle's future driving path.

[0009] The determination method includes: acquiring historical path data and historical driving style data from historical paths of the vehicle under different scenarios as a training dataset; and training the training dataset based on the preset algorithm to obtain the preset model.

[0010] The preset algorithm is a random forest prediction algorithm, and the preset model is a random forest model. The step of training the training dataset based on the preset algorithm to obtain the preset model includes: setting model parameters, including the number of decision trees, where there are multiple decision trees; randomly selecting historical path data and historical driving style data of the vehicle in a preset scenario as training samples; training multiple decision trees according to the multiple randomly selected training samples, and outputting multiple vehicle control parameters corresponding to the multiple decision trees; taking the average of the multiple vehicle control parameters to output the vehicle control parameters corresponding to the preset scenario, thereby training the random forest model.

[0011] After training each decision tree based on the training samples and outputting multiple vehicle control parameters corresponding to multiple decision trees, the determination method includes: using a breadth-first search algorithm to evaluate and predict the multiple vehicle control parameters to output optimized vehicle control parameters, so as to train the random forest model.

[0012] The preset algorithm is a backpropagation neural network algorithm, and the preset model is a neural network model. Training the training dataset based on the preset algorithm to obtain the preset model includes: preprocessing the historical path data and historical driving style data in the training dataset; determining the neural network topology and the connection weights between neurons in the neural network; inputting the preprocessed historical path data and historical driving style data into the neural network topology to calculate a first output value; calculating the error between the first output value and the true value, where the true value is the actual vehicle control parameter of the path corresponding to the first output value in the historical path data and historical driving style data; if the error is greater than or equal to a set value, updating the model parameters of the neural network model, including the connection weights; adjusting the neural network topology according to the updated model parameters, calculating a second output value, until the error between the second output value and the true value is less than the set value, thus obtaining the trained neural network model.

[0013] The determination method includes: inputting the vehicle's current time data and current spatial data into the neural network model to predict the vehicle's future driving path, and determining the vehicle control parameters of the vehicle on the future driving path.

[0014] The preset algorithm is a random forest prediction algorithm and a backpropagation neural network algorithm. The step of training the training dataset based on the preset algorithm to obtain the preset model includes: using the random forest prediction algorithm to classify or regress the historical path data and the historical driving style data to obtain preliminary prediction results; and adjusting the weights of the preliminary prediction results based on the backpropagation neural network algorithm to train the preset model.

[0015] The present invention also provides a device for determining vehicle control parameters. The device includes an acquisition module and a parameter determination module. The acquisition module acquires the current time data and current spatial data of the vehicle; the parameter determination module inputs the current time data and current spatial data of the vehicle into a preset model trained by a preset algorithm using the vehicle's historical path data and historical driving style data, to determine the vehicle control parameters for the vehicle's future driving path.

[0016] The present invention also provides a vehicle. The vehicle includes the vehicle control parameter determination device described in the above embodiments.

[0017] The present invention also provides a non-volatile computer-readable storage medium containing a computer program. When the computer program is executed by one or more processors, it implements the method for determining vehicle control parameters as described in any of the above embodiments.

[0018] Thus, this invention trains a preset model using a prediction algorithm on historical path data and historical driving style data, and inputs the vehicle's current time data and current spatial data into the preset model. This enables a more accurate and simple prediction of the vehicle's future driving path and the vehicle control parameters for that path, thereby effectively improving vehicle performance, driving experience, and safety.

[0019] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, wherein:

[0021] Figure 1 This is one of the flowcharts illustrating the method for determining vehicle control parameters according to the present invention;

[0022] Figure 2 This is one of the structural schematic diagrams of the vehicle control parameter determination device of the present invention;

[0023] Figure 3This is a second flowchart illustrating the method for determining vehicle control parameters according to the present invention.

