Decision-making method and system for lane changing before vehicle drives out of expressway and storage medium

Abstract

The invention discloses a decision-making method and system for lane changing before a vehicle is driven out of an expressway and a storage medium, and relates to the field of intelligent traffic control. The method comprises the following steps: constructing a highway vehicle driving behavior model, simulating driving behaviors of a vehicle driving away from a highway under different state parameters and state variables, and determining an end distance probability distribution model and an end distance classification boundary value; training a success rate evaluation model to accurately predict the success rate and uncertainty of successful departure of the vehicle, determining a success rate evaluation model corresponding to the expressway vehicle driving behavior model, constructing a model library, and determining the driving behavior of the expressway vehicle according to the to-be-decided road condition in combination with the preset success rate and confidence. According to the method, the applicable expressway vehicle driving behavior model is selected from the model library, the starting distance and the ending distance are predicted by using the expressway vehicle driving behavior model, and the starting position of lane changing before the vehicle is driven out of the expressway is determined, so that the safety of the vehicle driving out of the expressway and the stability of traffic flow are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic control, and in particular to a decision method, system and storage medium for a vehicle to change lanes before leaving an expressway. Background Art

[0002] At present, the "vehicle-road-cloud integration" of intelligent connected vehicles has become the key to promoting the large-scale industrialization and application of intelligent connected vehicles. The "vehicle-road-cloud integration" of intelligent connected vehicles uses advanced wireless communications and new generation Internet technologies to realize dynamic real-time information interaction between vehicles and roads in all directions, and carries out vehicle collaborative safety and active road control on the basis of full-time and space dynamic traffic information collection and integration, fully realizing the effective coordination of people, vehicles and roads, ensuring traffic safety and improving traffic efficiency. By combining it with the current booming autonomous driving technology, it can not only promote the rapid implementation of autonomous driving technology, but also provide services for autonomous driving vehicles and manned vehicles, effectively solving traffic control problems in mixed traffic scenarios, which is conducive to promoting comprehensive traffic management and is an important development direction of the current intelligent transportation system.

[0003] At present, with regard to the question of how far in advance a vehicle should change lanes to the outside at the exit of a multi-lane highway, manually driven vehicles mainly rely on the driver's experience to determine, while most intelligent assisted driving vehicles build a lane change model and preset the lane change length in advance during the lane change model training, calibration, testing and verification stages. However, the existing methods for determining the lane change strategy before a vehicle exits the highway rarely involve vehicle type, road type and traffic flow conditions, and lack a decision-making method for dynamic adaptive decision-making based on vehicle type, road type and traffic flow conditions, especially for the scenario where the innermost lane of the main road is a dedicated lane for autonomous driving.

[0004] Therefore, it is urgent to propose a decision method, system and storage medium for changing lanes before a vehicle exits the highway, so that when the target vehicle drives from the dedicated lane to the target exit ramp, the traffic flow interference area will not be expanded due to changing lanes too early, nor will the vehicle miss the exit due to changing lanes too late, effectively ensuring the driving safety of the vehicle before and after exiting the highway. Summary of the invention

[0005] The present invention aims to solve the above problems and provides a decision method, system and storage medium for changing lanes before a vehicle exits a highway, which solves the problem that it is difficult to scientifically determine the starting position of lane change when the target vehicle starts from the innermost dedicated autonomous driving lane and prepares to change lanes to the outermost exit ramp to exit the highway. By combining the traffic simulation method with the machine learning method, the highway vehicle driving behavior model is used to simulate and obtain the terminal distance probability distribution model and success rate evaluation model under different state parameters and state variables, and a model library is constructed to achieve the accurate determination of the starting position of the target vehicle leaving the highway, providing a basis for ensuring the safe driving of vehicles during lane change on the highway.

[0006] To achieve the above object, the present invention adopts the following technical solution:

[0007] A decision method for changing lanes before a vehicle exits an expressway is used to determine a starting position for changing lanes before the vehicle exits an expressway, comprising the following steps:

[0008] According to the road conditions to be decided, the state parameters of the road to be decided are determined, the traffic flow state of the road to be decided is obtained in real time, and the target success rate evaluation model is selected from the pre-built model library in combination with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters;

[0009] The selected target success rate evaluation model is used to predict the lane change success rate of the target vehicle starting from different starting positions on the decision-making road and heading to the exit ramp to leave the highway, and the starting position of the lane change before the target vehicle exits the highway is determined;

[0010] The method for constructing the model library includes:

[0011] Constructing a highway vehicle driving behavior model, presetting Q groups of state parameters, setting state variables for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model under each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter;

[0012] Wherein, for each of the Q groups of state parameters, the method for determining the success rate evaluation model includes:

[0013] A plurality of groups of state variables are set under the selected state parameters, and a plurality of simulation tests are carried out using a highway vehicle driving behavior model based on the plurality of groups of state variables to obtain simulation results. The termination distance probability distribution model and the termination distance classification boundary value are determined according to the simulation results of each state variable, the success rate and uncertainty of the target vehicle successfully leaving the exit ramp are calculated, and the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model is determined.

