A method and system for realizing intelligent control of LED vehicle lights

Abstract

The present invention relates to the field of LED light control, and specifically to a method and system for realizing intelligent LED light control. An intelligent LED light control system comprises: a target vehicle position information acquisition module, a target vehicle position prediction module, a position prediction model real-time training module, a to-be-processed marked vehicle set construction module, a marked vehicle status information construction module, an LED light control strategy selection module and an LED light control module. The present invention obtains all marked vehicles within the target vehicle range through the Internet of Vehicles, and determines the status of the target vehicle and all marked vehicles at the next monitoring time point through a time series analysis, and then outputs the LED light control strategy through an LED light control strategy selection model based on the status of the target vehicle and all marked vehicles at the next monitoring time point, so as to realize the adaptive change of the high and low beam lights during the vehicle driving process, and avoid affecting the driving of other vehicles.

Description

Technical Field

[0001] The present invention relates to the field of LED lamp control, and in particular to an intelligent LED vehicle lamp control method and system. Background Art

[0002] With the rapid development of the automobile industry, vehicle lighting systems play an increasingly important role in driving safety. Traditional vehicle headlight systems usually include two modes: high beam and low beam. Drivers need to manually switch between high beam and low beam according to road and traffic conditions. However, this manual switching method has many inconveniences and safety hazards. Especially when driving at night, frequent switching between high beam and low beam not only increases the driver's operating burden, but may also cause dazzle to other vehicles due to untimely or inaccurate switching, thereby causing traffic accidents. In addition, the traditional manual switching method cannot be adaptively adjusted according to the vehicle conditions in the real-time environment, and it is difficult to provide the best lighting effect under complex and changing road conditions. Summary of the invention

[0003] The present invention obtains all marked vehicles within the target vehicle range through the Internet of Vehicles, and determines the status of the target vehicle and all marked vehicles at the next monitoring time point through temporal analysis, and then outputs the LED headlight control strategy through the LED headlight control strategy selection model based on the status of the target vehicle and all marked vehicles at the next monitoring time point, which can realize the adaptive change of high and low beam lights during vehicle driving and avoid affecting the driving of other vehicles.

[0004] A method for realizing intelligent LED vehicle light control, comprising:

[0005] Obtaining the target vehicle position information of the target vehicle at the current monitoring time point; combining the target vehicle position information obtained at the current monitoring time point with the target vehicle position information obtained N-1 times before to form a first position time series data set, and then sending the first position time series data set to the position prediction model for processing, and outputting the predicted position of the target vehicle; and also including real-time training of the position prediction model;

[0006] Vehicles within the preset range of the target vehicle position information at the current monitoring time point are recorded as marked vehicles, and all marked vehicles are grouped into a marked vehicle set. The marked vehicles in the marked vehicle set are traversed, and for each marked vehicle, the following operations are performed to obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, and the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N-1 times before are combined into a second position time series data set, and then the second position time series data set is sent to the position prediction model for processing, and the predicted position of the marked vehicle is output, and the predicted position of the marked vehicle is marked by the relative position relationship, and the relative position relationship includes the same direction, facing, and back; until all marked vehicles in the marked vehicle set have been traversed, the predicted positions of all marked vehicles with the relative position relationship of the same direction and facing are combined into a marked vehicle set to be processed;

[0007] Traverse the predicted positions of marked vehicles in the set of marked vehicles to be processed, and for each predicted position of the marked vehicle, perform the following operations: perform coordinate transformation operation on the predicted position of the marked vehicle based on the predicted position of the target vehicle, output the relative position of the marked vehicle, and add the relative position relationship corresponding to the predicted position of the marked vehicle to the end of the relative position of the marked vehicle to construct the marked vehicle status information; until all the predicted positions of the marked vehicles in the set of marked vehicles to be processed are traversed, output the status information of all marked vehicles;

[0008] The predicted position of the target vehicle and the state information of the marked vehicle are combined into state space information, and then the state space information is input into the LED headlight control strategy selection model for processing, and the LED headlight control strategy is output;

[0009] The LED lights are controlled based on the LED light control strategy to achieve adaptive changes of the LED lights.

