Interwoven area fine-grained cellular automaton simulation method considering lane changing probability

Abstract

The invention discloses an interlacing area fine-grained cellular automaton simulation method considering lane change probability, which comprises the following steps: 1) on the basis of a classical NaSch model, introducing aggressive degree characteristic parameters to describe the risk behavior tendency of a driver; (2) submeter-level fine division is conducted on the cellular size in the transverse dimension and the longitudinal dimension, a transverse free lane changing rule of a vehicle under fine-grained division is provided, and a longitudinal updating rule is improved; 3) fitting the lane change probability of the vehicle at different positions of the interweaving area by analyzing real interweaving area trajectory data, 4) replacing a lane change decision model of a traditional cellular automaton model by using a GBDT + LR algorithm; and 5) constructing a complete interlaced area fine-grained cellular automaton model considering the lane changing probability based on the lane changing decision algorithm. According to the method, the behavior difference of the driver, the continuous dynamic nature of the lane changing behavior of the vehicle, the lane changing behavior difference of the vehicle at different positions and the complex diversity of the influence factors of the forced lane changing decision of the driver are fully considered.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic simulation, and in particular to a fine-grained cellular automata simulation method for an weaving area considering lane-changing probability. Background Art

[0002] As a high-class road in the urban internal traffic network, the urban expressway aims to provide a fast, efficient and safe vehicle passing environment. However, the current supply capacity of urban expressways is difficult to meet the growing travel demand, resulting in frequent traffic congestion on urban expressways during peak hours. In addition, due to the limitation of urban land use and high-density traffic demand, exit ramps often follow entrance ramps closely, and the entrance ramp and the exit ramp are directly connected by a short auxiliary lane, forming a typical expressway weaving area. In the expressway weaving area, the concentrated lane-changing behavior of vehicles leads to a decrease in speed and traffic congestion, making the weaving area the main congestion point and accident-prone area of the expressway.

[0003] The weaving area is a key node for vehicles to enter or leave the main line of the expressway. Merging vehicles and diverging vehicles need to complete forced lane changes within a limited space to achieve the purpose of entering or leaving the expressway. The movement directions of these two types of lane-changing behaviors are opposite and the trajectories intersect, making the traffic flow operation in the weaving area highly complex and uncertain. Especially in a short weaving area, the frequent acceleration, deceleration and forced lane-changing behaviors of vehicles make the traffic conditions in the weaving area more complex, which will not only reduce the road passing capacity, but also increase the incidence of traffic accidents, becoming a key factor affecting the safety and efficiency of urban expressways. Therefore, in-depth research on vehicle lane-changing behaviors and traffic conflict phenomena in the weaving area can not only effectively improve the lane-changing efficiency and safety of vehicles, but also help to alleviate the traffic congestion problem of urban expressways and improve the overall efficiency of the urban traffic system.

[0004] The core purpose of studying vehicle lane-changing conflicts is to predict potential vehicle lane-changing conflicts in advance and then formulate effective active guidance measures to minimize the incidence of traffic accidents. In this context, microscopic traffic flow simulation models represented by cellular automata have become key tools for predicting potential vehicle conflicts. The cellular automata model can accurately depict the behavior patterns and decision-making processes of individual vehicles by finely simulating the movement process of vehicles and the complex interaction between vehicles, so as to accurately predict potential conflicts and collision risks in traffic flow. This simulation method can not only deduce the dynamic change trend of traffic flow, but also provide a scientific basis for formulating active control strategies, enabling it to implement effective guidance measures in advance, reduce the risk of traffic accidents, and ensure road traffic safety.

[0005] However, the existing methods still have the following deficiencies:

[0006] 1. The traditional cellular automaton model cannot accurately express key motion parameters such as vehicle speed and acceleration in the setting of cell size, and fails to fully consider the individual behavior differences of drivers. These limitations cause deviations between the simulation results and the actual situation, making it impossible to accurately deduce the changes in the motion state of vehicles in the weaving area. As a result, the lane-changing conflict prediction based on cellular automata is not accurate enough. There is a lack of a refined cellular automaton model that can accurately simulate the microscopic motion characteristics of vehicles to achieve an accurate deduction of the motion of vehicles in the weaving area.

[0007] 2. The traditional cellular automaton model adopts fixed lane-changing rules, which are suitable for simulating free lane-changing behaviors on basic sections. However, such lane-changing rules do not fully consider the complexity of forced lane-changing behaviors of vehicles in the weaving area. In particular, forced lane-changing decisions are largely affected by traffic conditions, vehicle positions, and driver personality characteristics, resulting in the model being unable to accurately reflect the actual lane-changing behaviors of drivers in the weaving area. It is necessary to construct a lane-changing decision model that is more in line with the actual lane-changing characteristics of the weaving area to replace the fixed lane-changing rules of the traditional cellular automaton model, achieve an accurate simulation of the lane-changing behaviors of vehicles in the weaving area, and improve the deduction ability of the cellular automaton model. Summary of the Invention

[0008] The purpose of the present invention is to overcome the above-mentioned shortcomings and deficiencies of the prior art and provide a fine-grained cellular automaton simulation method for the weaving area considering lane-changing probability.

[0009] The present invention proposes a parameter of aggressiveness to characterize the behavior characteristics of drivers, and improves the formula for predicting the dynamic spacing at the next moment of the cellular automaton; divides the cell size into sub-meter-level fine divisions from both horizontal and vertical dimensions, and improves the vehicle motion rules; deeply explores the lane-changing behavior characteristics of vehicles in the weaving area based on actual trajectory data, and extracts the lane-changing probability distribution of the weaving area; constructs a lane-changing decision model for the weaving area based on the GBDT-LR algorithm, realizing a lane-changing decision that is more in line with the psychology and cognitive logic of drivers; the established fine-grained cellular automaton model for the weaving area can more realistically reflect the vehicle motion characteristics of the weaving area.

[0010] The present invention is realized through the following technical solutions:

[0011] A fine-grained cellular automaton simulation method for the weaving area considering lane-changing probability, comprising the following steps:

[0012] S1. On the basis of the classical NaSch model, in order to reflect the behavioral differences of drivers, introduce a characteristic parameter of aggressiveness to describe the risk-taking behavior tendency of drivers. The smaller this parameter is, the more conservative the driver is. Improve the formula for predicting the dynamic spacing at the next moment of the cellular automaton, reflecting the requirements for safety spacing during following driving and the requirements for safety spacing on the target lane during lane-changing by different drivers;

[0013] S2. Subdivide the cell size at the sub - meter level in both the horizontal and vertical dimensions. Refine a classic cell of 7.5m×3.5m into 5×5 fine cells. Introduce the concept of lateral speed considering the continuity and dynamics of vehicle lane - changing behavior. Propose the lateral free lane - changing rule for vehicles under fine - grained division and improve the longitudinal update rule, and propose the complete process of state evolution of the refined cellular automaton;