[0014] The highway vehicle driving behavior model is provided with a highway main road model and a vehicle model; the highway main road model is provided with a plurality of lanes, wherein the innermost lane is a dedicated lane and the outermost lane is an exit ramp; the vehicle model includes a target vehicle and a plurality of background vehicles, wherein the target vehicle travels in a dedicated lane, the background vehicles are other vehicles in the highway main road model except the target vehicle, and the target vehicle is provided with a vehicle-road dynamic real-time information interaction vehicle-mounted terminal, while the background vehicles are not provided with a vehicle-road dynamic real-time information interaction vehicle-mounted terminal;

[0015] The state parameters include the target vehicle type, the total number of lanes Z, the maximum speed limit and the minimum speed limit of each lane in the main road;

[0016] The state variables include the traffic flow state and the target vehicle lane change starting distance; the traffic flow state includes the average flow rate of a single lane during the simulation process and the average speed of a single lane The starting distance d s is the longitudinal distance between the target vehicle at the start of lane change in the innermost lane and the tip of the exit ramp diversion area. The end distance d e It is the longitudinal distance between the target vehicle at the position where it completes lane change in the outermost lane and the tip of the exit ramp diversion area.

[0017] Wherein, for each of the M groups of state parameters, the method for determining the success rate evaluation model includes:

[0018] Step a, constructing a highway vehicle driving behavior model and setting state parameters of the highway vehicle driving behavior model;

[0019] Step b, set multiple groups of state variables and the number of state variable combinations N, and determine the state variables of each group of state parameters and d s ,in, is the average flow rate of a single lane, is the average speed of a single lane, d s is the lane change starting distance of the target vehicle;

[0020] Step c, for the selected state variables, using the highway vehicle driving behavior model to perform n simulation tests, during each simulation test, by simulating the process of all vehicles on the main road of the highway changing lanes from the dedicated lane to the exit ramp, and simulating to obtain n simulation results, the simulation results include the termination distance and the departure result, wherein the departure result is represented by a Boolean value, if the target vehicle successfully leaves, the Boolean value is 1, if the target vehicle fails to leave, the Boolean value is 0;

[0021] Step d, according to the n termination distances obtained by multiple simulation tests of the current state variables, a probability density estimation method is used to determine the termination distance probability distribution model p(d e );

[0022] Step e: according to the simulation results obtained by multiple simulation tests of the current state parameters, the departure result corresponding to the termination distance is used as a Boolean value label to obtain n termination distance samples, and the termination distance samples are used to train a preset classifier. The training classifier predicts whether the target vehicle successfully leaves the exit ramp based on the termination distance, obtains the termination distance critical value between successfully leaving the exit ramp and unsuccessfully leaving the exit ramp, and determines the termination distance classification boundary value d eb and the uncertainty of the classification boundary σ eb ;

[0023] Step f, based on the termination distance probability distribution model in S1.4 and the termination distance classification boundary value d in S1.5 eb , determine the success rate and uncertainty of the target vehicle successfully leaving the exit ramp under the current state variables;

[0024] Step g, judging whether the current number of simulations exceeds the preset number of state variable combinations N, if it has exceeded the preset number of state variable combinations N, then proceeding to step h, otherwise, selecting the next set of state variables and returning to steps c to f to continue the simulation test until it exceeds the preset number of state variable combinations N, and obtaining N simulation data samples, the simulation data samples are represented by (X i ,y i ), where X i is the state variable of the ith simulation test, y i is the success rate of the ith simulation test, and enters step h;

[0025] Step h: Determine the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model

[0026] Among them, in step f, the success rate of the target vehicle successfully leaving the exit ramp under the current state parameters is:

[0027]

[0028] In the formula, c i is the success rate of the target vehicle successfully leaving the exit ramp in the i-th simulation test; p is the probability distribution model of the termination distance; Z is the total number of lanes, z is the lane number, where z = 1, 2, ..., Z; v zmax is the maximum speed limit of the lane, v zmin is the minimum speed limit for the lane; is the termination distance obtained from the ith simulation test, is the average flow rate of a single lane in the ith simulation test, is the average speed of a single lane in the ith simulation test, is the lane change starting distance of the target vehicle in the ith simulation test, and M is the highway vehicle driving behavior model in the ith simulation test;

[0029] The uncertainty corresponding to the success rate of the target vehicle successfully leaving the exit ramp under the current state variable is:

[0030]

[0031] In the formula, is the uncertainty of the target vehicle successfully leaving the exit ramp in the i-th simulation test.

[0032] Wherein, in the step h, a success rate evaluation model is constructed based on a Gaussian process regression model, wherein the success rate evaluation model is trained using simulation data samples, the state variable is used as the input feature of the success rate evaluation model, and the success rate is used as the target variable output by the success rate evaluation model;

[0033] During the training process of the success rate evaluation model, the state variables of the simulation data samples are input into the success rate evaluation model, and the success rate evaluation model is used to calculate the success rate estimation value according to the state variables. The accuracy calculated by the success rate evaluation model is obtained in combination with the success rate corresponding to the simulation data samples, and the accuracy calculated by the success rate evaluation model is compared with the preset accuracy. If the accuracy calculated by the success rate evaluation model does not reach the preset accuracy, the hyperparameters of the success rate evaluation model are optimized and the success rate evaluation model is continued to be trained. Otherwise, the trained success rate evaluation model is obtained.

[0034] Wherein, the success rate evaluation model is:

[0035]

[0036] in,

[0037] The radial basis kernel function k(X,X′) is:

[0038]

[0039] Where X and X′ are state variables in the simulation data sample, y(X) is the success prior distribution; m(X) is the mean function; k(X,X′) is the kernel function; is the variance of the signal; l is the length scale parameter, which is used to control the smoothness of the radial basis kernel function;

[0040] The hyperparameter optimization of the success rate evaluation model adopts the log-likelihood function, which is:

[0041]

[0042] Where p(y|X,θ) is the likelihood function; θ is the set of hyperparameters to be optimized; T is the transposed matrix; K is the covariance matrix calculated by the kernel function; is the variance of the noise; y is the success rate sample;

[0043] The log-likelihood function is optimized based on the gradient descent method. After optimization, for the new input feature X * , predict the new input feature X * The mean μ(X * ) and variance σ 2 (X * )for:

[0044]

[0045] Where I is the unit matrix.