[0010] As a preferred embodiment of the present invention, training the location prediction model specifically includes the following steps:

[0011] A number of position prediction training samples are obtained, each of which includes N+1 vehicle position information arranged in chronological order. All position prediction training samples are combined into a position prediction training set, and then the position prediction training set is sent to a parameter-initialized position prediction model for training. During the training, the first N vehicle position information in the position prediction training set is used as the input of the position prediction model, and the last vehicle position information in the position prediction training set is used as the target output to determine whether the training conditions are met. If the training conditions are met, the trained position prediction model is output, otherwise, the position prediction model is continued to be trained through the position prediction training set.

[0012] As a preferred embodiment of the present invention, real-time training is performed on the location prediction model, specifically comprising the following steps:

[0013] When obtaining the target vehicle position information at the current monitoring time point, calculate the distance between the target vehicle position information at the current monitoring time point and the predicted position of the target vehicle, and record it as the judgment distance. Then compare the judgment distance with the judgment threshold. If the judgment distance is less than the judgment threshold, no operation is performed. If the judgment distance is not less than the judgment threshold, obtain the time series training samples of all marked vehicles through the Internet of Vehicles. The time series training samples of the marked vehicles are composed of the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N times before. The time series training samples of all marked vehicles are combined into a time series training set, and the position prediction model is trained through the time series training set.

[0014] As a preferred embodiment of the present invention, the training of the model for selecting the control strategy of the LED headlights specifically includes the following steps:

[0015] Step S1: constructing an LED headlight control strategy network and an LED headlight control strategy evaluation network;

[0016] Step S2: randomly generate several copies of simulated state space information, and the form of the simulated state space information is consistent with that of the state space information. For each copy of the simulated state space information, perform the following operations: send the simulated state space information to the LED headlight control strategy network for processing, output the simulated LED headlight control strategy, obtain the reward value after executing the simulated LED headlight control strategy, and then combine the simulated state space information and the simulated LED headlight control strategy into an LED headlight control strategy evaluation training sample; divide all LED headlight control strategy evaluation training samples into an LED headlight control strategy evaluation training set and an LED headlight control strategy evaluation test set, and then combine the LED headlight control strategy evaluation training set into an LED headlight control strategy evaluation training set. The LED headlight control strategy evaluation network is sent to be trained. During the training, the reward value is used as the target output to determine whether the training conditions are met. If the training conditions are met, the training of the LED headlight control strategy evaluation network is completed. Otherwise, the LED headlight control strategy evaluation network is continued to be trained through the LED headlight control strategy evaluation training set. Each LED headlight control strategy evaluation training sample in the LED headlight control strategy evaluation test set is input into the trained LED headlight control strategy evaluation network for processing, and the target reward value is output. Based on the target reward value and the simulated LED headlight control strategy, the parameters in the LED headlight control strategy network are adjusted by the gradient descent method.

[0017] Step S3: Repeat step S2 until the number of repetitions of step S2 reaches a preset number, and use the LED headlight control strategy network as the LED headlight control strategy selection model.

[0018] As a preferred embodiment of the present invention, it also includes simulating and optimizing the hyper parameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network, and the specific steps are as follows:

[0019] Construct several hyperparameter simulation individuals, and combine all the hyperparameter simulation individuals into a simulation population;

[0020] Calculate the fitness value of each hyperparameter simulation individual. Specifically, set the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network through the hyperparameter simulation individual, train the set LED headlight control strategy network and the LED headlight control strategy evaluation network, and use the accuracy of the LED headlight control strategy selection model as the fitness value of the hyperparameter simulation individual.

[0021] Iteratively optimize the simulated population through a population optimization algorithm;

[0022] When the number of iterations reaches the maximum number of iterations, the hyperparameter simulation individual with the largest output fitness value sets the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network.

[0023] As a preferred embodiment of the present invention, the LED headlight control strategy network and the LED headlight control strategy evaluation network are constructed based on a BP neural network.

[0024] As a preferred embodiment of the present invention, the swarm optimization algorithm adopts the sparrow search algorithm.