[0024] Figure 4 This is a second schematic diagram of the vehicle control parameter determination device of the present invention;

[0025] Figure 5 This is the third flowchart illustrating the method for determining vehicle control parameters according to the present invention;

[0026] Figure 6 This is a schematic diagram of the process of modeling using the random forest prediction algorithm in the method for determining vehicle control parameters of the present invention;

[0027] Figure 7 This is the fourth flowchart illustrating the method for determining vehicle control parameters according to the present invention;

[0028] Figure 8 This is a schematic diagram of the process of modeling using the backpropagation neural network algorithm in the method for determining vehicle control parameters of the present invention;

[0029] Figure 9 This is the fifth flowchart illustrating the method for determining vehicle control parameters according to the present invention. Detailed Implementation

[0030] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the embodiments of the present invention, and should not be construed as limiting the embodiments of the present invention.

[0031] Please see Figure 1 This invention provides a method for determining vehicle control parameters. The method for determining vehicle control parameters includes:

[0032] 01: Obtain the vehicle's current time and location data;

[0033] 02: Input the vehicle's current time data and current spatial data into a preset model trained by a preset algorithm based on the vehicle's historical path data and historical driving style data to determine the vehicle control parameters for the future driving path.

[0034] Please see Figure 2 The present invention also provides a vehicle control parameter determination device 10. The determination device 10 includes an acquisition module 11 and a parameter determination module 12.

[0035] Step 01 can be implemented by the acquisition module 11, and step 02 can be implemented by the parameter determination module 12. That is, the acquisition module 11 is used to acquire the vehicle's current time data and current spatial data; the parameter determination module 12 is used to input the vehicle's current time data and current spatial data into a preset model trained by a preset algorithm based on the vehicle's historical path data and historical driving style data, and to determine the vehicle control parameters for the vehicle's future driving path.

[0036] Specifically, the vehicle's current time data refers to the time information corresponding to the vehicle's current travel path. For example, the time information can be summer or winter, rush hour, and date. The vehicle's current spatial data refers to the spatial information corresponding to the vehicle's current travel path. For example, the spatial information can be the road network structure information, geographical feature information, and traffic flow distribution information of the current travel path.

[0037] The vehicle control parameters of this invention include parameters such as energy recovery intensity, suspension height and stiffness, steering assist level, power performance and crawl speed.

[0038] The vehicle's future travel path refers to the route the vehicle will take within a preset range in the future. This preset range can refer to a preset mileage range, a preset time range, or other ranges; no restrictions are placed here.

[0039] Understandably, this invention first considers the impact of time factors on route prediction, predicting the future travel route corresponding to the current time based on time data in historical route data, and then adjusting vehicle control parameters. That is, historical route data includes time data, which specifically may include historical route time distribution data, traffic congestion time pattern data, weather and seasonal variation data, or other data, without limitation. The historical route time distribution can analyze the distribution of historical route data throughout the day, such as peak hours and off-peak hours, to understand the differences in route traffic flow at different times. Traffic congestion time pattern data can be used to analyze the time patterns of traffic congestion based on historical data, such as morning peak hours and evening peak hours, to predict the congestion situation of future travel routes. Weather and seasonal variation data can consider the impact of weather and seasonal changes on the route; for example, rain and snow may cause slippery roads, while winter may lead to icy roads.

[0040] Thus, by taking time into account, the present invention can more accurately predict the future driving path corresponding to the current time data based on the time data in historical path data, and adjust the vehicle control parameters accordingly.

[0041] Secondly, this invention considers the impact of spatial factors on path prediction. It predicts the future driving path corresponding to the current spatial data based on spatial data from historical path data, and then adjusts the vehicle control parameters accordingly. That is, historical path data includes spatial data, which specifically may include road network structure data, geographic feature data, and traffic flow distribution data, or other data, without limitation. Road network structure data can analyze the structure and connectivity of the road network, understanding the distances between different roads, traffic conditions, and traffic rules. Geographic feature data can consider the impact of geographic features on the path, such as hills, rivers, and bridges, to predict the difficulty and risk of future driving paths. Traffic flow distribution data can analyze the traffic flow distribution of historical path data on different road segments, understanding the congestion level and capacity of different road segments.

[0042] Thus, by taking spatial dimension factors into account, the present invention can more accurately predict the future driving path corresponding to the current spatial data based on the spatial data in historical path data, and adjust the vehicle control parameters accordingly.