[0014] S3. By analyzing the trajectory data of the real weaving section, fit the lane - changing probability of vehicles at different positions in the weaving section. Use the cumulative integral of the lane - changing frequency distribution as the lane - changing probability at different positions, replacing the fixed lane - changing probability in the traditional cellular automaton model, so as to more precisely describe the differences in lane - changing behavior of vehicles at different positions in the weaving section;

[0015] S4. Use the GBDT + LR algorithm to replace the lane - changing decision - making model of the traditional cellular automaton model. Through the GBDT algorithm, perform feature selection and combination on the data, and use the LR algorithm to learn and predict the output of GBDT to assist the cellular automaton in making forced lane - changing decisions that conform to the driver's rules. The GBDT - LR lane - changing decision - making model predicts the driver's lane - changing result through a series of eigenvalue (position, speed, acceleration, acceleration of the vehicle in front and behind in the target lane, speed difference with the vehicle in front and behind in the target lane, distance from the vehicle in front and behind in the target lane, etc.). When the prediction result is lane - changing, the model continues to check whether it meets the lane - changing probability distribution of the cell position in that lane;

[0016] S5. Based on the above - mentioned lane - changing decision - making algorithm, construct a complete fine - grained cellular automaton model for the weaving section considering lane - changing probability, and propose specific update rules for merging, diverging, and straight - moving vehicles in the improved cellular automaton model of the weaving section, which can achieve high - precision simulation of the actual forced lane - changing characteristics of vehicles in the weaving section.

[0017] Furthermore, step S1 specifically introduces an aggressiveness characteristic parameter in the formula for estimating the dynamic distance to describe the adventurous behavior tendency of the driver. The steps are as follows:

[0018] S101. Introduce a parameter β∈[0,1]. The aggressiveness β describes the adventurous behavior tendency of the driver. The smaller the parameter β, the more conservative the driver is, the more insensitive the driver is to the speed of the vehicle in front, and the driver tends to maintain a larger safety distance from the vehicle in front; the larger the parameter β, the more aggressive the driver is, and the driver tends to follow the vehicle in front closely;

[0019] S102. The dynamic distance d with the vehicle in front at the next moment in the classic NaSch model is expressed as:

[0020]

[0021] In the formula, d i,j(t) represents the actual distance from the vehicle in front at the current moment, v i,j-1 (t) represents the speed of the vehicle in front at the current moment, represents the desired speed of the current vehicle.

[0022] Desired speed is:

[0023]

[0024] In the formula, v i,j (t) represents the speed of the vehicle at the current moment, v max is the maximum speed, and a is the vehicle acceleration.

[0025] Introduce the aggressiveness parameter β, and the estimated dynamic distance from the vehicle in front at the next moment is expressed as:

[0026]

[0027] In the formula, βv i,j-1 (t) represents the perceived speed of the vehicle with respect to the vehicle in front. The parameter β = 0 means that the driver is very conservative and regards the vehicle in front as stationary. At this time, the estimated dynamic distance from the vehicle in front becomes the static distance; β = 1 means that the driver is very aggressive and can fully perceive the speed of the vehicle in front. At this time, the above rule becomes the classic NaSch model;

[0028] S103. If indicates that the driver believes there will be no collision with the vehicle in front, it can be deduced that

[0029]

[0030] That is, after adding the driver parameter β, the safety condition threshold of the dynamic distance in the model changes from 0 to (1 - β)v i,j-1 (t).

[0031] Furthermore, in step S2, the cell size is finely divided at the sub - meter level in both the horizontal and vertical dimensions, and the lateral free - lane - changing rule and the longitudinal update rule of the vehicle under the fine - grained division are proposed. The specific steps are as follows:

[0032] S201. The cell size is finely divided at the sub - meter level in both the horizontal and vertical dimensions. A classic cell of 7.5m×3.5m is refined into 5×5 fine cells. That is, the size of the fine cell is cellLenth = 1.5m, cellWidth = 0.7m. In the refined cellular automaton model, the lateral width of a vehicle is 2.1m and it occupies 3 fine cells; the longitudinal length is 6m and it occupies 4 fine cells. The position of vehicle j in lane i is represented by the coordinates of the first cell in the front - left of the vehicle (x i,j (t), y i,j(t);

[0033] S202. The free lane-changing rules of the fine-grained cellular automaton are as follows: When the vehicle is in the two side lanes (i = 1, 3), the target lane for lane-changing is the middle lane. When the vehicle is in the middle lane (i = 2), it can change lanes to either the left or right lane. If only one adjacent lane meets the conditions, that lane is the target lane; if the driving conditions of both the left and right lanes meet the conditions, according to the principle that lane-changing to the left on the expressway has priority, the left lane is the target lane for lane-changing. The selection of the target lane is expressed as follows:

[0034]

[0035] where d i_target,j+1 (t) is the distance from the current time to the vehicle behind in the target lane, and v i_target,j+1 (t) is the speed of the vehicle behind in the target lane. is the estimated dynamic distance from the current time to the vehicle behind in the target lane at the next moment. indicates that the safety condition is met;

[0036] S203. During the vehicle driving process, it always hopes to move forward at the desired speed . When the driving speed is hindered by the vehicle in front and the driving conditions of the adjacent lane meet the requirements, a lane-changing motivation will be generated. After the lane-changing motivation is generated, it is necessary to check whether the safety condition and the lane-changing probability are met. The safety condition is that there is no collision with the vehicle behind in the target lane during the lane-changing process. The lane-changing decision conditions are as follows:

[0037]

[0038] where represents the estimated dynamic distance from the current time to the vehicle in front in the adjacent lane at the next moment, and d i±1,j (t) is the actual distance from the current time t to the vehicle C i±1,j-1 in the adjacent lane, and v i±1,j-1 (t) is the speed of the vehicle C i±1,j-1 in the adjacent lane at time t, and P c is the set lane-changing probability;

[0039] S204. When the lane-changing decision result is to execute the lane-changing action, the lateral movement rules of the fine-grained cellular automaton are as follows:

[0040] First, calculate the lateral distance d l from the current lateral position to the center position of the target lane:

[0041] d l = y i_target - (y i,j (t) + 1) (7)

[0042] where dl represents the distance between the vehicle center cell and the target lane center cell, y i_target represents the center position of the target lane. If the lateral speed is less than the lateral distance, the vehicle moves laterally according to the lateral speed; if the lateral speed is greater than the lateral distance, it decelerates to the center of the target lane. As follows:

[0043]

[0044] S205. The longitudinal motion rules of the fine-grained cellular automaton include vehicle acceleration, deterministic deceleration, random slowdown, and position update, where the deterministic deceleration is improved as follows:

[0045]

[0046] Furthermore, in step S3, by analyzing the real weaving section trajectory data, the lane-changing probability of the vehicle at different positions in the weaving section is fitted. The specific steps are as follows:

[0047] S301. Use the unscented Kalman filter algorithm to denoise and smooth the trajectory data, and more accurately estimate the actual state of the vehicle;

[0048] S302. Standardize the lane-changing point position, and use the ratio of the vehicle lane-changing point position to the length of the auxiliary lane in the weaving section as the lane-changing position, so as to define the lane-changing positions of all vehicles between 0 and 1;

[0049] S303. According to the smoothed trajectory data, draw the histograms of the lane-changing point positions and lane-changing frequencies of the merging and diverging lane-changes in the weaving section respectively. The result is that the merging lane-changing positions are unimodally distributed, concentrated in the upper and middle reaches of the weaving section, while the diverging lane-changing positions are relatively dispersed and bimodal;

[0050] S304. Nonlinearly fit the lane-changing frequency distribution functions of the merging and diverging positions respectively. The Gauss distribution model has the best fitting effect on the merging lane-changing position distribution, and the fitting function is as follows:

[0051]

[0052] The result of the Gauss bimodal fitting model for the diverging lane-changing position distribution is:

[0053]

[0054] In the formula, x represents the relative lane-changing position (taking values between 0 and 1), and y represents the corresponding lane-changing frequency (%).