[0025] The present invention also provides an intelligent LED vehicle light control system, comprising:

[0026] A target vehicle position information acquisition module, used to acquire the target vehicle position information of the target vehicle at the current monitoring time point;

[0027] The target vehicle position prediction module is used to combine the target vehicle position information obtained at the current monitoring time point with the target vehicle position information obtained N-1 times before to form a first position time series data set, and then send the first position time series data set to the position prediction model for processing, and output the predicted position of the target vehicle;

[0028] A location prediction model real-time training module, used for real-time training of the location prediction model;

[0029] A module for constructing a set of marked vehicles to be processed is used to record vehicles within a preset range of the target vehicle position information at the current monitoring time point as marked vehicles, form all marked vehicles into a marked vehicle set, traverse the marked vehicles in the marked vehicle set, and perform the following operations for each marked vehicle: obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, form a second position time series data set with the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N-1 times before, and then send the second position time series data set to the position prediction model for processing, output the predicted position of the marked vehicle, and mark the predicted position of the marked vehicle by the relative position relationship, and the relative position relationship includes the same direction, facing and back; until all marked vehicles in the marked vehicle set have been traversed, the predicted positions of all marked vehicles with the relative position relationship of the same direction and facing are formed into a set of marked vehicles to be processed;

[0030] The marked vehicle status information construction module is used to traverse the predicted marked vehicle positions in the marked vehicle set to be processed, and for each predicted marked vehicle position, perform the following operations: perform coordinate transformation operations on the predicted marked vehicle position based on the predicted target vehicle position, output the relative position of the marked vehicle, and add the relative position relationship corresponding to the predicted marked vehicle position at the end of the relative marked vehicle position to construct the marked vehicle status information; until all the predicted marked vehicle positions in the marked vehicle set to be processed are traversed, the status information of all marked vehicles is output;

[0031] The LED headlight control strategy selection module is used to combine the target vehicle predicted position and the marked vehicle state information into state space information, and then input the state space information into the LED headlight control strategy selection model for processing, and output the LED headlight control strategy;

[0032] The LED headlight control module is used to control the LED headlight based on the LED headlight control strategy to achieve adaptive changes of the LED headlight.

[0033] The present invention has the following advantages:

[0034] 1. The present invention obtains all marked vehicles within the target vehicle range through the Internet of Vehicles, and determines the status of the target vehicle and all marked vehicles at the next monitoring time point through temporal analysis, and then outputs the LED headlight control strategy through the LED headlight control strategy selection model based on the status of the target vehicle and all marked vehicles at the next monitoring time point, which can realize the adaptive change of the high and low beam lights during the driving of the vehicle and avoid affecting the driving of other vehicles.

[0035] 2. The present invention obtains the actual position changes of other vehicles through the Internet of Vehicles, and performs real-time training on the position prediction model based on the actual position changes of other vehicles, which can make the position prediction model's prediction of the vehicle position more in line with the actual geographical situation, thereby improving the prediction accuracy of the position prediction model.

[0036] 3. The present invention simulates and optimizes the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network through the sparrow search algorithm, so that the trained LED headlight control strategy selection model can have a higher accuracy rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of the structure of an intelligent LED vehicle light control system used in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0039] Embodiment 1, a method for realizing intelligent LED vehicle light control, comprising:

[0040] Obtain the target vehicle position information of the target vehicle at the current monitoring time point. It should be noted that the target vehicle is a vehicle driven by the user, and a position sensor is set on the vehicle. The position information of the vehicle can be obtained through the position sensor and uploaded to the Internet of Vehicles; the target vehicle position information obtained at the current monitoring time point and the target vehicle position information obtained N-1 times before form a first position time series data set, and then send the first position time series data set to the position prediction model for processing, and output the predicted position of the target vehicle. It should be noted that on a road section with a fixed trend, the vehicle's driving trajectory can be predicted in a time series manner; in order to improve the real-time prediction accuracy of the position prediction model, real-time training of the position prediction model is also included;