[0046] The embodiment of the present invention also provides a lane change decision system for a vehicle before exiting a highway, comprising a model selection module, an output module and a model library component module;

[0047] The model selection module is configured to determine the state parameters of the road to be decided according to the road conditions to be decided, obtain the traffic flow state of the road to be decided in real time, and select the target success rate evaluation model from the pre-built model library in combination with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters;

[0048] The output module is configured to use the selected target success rate evaluation model to predict the starting distance and the ending distance of the target vehicle successfully leaving the exit ramp in the road to be decided, and determine the starting position of the target vehicle changing lanes before leaving the highway.

[0049] Wherein, the model selection module is in communication connection with the model library component module, and the model library component module is configured as follows:

[0050] Constructing a highway vehicle driving behavior model, presetting Q groups of state parameters, setting state variables for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model constructed according to each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter;

[0051] Wherein, for each of the Q groups of state parameters, the method for determining the success rate evaluation model includes:

[0052] A plurality of groups of state variables are set under the selected state parameters, and a plurality of simulation tests are carried out using a highway vehicle driving behavior model based on the plurality of groups of state variables to obtain simulation results. The termination distance probability distribution model and the termination distance classification boundary value are determined according to the simulation results of each state variable, the success rate and uncertainty of the target vehicle successfully leaving the exit ramp are calculated, and the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model is determined.

[0053] Wherein, the decision module is provided with an output end, and the output end is preset in the vehicle to be changed lanes, and is used for the vehicle to be changed lanes to exit the expressway according to the lane change starting position.

[0054] An embodiment of the present invention further provides a computer-readable storage medium on which computer program instructions are stored. The program instructions are executed by a lane change decision system before a vehicle exits a highway to implement the steps of the above method.

[0055] The beneficial technical effects brought by the present invention are:

[0056] The present invention proposes a decision method, system and storage medium for a vehicle to change lanes before leaving an expressway. The method simulates the lane-changing and leaving behaviors of a target vehicle on an expressway under different state parameters and state variables, builds a model library, and combines the real-time traffic status to be applicable to a variety of traffic scenarios. The method realizes a quick decision on the starting position of the lane change before the vehicle leaves the expressway, and solves the problem that the starting position of the lane change when the vehicle leaves the expressway at the current stage needs to be preset based on manual experience, lacks scientific analysis, and is difficult to quickly and adaptively adjust the lane-changing decision plan according to the real-time road conditions. The method can provide real-time lane-changing starting position decision support for networked vehicles traveling on the expressway, guide the networked vehicles to change lanes and leave, and effectively improve the driving safety of vehicles on the expressway and the stability of traffic flow. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The flowchart of the decision method for changing lanes before a vehicle exits a highway according to the present invention is exemplarily shown.

[0058] Figure 2The flowchart illustrates a method for constructing a success rate evaluation model according to the present invention.

[0059] Figure 3 A schematic diagram of a highway vehicle driving behavior model according to this embodiment is exemplarily shown.

[0060] Figure 4 A schematic diagram of a lane change decision system before a vehicle exits a highway according to the present invention is exemplarily shown.

[0061] Figure 5 A schematic diagram of another lane change decision system before a vehicle exits a highway according to the present invention is exemplarily shown. DETAILED DESCRIPTION

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0063] Example 1

[0064] This embodiment proposes a decision method for changing lanes before a vehicle exits an expressway, which is used to determine a starting position for changing lanes before the vehicle exits an expressway.

[0065] Figure 1 The flowchart of the decision method for changing lanes before a vehicle exits a highway according to this embodiment is exemplified.

[0066] like Figure 1 As shown, the decision method for changing lanes before a vehicle exits an expressway in this embodiment includes steps S101 to S102.

[0067] In step S101, according to the road conditions to be decided, the state parameters in the road to be decided are determined, the traffic flow state of the road to be decided is obtained in real time, and the target success rate evaluation model is selected from the pre-built model library in combination with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters. Among them, the state parameters can be used to measure the objective environmental state that affects the vehicle lane change, and the state parameters can include multiple parameter combinations. Under different parameter combinations, there are differences in vehicle driving behavior, and the success rate and confidence of successfully leaving the exit ramp are different. By obtaining the real-time traffic flow state, combined with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters, a success rate evaluation model that matches the current state parameters, traffic flow state and success rate expectations can be selected in the model library to improve the accuracy of lane change decisions.

[0068] In step S102, the selected target success rate evaluation model is used to predict the lane change success rate of the target vehicle starting from different starting positions in the road to be decided and driving towards the exit ramp to leave the highway, and determine the starting position of the lane change before the target vehicle exits the highway. For example, if the predicted lane change success rates corresponding to the starting positions A, B, and C are 90%, 80%, and 70%, respectively, the starting position A with the highest lane change success rate can be selected as the starting position for the lane change decision.

[0069] The model library can be pre-built. By simulating the driving state of the vehicle under various state parameter combinations and conducting simulation experiments, a model library of an evaluation model that can predict the lane change success rate under various traffic conditions is constructed to achieve accurate judgment of the vehicle lane change decision in a real-time traffic environment.