[0055] S305. The lane-changing probability function is expressed as:

[0056] P change (i,k) = H(x)|x＝k (12)

[0057] Wherein, P change (i,k) is the lane-changing probability of the vehicle at position k in lane i, which is the cumulative function of the lane-changing frequency distribution; H(x) is the lane-changing frequency distribution function.

[0058] Furthermore, in step S4, the GBDT+LR algorithm is used to replace the lane-changing decision model of the traditional cellular automaton model.

[0059] The specific steps are as follows:

[0060] S401. Construct a gradient boosting decision tree (GBDT) model for vehicle lane-changing decision-making. Its main idea is to calculate the residual between the predicted value and the true value of the first t-1 trees, and then construct the t-th tree to fit the residual. The final total decision tree is obtained by weighted summation of the decision trees obtained in each round of training.

[0061] The input of the GBDT model of the present invention is a k-dimensional vector x = {x1, x2,..., x k} composed of a series of characteristic variables that affect the driver's lane-changing decision. The output is the predicted driver's lane-changing decision. y = 1 indicates lane-changing, and y = 0 indicates no lane-changing;

[0062] S402. The forced lane-changing decision characteristic variables x = {x1, x2,..., x k} of the present invention include the position ratio in the weaving area, the longitudinal speed and longitudinal acceleration of the lane-changing vehicle, the longitudinal accelerations of the vehicles in front of and behind the target lane, the longitudinal speed difference and longitudinal distance between the lane-changing vehicle and the vehicles in front of and behind the target lane, a total of 9 characteristic parameters.

[0063] So the decision data corresponding to a set of lane-changing decision characteristic parameters X n = {x1, x2,..., x9} is represented as follows:

[0064] (X n , Y n ) = (x n1 , x n2 , x n3 , x n4 , x n5 , x n6 , x n7 , x n8 , x n9 , Y n ) (13) Then the set of characteristic parameters of each lane-changing vehicle in the decision-making process can be represented as:

[0065] {(X n-30 , Y n-30 ), (Xn-20 , Y n-20 ), (X n-10 , Y n-10 ), (X n , Y n )} (14) where n represents the starting time of lane change, X n represents the set of characteristic parameters at time n, Y n represents the lane change decision label at time n, Y n = 1 indicates lane change, Y n-10 = Y n-20 = Y n-30 = 0 indicates no lane change;

[0066] S403. Divide the trajectory data set into two data sets of merging lane change and diverging lane change, extract the above-mentioned forced lane change decision characteristic parameter sets of each vehicle, and randomly divide the two data sets into a training set and a test set according to a ratio of 7:3. GBDT first trains on the original training data set to obtain a trained decision tree. Among them, the tree nodes of Tree 1 to Tree N represent N weak classifiers of GBDT. Traverse N trees. For each tree, there is one and only one leaf node that outputs a prediction result. Record the position of the leaf node to which the predicted probability value calculated by each tree in the model belongs as 1, otherwise take 0, to obtain the encoded feature vector of each original training sample, and construct a new training data;

[0067] S404. Construct a Logistic Regression (LR) linear classifier, add a sigmoid function mapping on the basis of the linear regression analysis result, and perform binary classification to solve the lane change decision result. Input the encoded feature vector in S403 into the LR model for the training of the final classifier;

[0068] S405. After the GBDT-LR lane change decision model trained by the data set can obtain the forced lane change characteristic parameters of the weaving vehicle C i,j at the current time t, make a lane change decision according to the input feature parameters x = {x1, x2,..., x k}, that is, classify the input feature parameters as lane change y = 1 or no lane change y = 0. When the output result of the GBDT-LR model is lane change, check whether it satisfies the lane change probability P change (i, k) of the cell position k in lane i. If it is satisfied, the lane change decision result is lane change; otherwise, the lane change decision result is no lane change.

[0069] Furthermore, in step S5, a complete fine-grained cellular automaton model of the weaving area considering the lane change probability is constructed. The specific steps are as follows:

[0070] S501. In the cellular automaton model of the weaving area, at the moment of initially generating each vehicle, the OD attribute of the vehicle is configured according to the initial lane number and the traffic flow ratios in all directions under this traffic state. The OD attribute of the vehicle represents the driving direction of the vehicle in this weaving area, and each vehicle strictly runs to the target road according to the OD attribute. The vehicle OD attribute allocation method is shown as follows:

[0071]

[0072] In the formula, i0 is the initial lane of the vehicle, P is the random probability rand(0,1), and MM, MR, RM, and RR respectively represent that the vehicle OD attribute is from the main road to the main road, from the main road to the ramp, from the ramp to the main road, and from the ramp to the ramp;

[0073] S502. The merging vehicle (OD = RM) accelerates forward on the ramp and starts to look for suitable lane-changing conditions to merge into the main road when entering the weaving area. If the lane-changing safety conditions are met, the lane-changing result is judged by the GBDT-LR lane-changing decision model and satisfies the lane-changing probability distribution P change (i,k), then the lane change is implemented; otherwise, it moves forward waiting for a lane-changing opportunity. If the merging vehicle still fails to change lanes successfully when it reaches the end of the weaving area, it needs to stop at the latest lane-changing point and wait for the opportunity to merge;

[0074] S503. The weaving lane change of the diverging vehicle (OD = MR) is carried out between the outer lane of the main road and the ramp. The vehicle needs to enter the outer lane of the main road in advance before performing the weaving lane change. The process of changing the diverging vehicle from the inner or middle lane of the main road to the outer lane of the main road is called the preparation for weaving process. If the lane-changing safety conditions are met, the lane-changing result is judged by the GBDT-LR lane-changing decision model and satisfies the lane-changing probability distribution P change (i,k), then the lane change is implemented; otherwise, it moves forward waiting for a lane-changing opportunity. There is also a latest lane-changing point for the weaving lane change of the diverging vehicle, that is, when the vehicle reaches the end of the weaving area and still has not completed the weaving lane change, it will stop and wait for a lane-changing opportunity;

[0075] S504. After the straight-through vehicle on the main road (OD = MM) enters the weaving area, in order to avoid the influence caused by the lane change of the merging vehicle, it tends to drive in the inner lane of the main road. Therefore, in the cellular automaton model of the weaving area, the straight-through vehicle on the main road is restricted to change lanes unidirectionally. For the straight-through vehicle in the middle lane of the main road, if the driving conditions in the left lane are better than those in the middle lane and the safety conditions are met, the vehicle is allowed to change lanes to the left; for the straight-through vehicle in the left lane of the main road, it is prohibited to change lanes to the right in the weaving area;

[0076] S505. The flow ratio of the straight-through vehicle on the ramp (OD = RR) is very small. This part of the vehicle only moves forward on the ramp in the weaving area and does not perform lane-changing operations.