[0041] Vehicles within a preset range of the target vehicle position information at the current monitoring time point are recorded as marked vehicles. The preset range can be a circular range with a diameter of 1 km. All marked vehicles are grouped into a marked vehicle set. The marked vehicles in the marked vehicle set may affect the control of the high and low beam lights of the target vehicle during driving. The marked vehicles in the marked vehicle set are traversed, and for each marked vehicle, the following operations are performed to obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, and the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained for the previous N-1 times are combined into a second position time series data set, and then the second position time series data set is combined into a second position time series data set. The data set is sent to the position prediction model for processing, and the predicted position of the marked vehicle is output, and the predicted position of the marked vehicle is marked by the relative position relationship. The relative position relationship includes same direction, facing and back-to-back. Same direction means that the target vehicle and the marked vehicle are moving in the same direction, facing means that the target vehicle and the marked vehicle are moving face to face, and back-to-back means that the target vehicle and the marked vehicle are moving back to back. It should be noted that in the back-to-back state, the high and low beam lights of the target vehicle and the marked vehicle will not affect each other; until all the marked vehicles in the marked vehicle set have been traversed, the predicted positions of all marked vehicles with the relative position relationship of same direction and facing are composed of the marked vehicle set to be processed;

[0042] Traverse the predicted positions of the marked vehicles in the set of marked vehicles to be processed, and for each predicted position of the marked vehicle, perform the following operations: perform coordinate conversion operation on the predicted position of the marked vehicle based on the predicted position of the target vehicle, output the relative position of the marked vehicle, and realize mapping the predicted position of the marked vehicle to a coordinate system with the predicted position of the target vehicle as the origin, and this coordinate system takes the forward direction of the target vehicle as the X-axis, and add the position relative relationship corresponding to the predicted position of the marked vehicle at the end of the relative position of the marked vehicle to construct the marked vehicle status information. It should be added that the relative position relationship can be expressed in the form of coding, such as "001" for the same direction and "002" for facing; until all the predicted positions of the marked vehicles in the set of marked vehicles to be processed are traversed, and the status information of all marked vehicles is output;

[0043] The predicted position of the target vehicle and the state information of the marked vehicle are combined into state space information, and then the state space information is input into the LED headlight control strategy selection model for processing, and the LED headlight control strategy is output. It should be added that the LED headlight control strategy here refers to the parameters of the LED light continuous control function at the next monitoring time point, and the LED light continuous control function can be a nonlinear function; and it should be noted that the LED headlight used here is an LED matrix light, which can realize the continuous change from low beam to high beam, rather than a simple switch between two discrete modes;

[0044] The LED lights are controlled based on the LED light control strategy to achieve adaptive changes of the LED lights.

[0045] This application obtains all marked vehicles within the target vehicle range through the Internet of Vehicles, and determines the status of the target vehicle and all marked vehicles at the next monitoring time point through timing analysis, and then outputs the LED headlight control strategy through the LED headlight control strategy selection model based on the status of the target vehicle and all marked vehicles at the next monitoring time point, which can realize the adaptive changes of high and low beam lights during vehicle driving and avoid affecting the driving of other vehicles.

[0046] Training the location prediction model includes the following steps:

[0047] Several position prediction training samples are obtained. The position prediction training samples include N+1 vehicle position information arranged in chronological order. It should be noted that these vehicle position information are also obtained according to the monitoring time points and are obtained by the developers based on actual vehicle driving. All position prediction training samples are combined into a position prediction training set, and then the position prediction training set is sent to the parameter-initialized position prediction model for training. During the training, the first N vehicle position information in the position prediction training set is used as the input of the position prediction model, and the last vehicle position information in the position prediction training set is used as the target output to determine whether the training conditions are met. The training conditions are determined by the developers, such as if the model accuracy meets the expectations. If the training conditions are met, the trained position prediction model is output, otherwise, the position prediction model continues to be trained through the position prediction training set.