[0043] In summary, temporal data and spatial data are crucial factors to consider when performing route prediction. By combining temporal and spatial data from historical route data, we can more accurately predict the future driving path corresponding to the current temporal and spatial data, and provide drivers with more reasonable driving suggestions and vehicle control parameter settings.

[0044] In addition, historical driving style data includes the driver's behavior and habits during driving, such as the frequency and manner of acceleration, deceleration, and turning, as well as the driver's vehicle control methods and driving decisions. This historical driving style data can be collected and recorded through vehicle sensors, driver assistance systems, and other devices. By combining the driver's historical driving style data, it is possible to more accurately predict future driving paths and provide the driver with more reasonable driving suggestions and vehicle control parameter settings.

[0045] Therefore, this invention can input the vehicle's current time and spatial data into a preset model trained using a preset algorithm based on the vehicle's historical path data and historical driving style data to determine the vehicle's control parameters for future driving paths. In other words, this invention can first train a preset model using a prediction algorithm on historical path data and historical driving style data, and then use this preset model to predict paths within a preset range to obtain the vehicle's future driving path and acquire the vehicle control parameters for that future path, thereby improving driving experience and safety.

[0046] This invention strongly binds vehicle control parameters to the path, and statistically learns from historical path data and historical driving style data to obtain information such as road surface roughness, GPS signal, vehicle attitude, and intersection location. Then, based on current time and spatial data, the path is predicted within a certain range to predict the vehicle's future travel path within a certain time or spatial range. Next, by matching the predicted future travel path with historical paths, information such as road surface roughness, GPS signal, vehicle attitude, and intersection location corresponding to the future travel path can be obtained, thus obtaining each control parameter of the current vehicle, and consequently, the overall vehicle control parameters.

[0047] The degree of road bumpiness reflects the quality and smoothness of the road surface. By monitoring the degree of road bumpiness, vehicle control parameters such as driving speed, acceleration, and braking can be adjusted to reduce the impact on the vehicle and passengers, thereby improving driving comfort and safety.

[0048] GPS signals provide a vehicle's precise location information. Through GPS signals, a vehicle can obtain information such as its current latitude, longitude, and altitude, thus aiding in route planning and navigation. This is crucial for setting vehicle control parameters such as steering, acceleration, and braking, ensuring the vehicle stays on the correct path and avoids deviations or collisions.

[0049] Vehicle attitude reflects information such as the vehicle's tilt angle and yaw rate. By monitoring vehicle attitude, the vehicle can adjust its driving posture to maintain a stable and safe driving state. This provides guidance for setting vehicle control parameters such as steering control and stability control, and can optimize the vehicle's driving stability and handling performance.

[0050] Intersection location information helps vehicles identify the location and type of road intersections, enabling route planning and navigation. This is crucial for setting vehicle control parameters such as steering and speed control, ensuring safe and smooth vehicle movement at intersections and preventing traffic congestion or accidents.

[0051] In summary, this invention strongly binds vehicle control parameter settings to the path and performs statistical learning on historical path data and historical driving style data to obtain information such as road bumpiness, GPS signal, vehicle posture, and intersection location. This can help the vehicle better adapt to road conditions and traffic conditions, optimize vehicle control parameter settings, and improve driving comfort and safety.

[0052] Thus, this invention trains a preset model using a prediction algorithm on historical path data and historical driving style data, and inputs the vehicle's current time data and current spatial data into the preset model. This enables a more accurate and simple prediction of the vehicle's future driving path and the vehicle control parameters for that path, thereby effectively improving vehicle performance, driving experience, and safety.

[0053] Please see Figure 3 In some implementations, the determination method includes:

[0054] 03: Obtain historical path data and historical driving style data of the vehicle under different scenarios as training datasets;

[0055] 04: The preset model is obtained by training the training dataset based on the preset algorithm.

[0056] Please see Figure 4 The vehicle control parameter determination device 10 also includes a modeling module 13.

[0057] Steps 03 and 04 can be implemented by modeling module 13. That is, modeling module 13 is used to acquire historical path data and historical driving style data of the vehicle in different scenarios as training datasets; and to train the training datasets based on a preset algorithm to obtain a preset model.