[0077] The present invention has the following advantages and effects compared with the prior art:

[0078] 1) The present invention establishes a fine-grained cellular automaton simulation model considering driver behavior characteristics. For the first time, a driver aggressiveness parameter is introduced and innovatively incorporated into the dynamic spacing calculation formula, effectively expressing the influence of driving behavior on the microscopic motion pattern of vehicles. At the same time, through the sub-meter-level fine division of the cell size in the horizontal and vertical directions and the improvement of the motion rules, the model realizes the simulation of the vehicle's close following and continuous lane-changing processes, enhances the system's maximum flow processing capacity and speed stability, and improves the simulation accuracy of the cellular automaton model.

[0079] 2) The present invention proposes a lane-changing decision-making method based on actual lane-changing probability and ensemble learning. Aiming at the limitation of the fixed lane-changing rule in the traditional cellular automaton model, this paper uses real trajectory data in the weaving area to extract the confluence and divergence lane-changing probability distributions of vehicles in the actual scenario, and replaces the fixed lane-changing probability with this. Combining with the GBDT-LR model, by learning the influence of characteristic parameters in a large amount of real data, a lane-changing decision more in line with the actual thinking of drivers is simulated, making the simulation results more in line with the lane-changing characteristics of the actual weaving area.

[0080] 3) The present invention is a further promotion and application of cellular automata in the field of traffic simulation of vehicle lane-changing behavior in the weaving area, which can achieve high-precision and fast simulation of the actual forced lane-changing characteristics of vehicles in the weaving area, and can provide technical model support for fields such as simulation analysis of forced lane-changing behavior in the weaving area, lane-changing conflict prediction, and lane-changing guidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 is a trajectory comparison diagram of the lane-changing process of the fine-grained cellular automaton in the embodiment of step S2 of the present invention;

[0082] Figure 2 Trajectory comparison before and after smoothing in step S301 of the present invention;

[0083] Figure 3 Lane-changing frequency distribution function fitting diagram in step S304 of the present invention;

[0084] Figure 4 is the overall framework of the fine-grained cellular automaton simulation method for the weaving area considering lane-changing probability of the present invention;

[0085] Figure 5 is a schematic diagram of the estimated dynamic spacing introducing the aggressiveness β of the present invention;

[0086] Figure 6 is a schematic diagram of the fine-grained cellular automaton in the weaving area of the present invention;

[0087] Figure 7It is the flow chart of the evolution process of the fine-grained cellular automaton vehicle driving state in the weaving area of the present invention;

[0088] Figure 8 It is the training schematic diagram of the GBDT-LR lane-changing decision-making model of the present invention;

[0089] Figure 9 It is the structural schematic diagram of the GBDT-LR lane-changing decision-making model of the present invention;

[0090] Figure 10 It is the schematic diagram of the forced lane-changing decision-making model in the fine-grained cellular automaton in the weaving area considering the lane-changing probability of the present invention;

[0091] Figure 11 It is the schematic diagram of the fine-grained cellular automaton in the weaving area of the embodiment of the present invention;

[0092] Figure 12 It is the flow chart of the merging vehicle update process of the fine-grained cellular automaton in the weaving area of the present invention;

[0093] Figure 13 It is the flow chart of the diverging vehicle update process of the fine-grained cellular automaton in the weaving area of the present invention;

[0094] Figure 14 It is the flow chart of the main road straight-going vehicle update process of the fine-grained cellular automaton in the weaving area of the present invention. Detailed implementation manners

[0095] The present invention will be further described in detail below in conjunction with specific embodiments.

[0096] Such as Figure 4 It is the overall framework of the present invention, which describes the overall framework of the fine-grained cellular automaton simulation method in the weaving area considering the lane-changing probability, and includes the following steps:

[0097] Step S1: On the basis of the classical NaSch model, in order to reflect the behavioral differences of drivers, a radical degree characteristic parameter is introduced to describe the adventurous behavior tendency of drivers, and the formula for predicting the dynamic distance at the next moment of the cellular automaton is improved, reflecting the requirements for the safety distance of different drivers during following driving and the requirements for the safety distance of the target lane during lane-changing.

[0098] Furthermore, step S1 includes the following sub-steps, and the predicted dynamic distance introducing the radical degree β is as Figure 5 shown.

[0099] S101. Introduce a parameter β ∈ [0, 1]. The degree of aggressiveness β describes the driver's tendency towards risky behavior. The smaller the parameter β, the more conservative the driver is, with a slower perception of the speed of the vehicle ahead and a tendency to maintain a larger safety distance from the vehicle ahead; the larger the parameter β, the more aggressive the driver is, and the driver is more inclined to follow the vehicle ahead closely.

[0100] S102. The dynamic distance from the vehicle ahead at the next moment in the classic NaSch model is expressed as:

[0101]

[0102] In the formula, d i,j (t) represents the actual distance from the vehicle ahead at the current moment, v i,j-1 (t) represents the vehicle speed of the vehicle ahead at the current moment, represents the desired speed of the current vehicle.

[0103] Desired speed is:

[0104]

[0105] In the formula, v i,j (t) represents the vehicle speed at the current moment, v max is the maximum speed, and a is the vehicle acceleration.

[0106] Introduce the aggressiveness parameter β. The estimated dynamic distance from the vehicle ahead at the next moment is expressed as:

[0107]

[0108] In the formula, βv i,j-1 (t) represents the perceived speed of the vehicle with respect to the vehicle ahead. The parameter β = 0 indicates that the driver is very conservative and regards the vehicle ahead as stationary. At this time, the estimated dynamic distance from the vehicle ahead becomes the static distance; β = 1 indicates that the driver is very aggressive and can fully perceive the speed of the vehicle ahead. At this time, the above rule becomes the classic NaSch model.

[0109] S103. If indicates that the driver believes that there will be no collision with the vehicle ahead, it can be derived that

[0110]

[0111] That is, after adding the driver parameter β, the safety condition threshold of the dynamic distance in the model changes from 0 to (1 - β)v i,j-1 (t).