[0048] Real-time training of the location prediction model includes the following steps:

[0049] When obtaining the target vehicle position information at the current monitoring time point, calculate the distance between the target vehicle position information at the current monitoring time point and the predicted position of the target vehicle, and record it as the judgment distance, then compare the judgment distance with the judgment threshold. The judgment threshold is determined by the developer. If the judgment distance is less than the judgment threshold, it means that the accuracy of the position prediction model meets expectations and no operation is performed; if the judgment distance is not less than the judgment threshold, it means that the accuracy of the position prediction model does not meet expectations, which may be due to the presence of multiple bends on the road. Then, the time series training samples of all marked vehicles are obtained through the Internet of Vehicles. The time series training samples of the marked vehicles are composed of the marked vehicle position information of the marked vehicles at the current monitoring time point and the marked vehicle position information obtained N times before. The actual position information obtained by these other vehicles can provide a reference for the road driving rules at the current geographical location, and then the position prediction model is trained in real time. The time series training samples of all marked vehicles are composed of a time series training set, and the position prediction model is trained through the time series training set.

[0050] This application obtains the actual position changes of other vehicles through the Internet of Vehicles, and performs real-time training on the position prediction model based on the actual position changes of other vehicles, which can make the position prediction model's prediction of the vehicle position more in line with the actual geographical situation, thereby improving the prediction accuracy of the position prediction model.

[0051] The model is selected for training the LED headlight control strategy, which includes the following steps:

[0052] Step S1: constructing an LED headlight control strategy network and an LED headlight control strategy evaluation network based on a BP neural network;

[0053] Step S2: randomly generate several copies of simulated state space information, which includes the vehicle distribution in various situations, and the form of the simulated state space information is consistent with that of the state space information. For each copy of the simulated state space information, perform the following operations: send the simulated state space information to the LED headlight control strategy network for processing, output the simulated LED headlight control strategy, and obtain the reward value after executing the simulated LED headlight control strategy. It should be noted that the execution of the simulated LED headlight control strategy is implemented according to the simulation software, and the reward value can be calculated by evaluating the glare degree of other vehicles using UGR (unified glare rating), and then the simulated state space information and the simulated LED headlight control strategy are combined into an LED headlight control strategy evaluation training sample; all LED headlight control strategy evaluation training samples are divided into an LED headlight control strategy evaluation training set and an LED headlight control strategy evaluation test set, and then the LE The D headlight control strategy evaluation training set is sent to the LED headlight control strategy evaluation network for training. During the training, the reward value is used as the target output to determine whether the training conditions are met. The training condition can be whether the accuracy of the LED headlight control strategy evaluation network output meets expectations. If the training conditions are met, the training of the LED headlight control strategy evaluation network is completed. Otherwise, the LED headlight control strategy evaluation network is continued to be trained through the LED headlight control strategy evaluation training set; each LED headlight control strategy evaluation training sample in the LED headlight control strategy evaluation test set is input into the trained LED headlight control strategy evaluation network for processing, and the target reward value is output. Based on the target reward value and the simulated LED headlight control strategy, the parameters in the LED headlight control strategy network are adjusted by the gradient descent method, so that the target reward value output by the LED headlight control strategy network and corresponding to the simulated LED headlight control strategy reaches the maximum;

[0054] Step S3: Repeat step S2 until the number of repetitions of step S2 reaches a preset number of times, and the preset number of times is extracted and set, and the LED headlight control strategy network is used as the LED headlight control strategy selection model.

[0055] In the process of training the LED headlight control strategy selection model, it also includes simulating and optimizing the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network. The hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network include the number of hidden layers, the number of neural nodes in the hidden layers, etc. The specific steps are as follows:

[0056] Construct several hyperparameter simulation individuals, and form a simulation population with all the hyperparameter simulation individuals, wherein the hyperparameter simulation individuals include hyperparameter values ​​set based on prior knowledge;

[0057] Calculate the fitness value of each hyperparameter simulation individual. Specifically, set the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network through the hyperparameter simulation individual, train the set LED headlight control strategy network and the LED headlight control strategy evaluation network, and use the accuracy of the LED headlight control strategy selection model as the fitness value of the hyperparameter simulation individual.