[0070] In an embodiment of the present invention, a specific method for constructing a model library may include: presetting Q groups of state parameters, respectively setting a highway vehicle driving behavior model for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model constructed according to each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter. It should be understood that for Q groups of state parameters, the model library may include Q success rate evaluation models.

[0071] Further, for each of the Q groups of state parameters, the method for determining the success rate evaluation model includes:

[0072] A plurality of groups of state variables are set under the selected state parameters, and a plurality of simulation tests are carried out using a highway vehicle driving behavior model based on the plurality of groups of state variables to obtain simulation results. The termination distance probability distribution model and the termination distance classification boundary value are determined according to the simulation results of each state variable, the success rate and uncertainty of the target vehicle successfully leaving the exit ramp are calculated, and the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model is determined.

[0073] Figure 2A method for constructing a success rate evaluation model according to an example of the present invention is specifically illustrated. In this example, a highway vehicle driving behavior model is constructed based on traffic simulation software, such as SUMO software. Multiple groups of state parameters of the highway vehicle driving behavior model are set. Multiple groups of state variables can be set under the state parameters of the selected group. Multiple simulation tests are performed using the highway vehicle driving behavior model to obtain the simulation results of each simulation test, including the termination distance and Boolean value of the target vehicle. The termination distance probability distribution model and the termination distance classification boundary value are determined based on all simulation results under each state variable, and the success rate and uncertainty of the target vehicle successfully leaving the exit ramp corresponding to each state variable are calculated to determine the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model.

[0074] Specifically, the following sub-steps may be included:

[0075] Step a: construct a highway vehicle driving behavior model based on traffic simulation software including SUMO, and set state parameters of the highway vehicle driving behavior model.

[0076] The highway vehicle driving behavior model M is provided with a highway main road model and a vehicle model, such as Figure 3 As shown, the highway main road model is provided with multiple lanes, wherein the innermost lane is a dedicated lane and the outermost lane is an exit ramp; the vehicle model includes a target vehicle and multiple background vehicles, wherein the target vehicle travels in the dedicated lane, the background vehicles are other vehicles in the highway main road model except the target vehicle, and the target vehicle uses a vehicle-road dynamic real-time information interaction vehicle-mounted terminal, while the background vehicles do not use a vehicle-road dynamic real-time information interaction vehicle-mounted terminal.

[0077] The state parameters include the target vehicle type, the total number of lanes, and the maximum speed limit and the minimum speed limit of each lane in the main road.

[0078] In this embodiment, the target vehicle type includes but is not limited to a networked manually driven vehicle, a networked autonomous driving vehicle, or a fleet consisting of a networked manually driven vehicle as the first vehicle and networked autonomous driving vehicles as the remaining vehicles. In the example of the present invention, the total number of lanes is set to 4, and the maximum speed limit and minimum speed limit of each lane in the main road are {v 1max ,v 1min ,v 2max ,v 2min ,v 3max ,v 3min ,v 4max ,v 4min}={120,100,120,90,100,80,100,60}, the unit is km / h. The upstream length of the simulation section in the highway main road model is set to 4km, the downstream length is set to 200m, the extended lane length is set to 100m, and the ramp length is set to 200m. The first lane on the left side of the highway main road model is set as the dedicated lane for the target vehicle. Background vehicles are not allowed to enter, and the target vehicle is not allowed to switch out before leaving the starting position of the highway lane change.

[0079] The highway main road model can select the lane change model LC2013 and the lane change model DK2008 in the traffic simulation software SUMO. In this example, the default vehicle lane change model LC2013 in the traffic simulation software SUMO is selected for simulation. During the simulation, when the target vehicle arrives at the lane change starting position, the target vehicle is allowed to leave the inner dedicated lane, and the autonomous lane change module is activated. There are four lane change decisions in this model. Here, since the target vehicle needs to be forced to exit from the front ramp exit, a strategic lane change is performed. In each simulation step, the speed change of the target vehicle itself and the obstructing vehicles in front and behind are calculated to promote the successful execution of the expected lane change action.

[0080] Set state variables, including traffic flow state and target vehicle lane change starting distance; the traffic flow state includes the average flow of a single lane during the simulation process and the average speed of a single lane The starting distance d s is the longitudinal distance between the target vehicle at the start of lane change in the innermost lane and the tip of the exit ramp diversion area. The end distance d e It is the longitudinal distance between the target vehicle at the position where it completes lane change in the outermost lane and the tip of the exit ramp diversion area.

[0081] In this embodiment, the average flow rate of a single lane is set The value range is [0,1800], the unit is vehicle / hour, the average speed of a single lane The value range of is [0,120], the unit is km / h, the value range of the starting distance of the target vehicle lane change is [200,3000], the unit is m, and the average flow rate of a single lane According to the value range, we can evenly divide it into n q The average flow value of a single lane is obtained by evenly dividing the average speed of a single lane according to the value range. v The target vehicle lane change starting distance is divided evenly according to the value range to obtain n single lane average speed values. d The starting distance value of the target vehicle lane change forms a total of N groups of state variables, N = n q n v n d, where n in this embodiment q =19,n v =13,n d =15.

[0082] Step b, set multiple groups of state variables and the number of state variable combinations N, and determine the state variables of each group of state parameters and d s ,in, is the average flow rate of a single lane, is the average speed of a single lane, d s is the lane change starting distance of the target vehicle.