[0058] Specifically, since different vehicle control parameters can be adjusted based on historical path data and historical driving style data under different vehicle scenarios, this invention can obtain historical path data and historical driving style data under different vehicle scenarios as a training dataset, and train a preset model on the training dataset based on a preset algorithm.

[0059] For example, the scenarios in which the present invention sets vehicle control parameters based on paths include, but are not limited to, the following scenarios:

[0060] (1) The scenario is that the road ahead is an intersection. In this scenario, the vehicle control parameters of the energy recovery intensity can be adjusted and controlled according to the distance to the intersection ahead and the deceleration pattern of the historical driving style, so as to reduce the vehicle speed and ensure driving safety.

[0061] (2) The scenario is that the road surface ahead is relatively uneven. In this scenario, the vehicle control parameters such as energy recovery intensity and suspension height stiffness can be adjusted and controlled according to the deceleration pattern of historical path data and driving style data to ensure vehicle passability and driving comfort.

[0062] (3) The scenario is that the slope or curvature of the road ahead is large. In this scenario, the vehicle's power parameters, Electronic Stability Program (ESP) mode, and steering assist mode vehicle control parameters can be adjusted and controlled based on the road slope and curvature calculated from historical path data to ensure the vehicle's power and driving comfort.

[0063] (4) The scenario is that the road ahead is either clear or congested. In this scenario, the road can be judged to be clear or congested based on historical path data such as vehicle speed, deceleration, time, and distance to the vehicle in front. The vehicle control parameters of the vehicle's crawl speed can be adjusted and controlled to reduce the driving difficulty in congested road conditions and improve driving comfort.

[0064] Thus, this invention uses historical path data and historical driving style data in various scenarios as training datasets, and trains the training datasets based on a preset algorithm to obtain a preset model. This allows the invention to analyze the vehicle's performance under different road conditions based on historical path data and historical driving style data, thereby adjusting the corresponding vehicle control parameters such as the vehicle's suspension system, steering system, and braking system to obtain a better driving experience and a higher safety factor.

[0065] Please see Figure 5 In some implementations, the preset algorithm is a random forest prediction algorithm, the preset model is a random forest model, and step 04 includes:

[0066] 041: Set model parameters, including the number of decision trees, which can be multiple;

[0067] 042: Randomly select historical path data and historical driving style data of vehicles in preset scenarios as training samples;

[0068] 043: Train multiple decision trees based on multiple randomly selected training samples, and output multiple vehicle control parameters corresponding to the multiple decision trees;

[0069] 044: Take the average value of multiple vehicle control parameters to output the vehicle control parameters corresponding to the preset scenario, so as to train a random forest model.

[0070] Please see Figure 4Steps 041, 042, 043, and 044 can be implemented by modeling module 13. That is, modeling module 13 is used to set model parameters, including the number of decision trees (multiple decision trees); randomly select historical path data and historical driving style data of the vehicle in a preset scenario as training samples; train multiple decision trees based on the multiple randomly selected training samples, outputting multiple vehicle control parameters corresponding to the multiple decision trees; and average the multiple vehicle control parameters to output vehicle control parameters corresponding to the preset scenario, thereby training a random forest model.

[0071] Specifically, the preset scenarios of the present invention include, but are not limited to, the scenarios described above, such as the road ahead being an intersection, the road surface being uneven, the road ahead having a large slope or path curvature, and the road ahead being either unobstructed or congested.

[0072] Please combine Figure 6 First, to build a random forest model, you can set the model parameters, including the number of decision trees (multiple decision trees are allowed). For example... Figure 6 The diagram shows m decision trees, where m is greater than 1. Figure 6 The training dataset in the code is the dataset used to train the model. In machine learning, the model builds a predictive model by learning the input-output relationships in the training dataset. The training dataset contains known inputs and outputs, and the model learns from this data to build a predictive model. Figure 6 The test set is a dataset used to test the model's performance. The test set contains input data that the model has never seen before and known output data; the model uses these to predict data and validate its accuracy. The main purpose of the test set is to evaluate the model's generalization ability, that is, its performance on unseen data. The training set is used to train the model, while the test set is used to evaluate its performance.

[0073] In addition, the model parameters of the present invention may also include parameters such as the maximum feature value, the minimum number of samples, and the maximum depth of the tree.