[0058] The simulated population is iteratively optimized using a sparrow search algorithm that simulates the foraging behavior of sparrows and performs global and local searches through the collaboration of foragers and followers; foragers are responsible for exploring new areas to find food (solutions), while followers follow foragers and perform local searches around them;

[0059] When the number of iterations reaches the maximum number of iterations, the hyperparameter simulation individual with the largest output fitness value sets the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network.

[0060] The present application simulates and optimizes the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network through a sparrow search algorithm, so that the trained LED headlight control strategy selection model can have a higher accuracy rate.

[0061] Embodiment 2, a smart LED car light control system, such as Figure 1 As shown, including:

[0062] The target vehicle position information acquisition module is used to obtain the target vehicle position information of the target vehicle at the current monitoring time point. It should be noted that the target vehicle is the vehicle driven by the user, and a position sensor is set on the vehicle. The position information of the vehicle can be obtained through the position sensor and uploaded to the Internet of Vehicles;

[0063] The target vehicle position prediction module is used to combine the target vehicle position information obtained at the current monitoring time point with the target vehicle position information obtained N-1 times before to form a first position time series data set, and then send the first position time series data set to the position prediction model for processing, and output the predicted position of the target vehicle. It should be noted that on a road section with a fixed trend, the vehicle's driving trajectory can be predicted in a time series manner;

[0064] A location prediction model real-time training module, used for real-time training of the location prediction model;

[0065] The module for constructing the set of marked vehicles to be processed is used to record the vehicles within the preset range of the target vehicle position information at the current monitoring time point as marked vehicles. The preset range can be a circular range with a diameter of 1 km. All marked vehicles are grouped into a marked vehicle set. The marked vehicles in the marked vehicle set may affect the control of the high and low beam lights of the target vehicle during driving. The marked vehicles in the marked vehicle set are traversed, and for each marked vehicle, the following operations are performed to obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, and the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained for the previous N-1 times are combined into a second position time series data set. Then, the second position time series data set is sent to the position prediction model for processing, and the predicted position of the marked vehicle is output, and the predicted position of the marked vehicle is marked by the relative position relationship. The relative position relationship includes same direction, facing and back-to-back. Same direction means that the target vehicle and the marked vehicle are moving in the same direction, facing means that the target vehicle and the marked vehicle are moving face to face, and back-to-back means that the target vehicle and the marked vehicle are moving back to back. It should be noted that in the back-to-back state, the high and low beam lights of the target vehicle and the marked vehicle will not affect each other; until all the marked vehicles in the marked vehicle set are traversed, the predicted positions of all marked vehicles whose relative positions are same direction and facing form a marked vehicle set to be processed;

[0066] The marked vehicle status information construction module is used to traverse the predicted positions of the marked vehicles in the set of marked vehicles to be processed, and for each predicted position of the marked vehicle, perform the following operations: perform coordinate conversion operation on the predicted position of the marked vehicle based on the predicted position of the target vehicle, output the relative position of the marked vehicle, and realize mapping the predicted position of the marked vehicle to a coordinate system with the predicted position of the target vehicle as the origin, and this coordinate system takes the forward direction of the target vehicle as the X-axis, and add the position relative relationship corresponding to the predicted position of the marked vehicle at the end of the relative position of the marked vehicle to construct the marked vehicle status information. It should be added that the relative position relationship can be expressed in the form of coding, for example, "001" is the same direction and "002" is facing; until all the predicted positions of the marked vehicles in the set of marked vehicles to be processed are traversed, and the status information of all marked vehicles is output;

[0067] The LED headlight control strategy selection module is used to combine the target vehicle predicted position and the marked vehicle state information into state space information, and then input the state space information into the LED headlight control strategy selection model for processing, and output the LED headlight control strategy. It should be added that the LED headlight control strategy here refers to the parameters of the LED light continuous control function at the next monitoring time point, and the LED light continuous control function can be a nonlinear function; and it should be noted that the LED headlight used here is an LED matrix light, which can realize the continuous change from low beam to high beam, rather than a simple switch between two discrete modes;

[0068] The LED headlight control module is used to control the LED headlight based on the LED headlight control strategy to achieve adaptive changes of the LED headlight.