[0083] Step c, for the selected state variables, use the highway vehicle driving behavior model to perform a preset total number of simulations, for example, 100 simulation tests. During each simulation test, the process of the target vehicle changing lanes from the dedicated lane to the exit ramp during the driving of all vehicles on the main road of the highway is simulated, and 100 simulation results are obtained by simulation. The simulation results include the termination distance and the departure result, wherein the departure result is represented by a Boolean value, and if the target vehicle successfully leaves, the Boolean value is 1, and if the target vehicle fails to leave, the Boolean value is 0.

[0084] Step d, based on the 100 termination distances obtained from multiple simulation tests of the current state variables, a probability density estimation method is used to determine the termination distance probability distribution model p(d e ).

[0085] In this embodiment, the termination distance d obtained from 100 simulation tests is e1 ,d e2 ,…,d e100 , the nonparametric kernel density estimation termination distance probability distribution model p(d e ), this method does not need to assume the probability distribution form expressed by a few parameters, but directly estimates the probability density function from the data, which can maximize the approximation of the true probability density distribution in the infinite-dimensional parameter space.

[0086] Since kernel density estimation discretizes the infinite-dimensional parameter space into finite dimensions, that is, discretizes the infinite-dimensional parameter space into several intervals, and then determines their respective probability distribution values ​​according to the frequency of sample points in each interval.

[0087] When the histogram distribution model is used as the kernel function, the center point and interval length of each interval are used as the hyperparameters of the kernel function to determine the numerical expression of the specific discrete dimension. If the interval length is h, the probability density of any point can be determined by the frequency of the sample points obtained by taking itself as the center point of the interval, that is:

[0088]

[0089] Among them, 1(A) is the indicative function,

[0090]

[0091] Where p(x; h) is the probability distribution function of the termination distance; x is the sample data point whose probability density is to be estimated; i is the serial number of the data point; x i is the i-th data point in the simulated data sample; h is the interval length; n is the number of simulated data samples; A is a Boolean value.

[0092] After testing, the histogram distribution model meets the normalization requirements.

[0093] At the same time, the Gaussian function can also be used as the kernel function, that is:

[0094]

[0095] Where h is the bandwidth parameter, usually called the smoothing parameter, which is used to adjust the smoothness of the histogram distribution model, thereby affecting the accuracy of the probability density function.

[0096] Step e: Based on the simulation results obtained from multiple simulation tests of the current state parameters, the departure result corresponding to the termination distance is used as a Boolean value label to obtain 100 termination distance samples. The termination distance samples are used to train a preset classifier, and the training classifier predicts whether the target vehicle successfully leaves the exit ramp based on the termination distance.

[0097] In this embodiment, the classifier is specifically a binary classification algorithm logistic regression learning classification model. During the training process, the termination distance d e As the input feature of the classifier, the success mark is used as the target variable. At the same time, in order to improve the generalization ability of the classifier, cross-validation is used to select the optimal model parameters of the classifier. The input terminal distance samples are divided and marked as the terminal distance samples of "successfully leaving the exit ramp" and the terminal distance samples of "unsuccessfully leaving the exit ramp". The critical value of the terminal distance between successfully leaving the exit ramp and unsuccessfully leaving the exit ramp is obtained, and the terminal distance classification boundary value d is determined. eb and the uncertainty of the classification boundary σ eb .

[0098] Step f, based on the termination distance probability distribution model in S1.4 and the termination distance classification boundary value d in S1.5 eb , through a simple cumulative probability calculation, the success rate of the target vehicle successfully leaving the exit ramp under the current state variable is determined as:

[0099]

[0100] In the formula, c i is the success rate of the target vehicle successfully leaving the exit ramp in the i-th simulation test; p is the probability distribution model of the termination distance; Z is the total number of lanes, z is the lane number, z=1,2,…,Z, v zmax is the maximum speed limit of the lane, v zmin is the minimum speed limit for the lane; is the termination distance obtained from the ith simulation test, is the average flow rate of a single lane in the ith simulation test, is the average speed of a single lane in the ith simulation test, is the lane change starting distance of the target vehicle in the i-th simulation test, and M is the highway vehicle driving behavior model of the i-th simulation test.

[0101] The uncertainty corresponding to the success rate of the target vehicle successfully leaving the exit ramp under the current state variable is:

[0102]

[0103] In the formula, is the uncertainty of the target vehicle successfully leaving the exit ramp in the i-th simulation test.

[0104] Step g, judging whether the value of the current simulation number i exceeds the preset state variable combination number N, if it has exceeded the preset state variable combination number N, then proceeding to step h, otherwise, selecting the next set of state variables and returning to steps c to f to continue the simulation test until it exceeds the preset state variable combination number N, and obtaining N simulation data samples. The simulation data sample is represented by (X i ,y i ), where X i is the state variable of the ith simulation test, y i is the success rate of the ith simulation test.

[0105] When i>N, go to step h.

[0106] Step h: Determine the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model

[0107] In this embodiment, a success rate evaluation model is constructed based on a Gaussian process regression model, and the success rate evaluation model is trained using simulation data samples. As the input feature of the success rate evaluation model, construct the feature vector Success rate i The target variable y as the output of the success rate evaluation model i .

[0108] During the training process of the success rate evaluation model, the state variables of the simulation data samples are input into the success rate evaluation model, and the success rate evaluation model is used to calculate the success rate estimation value according to the state variables. The accuracy calculated by the success rate evaluation model is obtained in combination with the success rate corresponding to the simulation data sample input, and the accuracy calculated by the success rate evaluation model is compared with the preset accuracy. If the accuracy calculated by the success rate evaluation model does not reach the preset accuracy, the hyperparameters of the success rate evaluation model are optimized and the success rate evaluation model is continued to be trained. Otherwise, the trained success rate evaluation model is obtained.