[0074] Then, historical path data and historical driving style data of the vehicle under a preset scenario are randomly selected as training samples, i.e., random sampling. Multiple decision trees are trained based on multiple randomly selected training samples, and multiple vehicle control parameters corresponding to multiple decision trees are output. The average of multiple vehicle control parameters is taken to output the vehicle control parameters corresponding to the preset scenario, thus training a random forest model.

[0075] In other words, when constructing the decision tree, this invention first randomly selects a portion of samples with replacement from historical path data and historical driving style data. Similarly, it randomly selects a portion of features from all feature values, such as historical road slope features, path curvature, road surface unevenness, GPS signal, vehicle attitude signal, current vehicle speed, intersection distance, distance to the vehicle in front, deceleration, and current time. That is, each decision tree is randomly sampled from the training dataset for model training, and finally, the average output of all decision trees is taken as the output vehicle control parameters corresponding to the preset scenario.

[0076] Understandably, since each decision tree is trained by randomly sampling from the dataset, the model parameters of each tree are different. Therefore, when a new test data is input, the prediction results of each tree will be different. By averaging the prediction results of multiple trees, a relatively robust and accurate prediction result can be obtained.

[0077] Furthermore, each decision tree in a random forest is independent and unaffected by other trees. This helps avoid overfitting, which occurs when a model performs well on training data but poorly on test data.

[0078] Furthermore, since each decision tree is trained using random sampling, each tree is likely to capture different data features and patterns. By averaging these features and patterns, a more accurate and comprehensive prediction result can be obtained. Because each decision tree is trained using random sampling, computational resources and time consumption can be reduced. At the same time, each tree in a random forest can be trained in parallel, further improving computational efficiency.

[0079] Thus, by constructing multiple decision trees and training each decision tree through random sampling to obtain a random forest model, this invention can improve the robustness and accuracy of the trained random forest model, prevent overfitting, reduce computational resource consumption, and improve computational efficiency.

[0080] In some implementations, after step 043, the determination method includes:

[0081] 045: A breadth-first search algorithm is used to evaluate and predict multiple vehicle control parameters to output optimized vehicle control parameters, which are then used to train a random forest model.

[0082] Please see Figure 4 Step 045 can be implemented by modeling module 13. That is, modeling module 13 is used to evaluate and predict multiple vehicle control parameters using a breadth-first search algorithm to output optimized vehicle control parameters, so as to train a random forest model.

[0083] In other words, the present invention can use a prediction algorithm that combines random forest algorithm and breadth-first search algorithm to train the model and obtain the final random forest model.

[0084] Breadth-First Search (BFS) is an algorithm for traversing or searching a tree or graph. This algorithm starts with the root node (or any node in the case of a graph), explores the nearest nodes, and then moves on to their unvisited neighbors. This process continues until the necessary information is found, or all possible nodes have been checked.

[0085] In prediction algorithms that combine random forests and breadth-first search, the role of breadth-first search can be summarized as follows:

[0086] (1) Feature selection: In the feature space, breadth-first search can be used to select the features most relevant to the target variable. This helps reduce the number of features and improves the efficiency and accuracy of the model.

[0087] (2) Searching the decision tree: In a random forest, each decision tree may have many nodes. Breadth-first search can be used to guide the search process, starting from the root node, exploring all child nodes of each node, and then moving on to the next unvisited node.

[0088] (3) Model optimization: Breadth-first search can be used to optimize model parameters, such as learning rate and number of iterations. This helps to find the optimal model configuration, thereby improving model performance.

[0089] Understandably, since breadth-first search can quickly explore a large number of possibilities, and random forest can use this information to generate more accurate and stable predictions, the prediction method of combining random forest algorithm and breadth-first search algorithm in this invention may help improve the prediction accuracy and efficiency of the trained random forest model.

[0090] Therefore, the breadth-first search algorithm can be used to search for multiple vehicle control parameters output by multiple decision trees. Then, during the search process, a prediction algorithm is used to evaluate and predict the merits of each vehicle control parameter, so as to quickly find the optimal or near-optimal vehicle control parameters.