[0112] Step S2: Sub - meter - level fine - divide the cell size in both the horizontal and vertical dimensions. Refine a classic cell of 7.5m×3.5m into 5×5 fine cells. Considering the continuity and dynamics of vehicle lane - changing behavior, introduce the concept of horizontal speed, propose the horizontal free - lane - changing rule for vehicles under fine - grained division, and improve the vertical update rule, and propose the complete state - evolution process of the refined cellular automaton. The specific steps are as follows:

[0113] S201: Sub - meter - level fine - divide the cell size in both the horizontal and vertical dimensions. Refine a classic cell of 7.5m×3.5m into 5×5 fine cells, that is, the size of the fine cell is cellLenth = 1.5m, cellWidth = 0.7m. In the refined cellular automaton model, the lateral width of a vehicle is 2.1m, occupying 3 fine cells; the longitudinal length is 6m, occupying 4 fine cells. The position of vehicle j in lane i is represented by the coordinates of the first cell at the vehicle's left front (x i,j (t), y i,j (t). The division of the fine - grained cellular automaton is as Figure 6 shown;

[0114] S202: The free - lane - changing rule of the fine - grained cellular automaton is as follows: When the vehicle is in the two side lanes (i = 1, 3), the target lane for lane - changing is the middle lane. When the vehicle is in the middle lane (i = 2), it can change lanes to both the left and right lanes. If only one adjacent lane meets the conditions, then that lane is the target lane; if the driving conditions of both the left and right lanes meet the conditions, according to the principle that lane - changing to the left is preferred on the expressway, the left lane is the target lane for lane - changing. The selection of the target lane is expressed as follows:

[0115]

[0116] where d i_target,j+1 (t) is the distance from the vehicle at the current moment to the vehicle behind in the target lane, v i_target,j+1 (t) is the speed of the vehicle behind in the target lane, is the estimated dynamic distance from the vehicle to the vehicle behind in the target lane at the next moment, indicates meeting the safety conditions;

[0117] S203: During the vehicle's driving process, it always hopes to move forward at the desired speed . When the driving speed is hindered by the vehicle in front and the driving conditions of the adjacent lane meet the requirements, a lane - changing motivation will be generated. After the lane - changing motivation is generated, it is necessary to check whether the safety conditions and the lane - changing probability are met. The safety condition is that there is no collision with the vehicle behind in the target lane during the lane - changing process. The lane - changing decision conditions are as follows:

[0118]

[0119] Wherein, represents the estimated dynamic distance from the vehicle in the adjacent lane at the next moment, and d i±1,j (t) is the actual distance from the vehicle in the adjacent lane ahead C i±1,j-1 at time t, and v i±1,j-1 (t) is the speed of the vehicle in the adjacent lane ahead C i±1,j-1 at time t, and P c is the set lane-changing probability;

[0120] S204. When the lane-changing decision result is to execute the lane-changing action, the lateral motion rule of the fine-grained cellular automaton is as follows: First, calculate the lateral distance d l between the current lateral position and the center position of the target lane:

[0121] d l = y i_target - (y i,j (t) + 1) (7)

[0122] Wherein, d l represents the distance between the vehicle center cell and the target lane center cell, and y i_target represents the center position of the target lane. If the lateral speed is less than the lateral distance, perform lateral motion according to the lateral speed; if the lateral speed is greater than the lateral distance, decelerate to the center of the target lane. As follows:

[0123]

[0124] S205. The longitudinal motion rules of the fine-grained cellular automaton include vehicle acceleration, deterministic deceleration, random slowdown, and position update, where the deterministic deceleration is improved as:

[0125]

[0126] The complete state evolution flowchart of the fine-grained cellular automaton is as Figure 7 shown.

[0127] According to the above method, use the fine-grained cellular automaton to simulate the vehicle lane-changing process in a high-density scenario, and the obtained lateral trajectory details are as Figure 1 shown.

[0128] Step S3. By analyzing the real weaving section trajectory data, fit the lane-changing probability of vehicles at different positions in the weaving section, and use the cumulative integral of the lane-changing frequency distribution as the lane-changing probability at different positions to replace the fixed lane-changing probability in the traditional cellular automaton model, so as to more precisely describe the lane-changing behavior differences of vehicles at different positions in the weaving section. The specific steps are as follows:

[0129] S301. Use the unscented Kalman filter algorithm to denoise and smooth the trajectory data, and more accurately estimate the actual state of the vehicle.Figure 2 It is the comparison of the vehicle trajectories and speeds before and after the smoothing operation.

[0130] S302. Standardize the lane-changing point positions. Take the ratio of the vehicle lane-changing point position to the length of the auxiliary lane in the weaving area as the lane-changing position, so as to define the lane-changing positions of all vehicles between 0 and 1.

[0131] S303. According to the smoothed trajectory data, respectively draw the histograms of the lane-changing point positions and lane-changing frequencies for the merging and diverging lane-changes in the weaving area. The result is that the merging lane-changing positions show a single-peak distribution, concentrated in the upper and middle reaches of the weaving area, while the diverging lane-changing positions are relatively dispersed and show a double-peak state.

[0132] S304. Non-linearly fit the lane-changing frequency distribution functions for the merging and diverging positions respectively. The Gauss distribution model has the best fitting effect on the distribution of the merging lane-changing positions, and the fitting function is as follows:

[0133]

[0134] The result of the Gauss double-peak fitting model for the diverging lane-changing position distribution is:

[0135]

[0136] In the formula, x represents the relative lane-changing position (taking values between 0 and 1), and y represents the corresponding lane-changing frequency (%).

[0137] The fitting images of the lane-changing frequency distribution functions for merging and diverging are as Figure 3 shown:

[0138] S305. The lane-changing probability function is expressed as:

[0139] P change (i,k) = H(x) x＝k (12)

[0140] In the formula, P change (i,k) is the lane-changing probability of the vehicle at the position k in the i-th lane, which is the cumulative function of the lane-changing frequency distribution; H(x) is the lane-changing frequency distribution function.

[0141] Step S4. Use the GBDT+LR algorithm to replace the lane-changing decision-making model of the traditional cellular automaton model. Through the GBDT algorithm, feature selection and combination are performed on the data, and the LR algorithm learns and predicts the output of the GBDT to assist the cellular automaton in making a forced lane-changing decision that conforms to the driver's pattern. When the prediction result is a lane change, the model continues to check whether it meets the lane-changing probability distribution of the cell position in this lane. The specific steps are as follows:

[0142] S401. Construct a Gradient Boosting Decision Tree (GBDT) model for vehicle lane - changing decision - making. Its main idea is to calculate the residual between the predicted values of the first t - 1 trees and the true values, and then construct the t - th tree to fit the residual. The final total decision tree is obtained by weighted summation of the decision trees obtained in each round of training.