[0109] When the success rate evaluation model is constructed using the Gaussian process regression model, the expression is:

[0110]

[0111] Where y(X) is the prior distribution of success; m(X) is the mean function, which is assumed to be the zero function, that is, m(X) = 0; k(X, X′) is the kernel function, which is used to describe the correlation between input sample points, where X and X′ are the same or different feature vectors.

[0112] The radial basis kernel function is used as the kernel function of the Gaussian process regression model, and the radial basis kernel function k(X,X′) is:

[0113]

[0114] In the formula, is the variance of the signal; l is the length scale parameter, which is used to control the smoothness of the radial basis kernel function.

[0115] In order to improve the fitting effect of the success rate evaluation model, the maximum likelihood estimation or Bayesian optimization is used to optimize the hyperparameters of the success rate evaluation model. In this embodiment, the hyperparameter optimization of the success rate evaluation model adopts the log-likelihood function, and the log-likelihood function is:

[0116]

[0117] Where p(y|X,θ) is the likelihood function; θ is the set of hyperparameters to be optimized; T is the transposed matrix; K is the covariance matrix calculated by the function; is the variance of the noise.

[0118] The log-likelihood function is optimized based on the gradient descent method, with the goal of maximizing the log-likelihood function and obtaining the optimal hyperparameter value. When the success rate evaluation model is trained using simulated data samples, the success rate evaluation model can not only predict the success rate through learning, but also estimate the uncertainty of the success rate prediction value based on the correlation between the sample data. The optimized success rate evaluation model is effective for the new input feature X.* , predict the new input feature X * The mean μ(X * ) and variance σ 2 (X * )for:

[0119]

[0120] Where I is the identity matrix, and k is the covariance matrix obtained by substituting the known simulation samples into the kernel function.

[0121] In one embodiment of the present invention, according to the real-time average flow rate of a single lane of the road to be decided and the average speed of a single lane Average traffic volume per lane and the average speed of a single lane Input into the success rate evaluation model corresponding to the highway vehicle driving behavior model, and use the one-dimensional slice corresponding to the success rate evaluation model Combined with the preset confidence requirements, according to the 2sigma principle or the 3sigma principle, the lower bound of the credible range of the one-dimensional slice is obtained. When the 2sigma principle is used, the confidence level is 95%, that is, When the 3 sigma principle is used, the confidence level is 99%, that is,

[0122] To ensure satisfaction Starting distance d s , which is the final decision output starting distance result, that is, the best starting position.

[0123] Example 2

[0124] An embodiment of the present invention also provides a lane change decision system for a vehicle before leaving a highway.

[0125] Figure 4 A schematic diagram of a lane change decision system before a vehicle exits a highway according to the present invention is exemplarily shown.

[0126] like Figure 4 As shown, the lane change decision system 400 for a vehicle before exiting a highway includes a model selection module 401 , an output module 402 and a model library component module 403 .

[0127] Among them, the model selection module 401 is configured to determine the state parameters of the road to be decided according to the situation of the road to be decided, obtain the traffic flow state of the road to be decided in real time, and select the target success rate evaluation model from the pre-built model library based on the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters.

[0128] The output module configuration 402 is to use the selected target success rate evaluation model to predict the starting distance and the ending distance of the target vehicle successfully leaving the exit ramp in the road to be decided, and determine the starting position of the target vehicle to change lanes before leaving the highway.

[0129] The model library component module 403 is configured as follows: constructing a highway vehicle driving behavior model, presetting Q groups of state parameters, setting state variables for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model constructed according to each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter; wherein, for each group in the Q groups of state parameters, a method for determining the success rate evaluation model includes: setting multiple groups of state variables under selected state parameters, performing multiple simulation tests using the highway vehicle driving behavior model based on the multiple groups of state variables to obtain simulation results, determining a termination distance probability distribution model and a termination distance classification boundary value according to the simulation results of each state variable, calculating the success rate and uncertainty of the target vehicle successfully leaving the exit ramp, and determining the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model.

[0130] The model selection module 401 is in communication connection with the model library component module 403 .

[0131] Example 3

[0132] In this embodiment, another lane-changing decision system for a vehicle before exiting a highway is proposed, which is used to implement the lane-changing decision method for a vehicle before exiting a highway described in Embodiment 1.

[0133] Figure 5 A schematic diagram of another lane change decision system before a vehicle exits a highway according to the present invention is exemplarily shown.

[0134] like Figure 5 The other lane change decision system for a vehicle before exiting a highway comprises a simulation module, a termination distance prediction module, a success rate estimation module, a success rate evaluation model establishment module, a road status real-time acquisition module and a decision module.

[0135] The simulation module is preset with a highway vehicle driving behavior model, which is used to simulate the process of the target vehicle leaving the exit ramp using the highway vehicle driving behavior model, and obtain and store simulation results under different state parameters and state variables.

[0136] The simulation module is connected to the termination distance prediction module and the success rate estimation module respectively, and is used to transmit the simulation result to the termination distance prediction module and the success rate estimation module.

[0137] The termination distance prediction module is connected to the success rate estimation model and is used to determine the termination distance probability distribution model according to the simulation results obtained from the simulation module.

[0138] The success rate estimation module is connected to the success rate evaluation model establishment module, and is used to determine the success rate and uncertainty of the target vehicle successfully leaving the exit ramp based on the simulation results obtained from the simulation module and the termination distance probability distribution model determined by the termination distance prediction module.