[0069] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for realizing intelligent LED vehicle light control, characterized in that: include: Obtain target vehicle position information of the target vehicle at the current monitoring time point; The target vehicle position information obtained at the current monitoring time point and the target vehicle position information obtained N-1 times before are combined into a first position time series data set, and the first position time series data set is then sent to a position prediction model for processing to output a predicted position of the target vehicle; and real-time training of the position prediction model is also included; Vehicles within the preset range of the target vehicle position information at the current monitoring time point are recorded as marked vehicles, and all marked vehicles are grouped into a marked vehicle set. The marked vehicles in the marked vehicle set are traversed, and for each marked vehicle, the following operations are performed to obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, and the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N-1 times before are combined into a second position time series data set, and then the second position time series data set is sent to the position prediction model for processing, and the predicted position of the marked vehicle is output, and the predicted position of the marked vehicle is marked by the relative position relationship, and the relative position relationship includes the same direction, facing, and back; until all marked vehicles in the marked vehicle set have been traversed, the predicted positions of all marked vehicles with the relative position relationship of the same direction and facing are combined into a marked vehicle set to be processed; Traverse the predicted positions of marked vehicles in the set of marked vehicles to be processed, and for each predicted position of the marked vehicle, perform the following operations: perform coordinate transformation operation on the predicted position of the marked vehicle based on the predicted position of the target vehicle, output the relative position of the marked vehicle, and add the relative position relationship corresponding to the predicted position of the marked vehicle to the end of the relative position of the marked vehicle to construct the marked vehicle status information; until all the predicted positions of the marked vehicles in the set of marked vehicles to be processed are traversed, output the status information of all marked vehicles; The predicted position of the target vehicle and the state information of the marked vehicle are combined into state space information, and then the state space information is input into the LED headlight control strategy selection model for processing, and the LED headlight control strategy is output; Control the LED lights based on the LED light control strategy to achieve adaptive changes of the LED lights; Training the location prediction model includes the following steps: Obtain several position prediction training samples, which include N+1 vehicle position information arranged in chronological order, and form all the position prediction training samples into a position prediction training set. Then, the position prediction training set is sent to a position prediction model with parameter initialization for training. During the training, the first N vehicle position information in the position prediction training set is used as the input of the position prediction model, and the last vehicle position information in the position prediction training set is used as the target output to determine whether the training conditions are met. If the training conditions are met, the trained position prediction model is output, otherwise, the position prediction model is continuously trained through the position prediction training set. Real-time training of the location prediction model includes the following steps: When obtaining the target vehicle position information at the current monitoring time point, calculate the distance between the target vehicle position information at the current monitoring time point and the target vehicle predicted position, and record it as the judgment distance, then compare the judgment distance with the judgment threshold. If the judgment distance is less than the judgment threshold, no operation is performed; if the judgment distance is not less than the judgment threshold, obtain the time series training samples of all marked vehicles through the Internet of Vehicles. The time series training samples of the marked vehicles are composed of the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N times before. The time series training samples of all marked vehicles are combined into a time series training set, and the position prediction model is trained through the time series training set. The model is selected for training the LED headlight control strategy, which includes the following steps: Step S1: constructing an LED headlight control strategy network and an LED headlight control strategy evaluation network; Step S2: randomly generate several copies of simulated state space information, and the form of the simulated state space information is consistent with that of the state space information. For each copy of the simulated state space information, perform the following operations: send the simulated state space information to the LED headlight control strategy network for processing, output the simulated LED headlight control strategy, obtain the reward value after executing the simulated LED headlight control strategy, and then combine the simulated state space information and the simulated LED headlight control strategy into an LED headlight control strategy evaluation training sample; divide all LED headlight control strategy evaluation training samples into an LED headlight control strategy evaluation training set and an LED headlight control strategy evaluation test set, and then combine the LED headlight control strategy evaluation training set into an LED headlight control strategy evaluation training set. The LED headlight control strategy evaluation network is sent to be trained. During the training, the reward value is used as the target output to determine whether the training conditions are met. If the training conditions are met, the training of the LED headlight control strategy evaluation network is completed. Otherwise, the LED headlight control strategy evaluation network is continued to be trained through the LED headlight control strategy evaluation training set. Each LED headlight control strategy evaluation training sample in the LED headlight control strategy evaluation test set is input into the trained LED headlight control strategy evaluation network for processing, and the target reward value is output. Based on the target reward value and the simulated LED headlight control strategy, the parameters in the LED headlight control strategy network are adjusted by the gradient descent method. Step S3: Repeat step S2 until the number of repetitions of step S2 reaches a preset number, and use the LED headlight control strategy network as the LED headlight control strategy selection model.