[0091] Thus, this invention can quickly find the optimal or near-optimal vehicle control parameters through a breadth-first search algorithm, thereby improving the prediction accuracy and efficiency of the trained random forest model.

[0092] In addition, the random forest prediction algorithm used in the above embodiments of the present invention for setting path-based vehicle control parameters can be replaced by other prediction algorithms to train historical path data and historical driving style data, such as backpropagation (BP) neural network algorithm, regression analysis prediction algorithm and Markov chain model, etc.

[0093] This explanation uses the backpropagation neural network algorithm as an example; please refer to [link / reference]. Figure 7 In some implementations, the preset algorithm is a backpropagation neural network algorithm, the preset model is a neural network model, and step 04 further includes:

[0094] 046: Preprocess historical path data and historical driving style data in the training dataset;

[0095] 047: Determine the topology of the neural network and the connection weights between the neurons in the neural network;

[0096] 048: Input the preprocessed historical path data and historical driving style data into the neural network topology and calculate the first output value;

[0097] 049: Calculate the error between the first output value and the true value, where the true value is the actual vehicle control parameter of the path corresponding to the first output value in the historical path data and historical driving style data; if the error is greater than or equal to the set value, update the model parameters of the neural network model, including the connection weight values; adjust the neural network topology according to the updated model parameters, calculate the second output value, until the error between the second output value and the true value is less than the set value, and obtain the trained neural network model.

[0098] Please see Figure 4 Steps 046, 047, 048, and 049 can be implemented by modeling module 13. That is, modeling module 14 is used to preprocess historical path data and historical driving style data in the training dataset; determine the neural network topology and the connection weights between neurons in the neural network; input the preprocessed historical path data and historical driving style data into the neural network topology to calculate the first output value; calculate the error between the first output value and the true value, where the true value is the actual vehicle control parameter of the path corresponding to the first output value in the historical path data and historical driving style data; if the error is greater than or equal to a set value, update the model parameters of the neural network model, including the connection weights; adjust the neural network topology according to the updated model parameters, calculate the second output value, until the error between the second output value and the true value is less than the set value, thus obtaining the trained neural network model.

[0099] Specifically, firstly, the historical path data and historical driving style data in the training dataset are preprocessed. Preprocessing is a crucial step in neural network training, as it can improve model performance and accuracy. Common preprocessing methods include the following:

[0100] Data standardization / normalization: scaling data to a specific range (usually 0 to 1 or -1 to 1) so that features at different scales have the same weight and helps prevent overfitting;

[0101] Missing value handling: Missing data can be handled by methods such as interpolation, deletion, or filling.

[0102] Feature selection: Selecting the features most relevant to the target variable can reduce computational complexity and improve the model's generalization ability;

[0103] Feature engineering: Enhancing model performance by transforming, combining, or extracting new features. For example, operations such as convolution, pooling, and activation functions can be used to process image data;

[0104] Data splitting: The dataset is divided into training, validation and test sets to monitor the model's performance during training and to evaluate the model's generalization ability on the test set;

[0105] Label encoding: For classification problems, the labels are converted into numerical form (e.g., using one-hot encoding);

[0106] Batch size and batch gradient descent: When training a neural network, the parameters of batch size and batch gradient descent can be set to control the step size and frequency of each update, so as to improve the convergence speed and stability of the model.

[0107] Gradient clipping: To prevent gradient explosion during training, the magnitude of the gradient can be clipped at each step.

[0108] Learning rate decay: As training progresses, the learning rate can be gradually reduced to accelerate model convergence and reduce the risk of overfitting;

[0109] Regularization: By adding a regularization term to the loss function, the complexity of the model is penalized, thereby preventing overfitting. Common regularization methods include L1 regularization and L2 regularization.

[0110] After preprocessing the historical path data and historical driving style data in the training dataset, the topology of the neural network is determined and the connection weight values ​​between each neuron in the neural network are initially selected.

[0111] Then, after calculating the output values ​​of each layer through the neural network topology, the output values ​​are compared with the true values. If the model error corresponding to the error between the output value and the true value does not reach the ideal, the model parameters of the neural network model are updated. This process is repeated until convergence, resulting in a well-trained offline model with an error less than the set value.