[0139] The success rate evaluation model building module is connected to the decision-making module, and is used to use the simulation results and their corresponding success rates to train the success rate evaluation model, optimize the success rate evaluation model corresponding to the highway vehicle driving behavior model constructed by each state parameter, and build a success rate evaluation model construction model library.

[0140] The road status real-time acquisition module is connected to the success rate evaluation model establishment module, and is used to obtain the status parameters of the highway to be decided in real time, and transmit the type of target vehicle on the highway to be decided, the maximum speed limit and the minimum speed limit of each lane in the main road to the decision module.

[0141] The decision module is used to formulate a decision plan for the starting position of lane change before the vehicle exits the highway. According to the highway state parameters to be decided transmitted by the road state real-time acquisition module, combined with the preset requirements of the success rate and uncertainty of the target vehicle successfully leaving the exit ramp, a suitable highway vehicle driving behavior model is selected from the model library, and the highway vehicle driving behavior model is used to predict the starting distance and the ending distance of the target vehicle successfully leaving the exit ramp in the road to be decided, so as to determine the starting position of the lane change before the target vehicle exits the highway.

[0142] Example 4

[0143] In this embodiment, a computer-readable storage medium is proposed, on which computer program instructions are stored. The program instructions are executed by the lane change decision system before the vehicle exits the highway described in Example 2 or 3, and are used to implement the specific steps of the lane change decision method before the vehicle exits the highway described in Example 1.

[0144] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A decision method for changing lanes before a vehicle exits a highway, characterized in that: The method is used to determine the starting position of a lane change before a vehicle exits a highway, and includes the following steps: According to the road conditions to be decided, the state parameters of the road to be decided are determined, the traffic flow state of the road to be decided is obtained in real time, and the target success rate evaluation model is selected from the pre-built model library in combination with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters; The selected target success rate evaluation model is used to predict the lane change success rate of the target vehicle starting from different starting positions on the decision-making road and heading to the exit ramp to leave the highway, and the starting position of the lane change before the target vehicle exits the highway is determined; The method for constructing the model library includes: Constructing a highway vehicle driving behavior model, presetting Q groups of state parameters, setting state variables for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model under each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter; Wherein, for each of the Q groups of state parameters, the method for determining the success rate evaluation model includes: A plurality of groups of state variables are set under the selected state parameters, and a plurality of simulation tests are carried out using a highway vehicle driving behavior model based on the plurality of groups of state variables to obtain simulation results. The termination distance probability distribution model and the termination distance classification boundary value are determined according to the simulation results of each state variable, the success rate and uncertainty of the target vehicle successfully leaving the exit ramp are calculated, and the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model is determined.

2. The decision method for changing lanes before a vehicle exits a highway according to claim 1, characterized in that: The highway vehicle driving behavior model is provided with a highway main road model and a vehicle model; the highway main road model is provided with multiple lanes, wherein the innermost lane is a dedicated lane and the outermost lane is an exit ramp; the vehicle model includes a target vehicle and multiple background vehicles, wherein the target vehicle travels in a dedicated lane, the background vehicles are other vehicles in the highway main road model except the target vehicle, and the target vehicle uses a vehicle-road dynamic real-time information interaction vehicle-mounted terminal, while the background vehicles do not use a vehicle-road dynamic real-time information interaction vehicle-mounted terminal; The state parameters include the target vehicle type, the total number of lanes Z, the maximum speed limit and the minimum speed limit of each lane in the main road; The state variables include the traffic flow state and the target vehicle lane change starting distance; the traffic flow state includes the average flow rate of a single lane during the simulation process and the average speed of a single lane The starting distance d s is the longitudinal distance between the target vehicle at the start of lane change in the innermost lane and the tip of the exit ramp diversion area. The end distance d e It is the longitudinal distance between the target vehicle at the position where it completes lane change in the outermost lane and the tip of the exit ramp diversion area.