[0143] The input of the GBDT model of the present invention is a k - dimensional vector x = {x1, x2,..., x k} composed of a series of characteristic variables that affect the driver's lane - changing decision. The output is the predicted driver's lane - changing decision. y = 1 indicates lane - changing, and y = 0 indicates no lane - changing. The training schematic diagram of the GBDT lane - changing decision model is as Figure 8 shown;

[0144] S402. The forced lane - changing decision characteristic variables x = {x1, x2,..., x k} of the present invention include the weaving - area position ratio, the longitudinal speed and longitudinal acceleration of the lane - changing vehicle, the longitudinal accelerations of the vehicles in front of and behind the target lane, the longitudinal speed differences and longitudinal distances between the lane - changing vehicle and the vehicles in front of and behind the target lane, a total of 9 characteristic parameters, as shown in Table 1.

[0145] So the decision data corresponding to a set of lane - changing decision characteristic parameters X n = {x1, x2,..., x9} is represented as follows:

[0146] (X n , Y n ) = (x n1 , x n2 , x n3 , x n4 , x n5 , x n6 , x n7 , x n8 , x n9 , Y n ) (13) Then the characteristic parameter set of each lane - changing vehicle in the decision - making process can be represented as:

[0147] {(X n-30 , Y n-30 ), (X n-20 , Y n-20 ), (X n-10 , Y n-10 ), (X n , Y n )} (14) where n represents the starting moment of lane - changing, X n represents the characteristic parameter set at the n - th moment, Y n represents the lane - changing decision label at the n - th moment, Y n = 1 indicates lane - changing, Yn-10 = Y n-20 = Y n-30 = 0 indicates no lane change;

[0148] Table 1 Forced Lane Change Decision Feature Parameters

[0149]

[0150] S403. Divide the trajectory dataset into two datasets for merging lane change and splitting lane change, extract the above-mentioned forced lane change decision feature parameter sets of each vehicle, and randomly divide the two datasets into a training set and a test set according to a ratio of 7:3. GBDT first trains on the original training dataset to obtain a trained decision tree. Among them, the tree nodes from Tree 1 to Tree N represent N weak classifiers of GBDT. Traverse N trees. For each tree, there is exactly one leaf node that outputs the prediction result. Denote the position of the leaf node to which the predicted probability value calculated by each tree in the model belongs as 1, otherwise take 0, to obtain the encoded feature vector of each original training sample, and construct a new training data;

[0151] S404. Construct a Logistic Regression (LR) linear classifier, add a sigmoid function mapping on the basis of the linear regression analysis result, and perform binary classification to solve the lane change decision result. Input the encoded feature vector in S403 into the LR model for the training of the final classifier. The structural schematic diagram of the GBDT-LR lane change decision model is as Figure 9 shown.

[0152] S405. After obtaining the forced lane change feature parameters of the weaving vehicle C at the current moment t, the trained GBDT-LR lane change decision model based on the dataset can make a lane change decision according to the input feature parameters x = {x1, x2,..., x i,j}, that is, classify the input feature parameters as lane change y = 1 or no lane change y = 0. When the output result of the GBDT-LR model is a lane change, check whether it satisfies the lane change probability P k}(i, k) of the cell position k in lane i. If it is satisfied, the lane change decision result is a lane change; otherwise, the lane change decision result is no lane change. The above GBDT-LR forced lane change decision model process considering the lane change probability is as change shown. Figure 10 shown.

[0153] Step S5. Based on the above lane change decision algorithm, a complete fine-grained cellular automaton model for the weaving area considering the lane change probability is constructed, and specific update rules for merging, splitting, and straight-going vehicles in the improved cellular automaton model for the weaving area are proposed, which can achieve high-precision simulation of the actual forced lane change characteristics of vehicles in the weaving area. The specific steps are as follows:

[0154] S501. In the cellular automaton model of the weaving area of the embodiment of the present invention as shown in Figure 11 , the OD attributes of vehicles are configured according to the initial lane numbers and the traffic flow ratios in each direction under this traffic state at the moment of initially generating each vehicle. The OD attribute of a vehicle represents the driving direction of the vehicle in this weaving area, and each vehicle strictly runs to the target road according to the OD attribute. The vehicle OD attribute allocation method is shown as follows:

[0155]

[0156] In the formula, i0 is the initial lane of the vehicle, P is the random probability rand(0,1), MM, MR, RM, and RR respectively represent that the vehicle OD attribute is from the main road to the main road, from the main road to the ramp, from the ramp to the main road, and from the ramp to the ramp;

[0157] S502. The merging vehicle (OD = RM) accelerates forward on the ramp and starts to look for suitable lane-changing conditions to merge into the main road when entering the weaving area. If the lane-changing safety conditions are met, the lane-changing result is judged by the GBDT-LR lane-changing decision model and satisfies the lane-changing probability distribution P change (i,k), then the lane change is implemented; otherwise, it drives forward to wait for a lane-changing opportunity. If the merging vehicle still fails to change lanes successfully when it reaches the end of the weaving area, it needs to stop at the latest lane-changing point to wait for the merging opportunity. The flow chart of the merging vehicle update process is as shown in Figure 12 ;

[0158] S503. The weaving lane change of the diverging vehicle (OD = MR) is carried out between the outer lane of the main road and the ramp. The vehicle needs to enter the outer lane of the main road in advance before performing the weaving lane change. The process of changing the diverging vehicle from the inner or middle lane of the main road to the outer lane of the main road is called the preparation for weaving process. If the lane-changing safety conditions are met, the lane-changing result is judged by the GBDT-LR lane-changing decision model and satisfies the lane-changing probability distribution P change (i,k), then the lane change is implemented; otherwise, it drives forward to wait for a lane-changing opportunity. There is also a latest lane-changing point for the weaving lane change of the diverging vehicle, that is, when the vehicle reaches the end of the weaving area and still has not completed the weaving lane change, it will stop to wait for a lane-changing opportunity. The flow chart of the diverging vehicle update process is as shown in Figure 13 ;

[0159] S504. After the straight-through vehicle on the main road (OD = MM) enters the weaving area, in order to avoid the influence caused by the lane change of the merging vehicle, it tends to drive in the inner lane of the main road. Therefore, in the cellular automaton model of the weaving area, the straight-through vehicle on the main road is restricted to change lanes unidirectionally. For the straight-through vehicle in the middle lane of the main road, if the driving conditions in the left lane are better than those in the middle lane and the safety conditions are met, the vehicle is allowed to change lanes to the left; for the straight-through vehicle in the left lane of the main road, it is prohibited to change lanes to the right in the weaving area. The flow chart of the straight-through vehicle update process on the main road is as shown in Figure 14 ;

[0160] For S505, the flow proportion of ramp straight-through vehicles (OD = RR) is very small. These vehicles only move forward on the ramp within the weaving area and do not perform lane-changing operations.

[0161] As described above, the present invention can be preferably implemented.

[0162] The implementation manners of the present invention are not limited by the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement manners and are all included in the protection scope of the present invention.