[0112] The trained offline neural network model can be used to predict the future driving path of a vehicle and calculate vehicle control parameters. The flowchart based on the BP neural network algorithm is shown below. Figure 8 As shown.

[0113] Thus, this invention can not only train a random forest model on the training dataset using the random forest prediction algorithm to obtain a preset model for predicting vehicle control parameters corresponding to the future driving path, but also train a neural network model on the training dataset using the backpropagation neural network algorithm to predict vehicle control parameters corresponding to the future driving path.

[0114] In some implementations, the determination method includes:

[0115] 05: Input the vehicle's current time data and current spatial data into the neural network model to predict the vehicle's future driving path and determine the vehicle control parameters for the future driving path.

[0116] Please see Figure 2 Step 05 can be implemented by the parameter determination module 12. That is, the parameter determination module 12 is used to input the vehicle's current time data and current spatial data into the neural network model to predict the vehicle's future driving path and determine the vehicle control parameters for the future driving path.

[0117] In other words, the present invention can also directly predict the future driving path and calculate the vehicle control parameters of the future driving path through a neural network model, which is simpler and faster.

[0118] In addition, the present invention can also use a prediction algorithm combining random forest and BP neural network for model training.

[0119] Specifically, please refer to Figure 9 The preset algorithms are random forest prediction algorithm and backpropagation neural network algorithm. Step 04 includes:

[0120] 0491: Use the random forest prediction algorithm to classify or regress historical path data and historical driving style data to obtain preliminary prediction results;

[0121] 0492: Adjust the weights of the preliminary prediction results based on the backpropagation neural network algorithm to train and obtain the preset model.

[0122] Please combine Figure 4 Steps 0491 and 0492 can be implemented by modeling module 14. That is, modeling module 14 is used to classify or regress historical path data and historical driving style data using the random forest prediction algorithm to obtain preliminary prediction results; and to adjust the weights of the preliminary prediction results based on the backpropagation neural network algorithm to train a preset model.

[0123] Specifically, during model training, the random forest algorithm is first used to classify or regress historical path data and historical driving style data to obtain preliminary prediction results. Then, these preliminary prediction results are used as input to the backpropagation (BP) neural network algorithm, which further optimizes and adjusts the weights of the preliminary prediction results.

[0124] Specifically, the random forest algorithm can be used to extract features and patterns from historical path data and historical driving style data, and to perform classification or regression prediction on these data. The backpropagation (BP) neural network, on the other hand, can adjust and optimize weights based on the input data, making the model's predictions more accurate and stable.

[0125] Thus, this invention combines the random forest algorithm and the backpropagation (BP) neural network algorithm to train a pre-defined model, fully leveraging the advantages of both algorithms to improve the predictive performance of the pre-defined model. Simultaneously, this combined approach reduces model complexity, enhances its generalization ability, and enables the model to better adapt to various data and scenarios.

[0126] In addition, the specific method of using regression analysis prediction algorithms to train historical path data and historical driving style data is as follows: the gradient descent method is used to find the optimal solution. By iteratively calculating the gradient of the loss function, the minimum loss function is obtained, thereby finding the most suitable hypothesis function to fully fit the training dataset.

[0127] The specific method for training historical path data and historical driving style data using a Markov chain model is as follows: discretize time and state, calculate the probability of each state transition to obtain the state transition matrix, and finally obtain the Markov chain model.

[0128] This application also provides a vehicle that includes the vehicle control parameter determining device 10 described above.

[0129] The present invention also provides a non-volatile computer-readable storage medium containing a computer program. When the computer program is executed by one or more processors, it implements the method for determining vehicle control parameters as described in any of the above embodiments.

[0130] For example, when a computer program is executed by a processor, it implements the following steps for determining vehicle control parameters:

[0131] 01: Obtain the vehicle's current time and location data;

[0132] 02: Input the vehicle's current time data and current spatial data into a preset model trained by a preset algorithm based on the vehicle's historical path data and historical driving style data to determine the vehicle control parameters for the future driving path.

[0133] It is understood that a computer program includes computer program code. Computer program code can be in the form of source code, object code, executable files, or some intermediate form. Computer-readable storage media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, external hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc.