3. The decision method for changing lanes before a vehicle exits a highway according to claim 2, characterized in that: The method for determining the success rate evaluation model for each of the M groups of state parameters includes: Step a, constructing a highway vehicle driving behavior model and setting state parameters of the highway vehicle driving behavior model; Step b, set multiple groups of state variables and the number of state variable combinations N, and determine the state variables of each group of state parameters and d s ,in, is the average flow rate of a single lane, is the average speed of a single lane, d s is the lane change starting distance of the target vehicle; Step c, for the selected state variables, using the highway vehicle driving behavior model to perform n simulation tests, during each simulation test, by simulating the process of all vehicles on the main road of the highway changing lanes from the dedicated lane to the exit ramp, and simulating to obtain n simulation results, the simulation results include the termination distance and the departure result, wherein the departure result is represented by a Boolean value, if the target vehicle successfully leaves, the Boolean value is 1, if the target vehicle fails to leave, the Boolean value is 0; Step d, according to the n termination distances obtained by multiple simulation tests of the current state variables, a probability density estimation method is used to determine the termination distance probability distribution model p(d e ); Step e: according to the simulation results obtained by multiple simulation tests of the current state parameters, the departure result corresponding to the termination distance is used as a Boolean value label to obtain n termination distance samples, and the termination distance samples are used to train a preset classifier. The training classifier predicts whether the target vehicle successfully leaves the exit ramp based on the termination distance, obtains the termination distance critical value between successfully leaving the exit ramp and unsuccessfully leaving the exit ramp, and determines the termination distance classification boundary value d eb and the uncertainty of the classification boundary σ eb ; Step f, based on the termination distance probability distribution model in S1.4 and the termination distance classification boundary value d in S1.5 eb , determine the success rate and uncertainty of the target vehicle successfully leaving the exit ramp under the current state variables; Step g, judging whether the current number of simulations exceeds the preset number of state variable combinations N, if it has exceeded the preset number of state variable combinations N, then proceeding to step h, otherwise, selecting the next set of state variables and returning to steps c to f to continue the simulation test until it exceeds the preset number of state variable combinations N, and obtaining N simulation data samples, the simulation data samples are represented by (X i ,y i ), where X i is the state variable of the ith simulation test, y i is the success rate of the ith simulation test, and enters step h; Step h: Determine the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model 4. The decision method for changing lanes before a vehicle exits a highway according to claim 3, characterized in that: In step f, the success rate of the target vehicle successfully driving off the exit ramp under the current state parameters is: In the formula, c i is the success rate of the target vehicle successfully leaving the exit ramp in the i-th simulation test; p is the probability distribution model of the termination distance; Z is the total number of lanes, z is the lane number, where z = 1, 2, ..., Z; v zmax is the maximum speed limit of the lane, v zmin is the minimum speed limit for the lane; is the termination distance obtained from the ith simulation test, is the average flow rate of a single lane in the ith simulation test, is the average speed of a single lane in the ith simulation test, is the lane change starting distance of the target vehicle in the ith simulation test, and M is the highway vehicle driving behavior model in the ith simulation test; The uncertainty corresponding to the success rate of the target vehicle successfully leaving the exit ramp under the current state variable is: In the formula, is the uncertainty of the target vehicle successfully leaving the exit ramp in the i-th simulation test.

5. The decision method for changing lanes before a vehicle exits a highway according to claim 4, characterized in that: In the step h, a success rate evaluation model is constructed based on a Gaussian process regression model, wherein the success rate evaluation model is trained using simulation data samples, the state variables are used as input features of the success rate evaluation model, and the success rate is used as the target variable output by the success rate evaluation model; During the training process of the success rate evaluation model, the state variables of the simulation data samples are input into the success rate evaluation model, and the success rate evaluation model is used to calculate the success rate estimation value according to the state variables. The accuracy calculated by the success rate evaluation model is obtained in combination with the success rate corresponding to the simulation data samples, and the accuracy calculated by the success rate evaluation model is compared with the preset accuracy. If the accuracy calculated by the success rate evaluation model does not reach the preset accuracy, the hyperparameters of the success rate evaluation model are optimized and the success rate evaluation model is continued to be trained. Otherwise, the trained success rate evaluation model is obtained.

6. The decision method for changing lanes before a vehicle exits a highway according to claim 5, characterized in that: The success rate evaluation model is: in, The radial basis kernel function k(X, X′) is: Where X and X′ are state variables in the simulation data sample, y(X) is the success prior distribution; m(X) is the mean function; k(X, X′) is the kernel function; is the variance of the signal; l is the length scale parameter, which is used to control the smoothness of the radial basis kernel function; The hyperparameter optimization of the success rate evaluation model adopts the log-likelihood function, which is: Where p(y|X, θ) is the likelihood function; θ is the set of hyperparameters to be optimized; T is the transposed matrix; K is the covariance matrix calculated by the kernel function; is the variance of the noise, y is the success rate sample; The log-likelihood function is optimized based on the gradient descent method. After optimization, for the new input feature X * , predict the new input feature X * The mean μ(X * ) and variance σ 2 (X * )for: Where I is the unit matrix.

7. A lane change decision system for a vehicle before exiting a highway, characterized in that: It includes model selection module, output module and model library component module; The model selection module is configured to determine the state parameters of the road to be decided according to the road conditions to be decided, obtain the traffic flow state of the road to be decided in real time, and select the target success rate evaluation model from the pre-built model library in combination with the success rate and confidence of the target vehicle successfully leaving the exit ramp under the current state parameters; The output module is configured to use the selected target success rate evaluation model to predict the starting distance and the ending distance of the target vehicle successfully leaving the exit ramp in the road to be decided, and determine the starting position of the target vehicle changing lanes before leaving the highway. Wherein, the model selection module is in communication connection with the model library component module, and the model library component module is configured as follows: Constructing a highway vehicle driving behavior model, presetting Q groups of state parameters, setting state variables for each group of state parameters, performing multiple simulation tests using the highway vehicle driving behavior model, obtaining a success rate evaluation model corresponding to the highway vehicle driving behavior model constructed according to each state parameter, and constructing the model library based on the success rate evaluation model corresponding to each state parameter; Wherein, for each of the Q groups of state parameters, the method for determining the success rate evaluation model includes: A plurality of groups of state variables are set under the selected state parameters, and a plurality of simulation tests are carried out using a highway vehicle driving behavior model based on the plurality of groups of state variables to obtain simulation results. The termination distance probability distribution model and the termination distance classification boundary value are determined according to the simulation results of each state variable, the success rate and uncertainty of the target vehicle successfully leaving the exit ramp are calculated, and the success rate evaluation model corresponding to the state parameters of the current highway vehicle driving behavior model is determined.

8. The lane change decision system for a vehicle before exiting a highway according to claim 7, characterized in that: The decision module is provided with an output end, and the output end is preset in the vehicle to be changed lanes, and is used for the vehicle to be changed lanes to exit the expressway according to the lane change starting position.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: The program instructions are executed by the lane change decision system before the vehicle exits the highway to implement the steps of the method described in any one of claims 1 to 6.

Images