2. The method for realizing intelligent LED vehicle light control according to claim 1, characterized in that: It also includes simulation optimization of the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network. The specific steps are as follows: Construct several hyperparameter simulation individuals, and combine all the hyperparameter simulation individuals into a simulation population; Calculate the fitness value of each hyperparameter simulation individual. Specifically, set the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network through the hyperparameter simulation individual, train the set LED headlight control strategy network and the LED headlight control strategy evaluation network, and use the accuracy of the LED headlight control strategy selection model as the fitness value of the hyperparameter simulation individual. Iteratively optimize the simulated population through a population optimization algorithm; When the number of iterations reaches the maximum number of iterations, the hyperparameter simulation individual with the largest output fitness value sets the hyperparameters of the LED headlight control strategy network and the LED headlight control strategy evaluation network.

3. The method for realizing intelligent LED vehicle light control according to claim 2, characterized in that: The LED headlight control strategy network and the LED headlight control strategy evaluation network are constructed based on the BP neural network.

4. The method for realizing intelligent LED vehicle light control according to claim 2, characterized in that: The swarm optimization algorithm uses the sparrow search algorithm.

5. An intelligent LED car light control system, characterized in that: The system uses a method for realizing intelligent LED vehicle light control as described in any one of claims 1 to 4, including: A target vehicle position information acquisition module, used to acquire the target vehicle position information of the target vehicle at the current monitoring time point; The target vehicle position prediction module is used to combine the target vehicle position information obtained at the current monitoring time point with the target vehicle position information obtained N-1 times before to form a first position time series data set, and then send the first position time series data set to the position prediction model for processing, and output the predicted position of the target vehicle; A location prediction model real-time training module is used to perform real-time training on the location prediction model; A module for constructing a set of marked vehicles to be processed is used to record vehicles within a preset range of the target vehicle position information at the current monitoring time point as marked vehicles, form all marked vehicles into a marked vehicle set, traverse the marked vehicles in the marked vehicle set, and perform the following operations for each marked vehicle: obtain the marked vehicle position information of the marked vehicle at the current monitoring time point, form a second position time series data set with the marked vehicle position information of the marked vehicle at the current monitoring time point and the marked vehicle position information obtained N-1 times before, and then send the second position time series data set to the position prediction model for processing, output the predicted position of the marked vehicle, and mark the predicted position of the marked vehicle by the relative position relationship, and the relative position relationship includes the same direction, facing and back; until all marked vehicles in the marked vehicle set have been traversed, the predicted positions of all marked vehicles with the relative position relationship of the same direction and facing are formed into a set of marked vehicles to be processed; The marked vehicle status information construction module is used to traverse the predicted marked vehicle positions in the marked vehicle set to be processed, and for each predicted marked vehicle position, perform the following operations: perform coordinate transformation operations on the predicted marked vehicle position based on the predicted target vehicle position, output the relative position of the marked vehicle, and add the relative position relationship corresponding to the predicted marked vehicle position at the end of the relative marked vehicle position to construct the marked vehicle status information; until all the predicted marked vehicle positions in the marked vehicle set to be processed are traversed, the status information of all marked vehicles is output; The LED headlight control strategy selection module is used to combine the target vehicle predicted position and the marked vehicle state information into state space information, and then input the state space information into the LED headlight control strategy selection model for processing, and output the LED headlight control strategy; The LED headlight control module is used to control the LED headlight based on the LED headlight control strategy to achieve adaptive changes of the LED headlight.

Images