[0134] The computer-readable storage medium of the present invention uses the above-described method for determining vehicle control parameters to train a preset model using a prediction algorithm on historical path data and historical driving style data. The current time data and current spatial data of the vehicle are then input into the preset model, which can more accurately and easily predict the future driving path of the vehicle and the vehicle control parameters of the vehicle on the future driving path. This can effectively improve vehicle performance, driving experience and safety factor, thereby enhancing driving experience and safety.

Claims

1. A method for determining vehicle control parameters, characterized in that, The determination method includes: Obtain the current time data and current location data of the vehicle; The vehicle's current time data and current spatial data are input into a preset model trained by a preset algorithm based on the vehicle's historical path data and historical driving style data to determine the vehicle control parameters for the vehicle's future driving path. The determination method includes: The historical path data and historical driving style data of the vehicle under different scenarios are obtained as training datasets; The preset model is obtained by training the training dataset based on the preset algorithm. The preset algorithm is a backpropagation neural network algorithm, the preset model is a neural network model, and the process of training the training dataset based on the preset algorithm to obtain the preset model includes: Preprocess the historical path data and historical driving style data in the training dataset; Determine the topology of the neural network and the connection weights between the neurons in the neural network; The preprocessed historical path data and historical driving style data are input into the neural network topology to calculate the first output value. Calculate the error between the first output value and the true value, wherein the true value is the actual vehicle control parameter of the path corresponding to the first output value in the historical path data and the historical driving style data; If the error is greater than or equal to a set value, the model parameters of the neural network model are updated, and the model parameters include the connection weight values; The neural network topology is adjusted according to the updated model parameters, and a second output value is calculated until the error between the second output value and the true value is less than the set value, thus obtaining the trained neural network model.

2. The method for determining vehicle control parameters according to claim 1, characterized in that, The preset algorithm is a random forest prediction algorithm, the preset model is a random forest model, and the process of training the preset model on the training dataset based on the preset algorithm includes: Set model parameters, including the number of decision trees, wherein there are multiple decision trees; Randomly select historical path data and historical driving style data of the vehicle in a preset scenario as training samples; Multiple decision trees are trained based on multiple randomly selected training samples, and multiple vehicle control parameters corresponding to the multiple decision trees are output. The average value of multiple vehicle control parameters is taken to output the vehicle control parameters corresponding to the preset scenario, so as to train the random forest model.

3. The method for determining vehicle control parameters according to claim 2, characterized in that, After training each decision tree based on the training samples and outputting multiple vehicle control parameters corresponding to multiple decision trees, the determination method includes: A breadth-first search algorithm is used to evaluate and predict multiple vehicle control parameters to output optimized vehicle control parameters, thereby training the random forest model.

4. The method for determining vehicle control parameters according to claim 3, characterized in that, The determination method includes: The vehicle's current time data and current spatial data are input into the neural network model to predict the vehicle's future driving path and determine the vehicle control parameters for the future driving path.

5. The method for determining vehicle control parameters according to claim 1, characterized in that, The preset algorithm is a random forest prediction algorithm and a backpropagation neural network algorithm. The process of training the training dataset using the preset algorithm to obtain the preset model includes: The random forest prediction algorithm is used to classify or regress the historical path data and the historical driving style data to obtain preliminary prediction results. The weights of the preliminary prediction results are adjusted based on the backpropagation neural network algorithm to train the preset model.

6. A device for determining vehicle control parameters, characterized in that, The determining device is used to implement the determining method as described in any one of claims 1-5, the determining device comprising: The acquisition module is used to acquire the current time data and current spatial data of the vehicle; The parameter determination module is used to input the current time data and current spatial data of the vehicle into a preset model trained by a preset algorithm based on the historical path data and historical driving style data of the vehicle, and to determine the vehicle control parameters for the future driving path of the vehicle.

7. A vehicle, characterized in that, The vehicle includes the vehicle control parameter determination device as described in claim 6.

8. A non-volatile computer-readable storage medium containing a computer program, characterized in that, When the computer program is executed by one or more processors, it implements the method for determining vehicle control parameters as described in any one of claims 1-5.

Images