Claims

1. A fine-grained cellular automaton simulation method for weaving areas considering lane-changing probability, characterized in that The steps include: S1. The aggressiveness characteristic parameter is introduced to describe the driver's risk-taking behavior tendency. The smaller the parameter, the more conservative the driver is. The formula for estimating the dynamic distance at the next moment of the cellular automation is improved to reflect the requirements of different drivers for safe distance during the following driving process and the requirements for safe distance of the target lane during lane changing. S2. The cell size is divided into sub-meter level in both horizontal and vertical dimensions. A classic cell of 7.5m×3.5m is refined into 5×5 fine cells. The concept of lateral speed is introduced considering the continuity and dynamics of the vehicle lane-changing behavior. The lateral free lane-changing rules of vehicles under fine-grained division are proposed and the longitudinal update rules are improved. The complete state evolution process of the refined cellular automaton is proposed. S3. By analyzing the real weaving area trajectory data, the lane changing probability of vehicles at different positions in the weaving area is fitted, and the cumulative integral of the lane changing frequency distribution is used as the lane changing probability at different positions to replace the fixed lane changing probability in the traditional cellular automaton model, thereby accurately describing the differences in lane changing behaviors of vehicles at different positions in the weaving area; S4. Use GBDT+LR algorithm to replace the lane-changing decision model of the traditional cellular automaton model. The GBDT algorithm is used to select and combine features of the data. The LR algorithm is used to learn and predict the output of GBDT, and the cellular automaton is assisted to make a forced lane-changing decision that conforms to the driver's rules. The GBDT-LR lane-changing decision model predicts the driver's lane-changing result through a series of eigenvalues. When the predicted result is a lane-changing, the model continues to check whether the lane-changing probability distribution of the cellular position of the lane is satisfied. S5. Based on the above lane-changing decision algorithm, a complete fine-grained cellular automaton model of the weaving area taking into account the lane-changing probability is constructed, and specific update rules for merging, diverging, and straight-going vehicles in the improved cellular automaton model of the weaving area are proposed, which can achieve high-precision simulation of the actual forced lane-changing characteristics of vehicles in the weaving area.

2. The fine-grained cellular automation simulation method for weaving areas considering lane-changing probability according to claim 1 is characterized in that: In step S1, an aggressiveness characteristic parameter is introduced into the formula for estimating the dynamic distance to describe the driver's risk-taking behavior tendency. The specific steps are as follows: S101. Introduce parameter β∈[0,1]. The aggressiveness β describes the driver's risk-taking tendency. The smaller the parameter β, the more conservative the driver is. The driver is slow to perceive the speed of the vehicle ahead and tends to keep a larger safety distance from the vehicle ahead. The larger the parameter β, the more aggressive the driver is and the more likely the driver is to follow the vehicle closely. S102, Dynamic distance between the vehicle in front at the next moment in the classic NaSch model It is expressed as: Where, d i,j (t) represents the actual distance to the vehicle ahead at the current moment, v i,j-1 (t) represents the speed of the preceding vehicle at the current moment, Indicates the expected speed of the current vehicle; Expected speed for: In the formula, v i,j (t) represents the speed of the vehicle at the current moment, v max is the maximum speed, a is the vehicle acceleration; introducing the aggressiveness parameter β, the estimated dynamic distance from the vehicle in front at the next moment is expressed as: In the formula, βv i,j-1 (t) represents the perceived speed of the vehicle in front. Parameter β = 0 means that the driver is very conservative and regards the vehicle in front as stationary. In this case, the estimated dynamic distance to the vehicle in front becomes a static distance. β = 1 means that the driver is very aggressive and can fully perceive the speed of the vehicle in front. In this case, the above rule becomes the classic NaSch model. S103, if This means that the driver believes that there will be no collision with the vehicle in front. That is, after adding the driver parameter β, the safety condition threshold of the dynamic spacing in the model changes from 0 to (1-β)v i,j-1 (t).

3. The fine-grained cellular automaton simulation method for weaving areas considering lane-changing probability according to claim 1 is characterized in that: In step S2, the cell size is finely divided into sub-meter levels in both the horizontal and vertical dimensions, and the lateral free lane changing rules and longitudinal update rules of the vehicle under the fine-grained division are proposed. The specific steps are as follows: S201. Divide the cell size into sub-meter-level fine divisions in both the horizontal and vertical dimensions. A 7.5m×3.5m classic cell is refined into 5×5 fine cells, that is, the size of the fine cell is cellLenth=1.5m, cellWidth=0.7m. In the refined cellular automaton model, a vehicle has a horizontal width of 2.1m, occupying 3 fine cells; a longitudinal length of 6m, occupying 4 fine cells. The coordinates of the first cell in front of the left side of the vehicle represent the position of vehicle j in lane i (x i,j (t),y i,j (t); S202. The free lane-changing rules of the fine-grained cellular machine are as follows: when the vehicle is located in the lanes on both sides (i=1,3), the target lane for lane change is the middle lane; when the vehicle is located in the middle lane (i=2), both the left and right lanes can be changed. If only one adjacent lane meets the conditions, then this lane is the target lane; if the driving conditions of both the left and right lanes meet the conditions, according to the principle of priority for lane change to the left on the expressway, the left lane is the target lane for lane change; the selection of the target lane is expressed as follows: Where, d i_target,j+1 (t) is the distance between the vehicle behind the target lane at the current moment, v i_target,j+1 (t) is the speed of the vehicle behind in the target lane, To estimate the dynamic distance between the vehicle behind the target lane at the next moment, Indicates that safety conditions are met; S203, the vehicle always moves at the desired speed When the driving speed is hindered by the vehicle in front and the driving conditions of the adjacent lane meet the conditions, the lane change motivation will be generated; when the lane change motivation is generated, it is necessary to check whether the safety conditions and lane change probability are met; the safety condition is that there is no collision with the vehicle behind the target lane during the lane change process; the lane change decision conditions are as follows: In the formula, It indicates the estimated dynamic distance between the vehicle in front of the adjacent lane at the next moment, d i±1,j (t) is the distance between the front vehicle C in the adjacent lane at time t. i±1,j-1 The actual spacing, v i±1,j-1 (t) is the front vehicle C in the adjacent lane i±1,j-1 The speed at time t, P c is the set lane-changing probability; S204: When the lane-changing decision result is to execute the lane-changing action, the lateral motion rule of the fine-grained automatic cellular machine is as follows: First, the lateral distance d between the current lateral position and the center position of the target lane is calculated. l : d l =y i_target -(y i,j (t)+1) (7) Where, d l represents the distance between the center cell of the vehicle and the center cell of the target lane, y i_target Indicates the center position of the target lane; If the lateral speed is less than the lateral distance, the vehicle moves laterally according to the lateral speed; if the lateral speed is greater than the lateral distance, the vehicle decelerates and moves to the center of the target lane. S205. The longitudinal motion rules of the fine-grained automatic cellular machine include vehicle acceleration, deterministic deceleration, random slowing down and position update, among which the deterministic deceleration is improved as follows:

4. The fine-grained cellular automaton simulation method for weaving areas considering lane-changing probability according to claim 1 is characterized in that: In step S3, by analyzing the real weaving area trajectory data, the lane change probability of the vehicle at different positions in the weaving area is fitted; the specific steps are as follows: S301, using an unscented Kalman filter algorithm to perform noise reduction and smoothing on the trajectory data to more accurately estimate the actual state of the vehicle; S302, standardizing the lane change point position, taking the ratio of the lane change point position of the vehicle to the length of the auxiliary road in the weaving area as the lane change position, thereby defining the lane change position of all vehicles between 0 and 1; S303, plotting histograms of lane-changing point positions and lane-changing frequencies of merging and diverging lane-changing in the weaving area according to the smoothed trajectory data, and the result is that the merging lane-changing position is unimodal and concentrated in the middle and upper reaches of the weaving area, while the diverging lane-changing position is relatively dispersed and bimodal; S304, performing nonlinear fitting on the lane-changing frequency distribution functions of the merging and diverging positions respectively. The Gauss distribution model has the best fitting effect on the distribution of the merging and lane-changing positions, and the fitting function is as follows: The Gauss bimodal fitting model results of the divergence lane-changing position distribution are: Where x represents the relative position of lane change (value between 0 and 1), and y represents the corresponding lane change frequency (%); S305. The lane change probability function is expressed as: P change (i,k)=H(x)| x＝k (12) Where P change (i,k) is the lane-changing probability of the vehicle at position k in lane i, which is the cumulative function of the lane-changing frequency distribution; H(x) is the lane-changing frequency distribution function.

5. The fine-grained cellular automation simulation method for weaving areas considering lane-changing probability according to claim 1 is characterized in that: Step S4 uses the GBDT+LR algorithm to replace the lane-changing decision model of the traditional cellular automaton model; the specific steps are as follows: S401, constructing a gradient boosting decision tree GBDT model for vehicle lane change decision, which is to calculate the residual between the predicted value and the true value of the first t-1 trees, and then construct the tth tree to fit the residual, and the final total decision tree is obtained by weighted summing the decision trees obtained in each round of training; The input of the GBDT model is a k-dimensional vector x={x1,x2,...,x k }, output the predicted lane-changing decision of the driver, y=1 indicates lane-changing, y=0 indicates no lane-changing; S402, lane change decision feature variable x={x1,x2,...,x k }Including the position ratio of the weaving area, the longitudinal speed and longitudinal acceleration of the lane-changing vehicle, the longitudinal acceleration of the front and rear vehicles in the target lane, the longitudinal speed difference and longitudinal distance between the lane-changing vehicle and the front and rear vehicles in the target lane, a total of 9 characteristic parameters; A set of lane-changing decision feature parameters X n ={x1,x2,...,x9} The corresponding decision data is expressed as follows: (X n ,Y n )=(x n1 ,x n2 ,x n3 ,x n4 ,x n5 ,x n6 ,x n7 ,x n8 ,x n9 ,Y n ) (13) Then the characteristic parameter set of each lane-changing vehicle in the decision-making process can be expressed as: {(X n-30 ,AND n-30 ),(X n-20 ,AND n-20 ),(X n-10 ,AND n-10 ),(X n ,AND n )} (14) Where n represents the starting time of lane change, X n represents the characteristic parameter set at time n, Y n represents the lane-changing decision label at time n, Y n =1 means changing lanes, Y n-10 =Y n-20 =Y n-30 =0 means no lane change; S403, dividing the trajectory data set into two data sets of merging lane change and diverging lane change, extracting the above-mentioned forced lane change decision feature parameter set of each vehicle, and randomly dividing the two data sets into a training set and a test set in a ratio of 7:3; GBDT first trains the original training data set to obtain a trained decision tree; wherein the tree nodes from Tree 1 to Tree N represent N weak classifiers of GBDT, traversing N trees, for each tree, there is only one leaf node outputting the prediction result, and the leaf node position to which the prediction probability value calculated by each tree in the model belongs is recorded as 1, otherwise it is taken as 0, and the feature vector after encoding of each original training sample is obtained to construct new training data; S404, constructing a logistic regression linear classifier, adding a sigmoid function mapping based on the linear regression analysis result, and performing a binary classification solution on the lane change decision result; The feature vector encoded by S403 is input into the LR model to train the final classifier; S405, the GBDT-LR lane-changing decision model trained with the data set can obtain the weaving vehicle C at the current time t i,j After the forced lane change characteristic parameters are obtained, according to the input characteristic parameters x={x1,x2,...,x k } Make a lane-changing decision, i.e. classify the input feature parameters as lane-changing y=1 or not changing lanes y=0; When the GBDT-LR model outputs a lane change, check whether the lane change probability P of the cell position k in lane i is satisfied. change (i, k), if it is satisfied, the lane changing decision result is lane changing; otherwise, the lane changing decision result is not lane changing.

6. The fine-grained cellular automation simulation method for weaving areas considering lane-changing probability according to claim 1 is characterized in that: Step S5 constructs a complete fine-grained cellular automaton model of the weaving area taking into account the lane-changing probability; the specific steps are as follows: S501. In the cellular automaton model of the weaving area, when each vehicle is initially generated, the vehicle OD attribute is configured according to the initial lane number and the ratio of the traffic volume in each direction under the traffic state; the OD attribute of the vehicle indicates the driving direction of the vehicle in the weaving area, and each vehicle runs to the target road according to the OD attribute; the vehicle OD attribute allocation method is as follows: Where i0 is the initial lane of the vehicle, P is the random probability rand(0,1), MM, MR, RM, and RR represent the vehicle OD attributes from main road to main road, main road to ramp, ramp to main road, and ramp to ramp, respectively; S502: The merging vehicle (OD=RM) accelerates on the ramp and begins to look for suitable lane-changing conditions when entering the weaving area to merge into the main road. If the lane-changing safety conditions are met, the lane-changing result is determined by the GBDT-LR lane-changing decision model and the lane-changing probability distribution P is met. change (i,k), then change lanes; otherwise, drive forward and wait for a lane-changing opportunity. If the merging vehicle fails to change lanes after driving to the end of the weaving area, it needs to stop at the latest lane-changing point and wait for the opportunity to merge. S503: The weaving lane change of the diverging vehicle (OD=MR) is performed between the outer lane of the main road and the ramp. The vehicle needs to enter the outer lane of the main road in advance before the weaving lane change. The process of changing the diverging vehicle from the inner lane or middle lane of the main road to the outer lane of the main road is called the preparation for weaving process. If the lane change safety conditions are met, the lane change result is determined by the GBDT-LR lane change decision model, and the lane change probability distribution P is satisfied. change (i,k), then change lanes; otherwise, drive forward and wait for the opportunity to change lanes; There is also a latest lane-changing point for the weaving lane-changing of diverted vehicles, that is, when a vehicle reaches the end of the weaving area and has not yet completed the weaving lane-changing, it will stop and wait for the opportunity to change lanes; S504. After entering the weaving area, straight vehicles on the main road tend to drive in the inner lane of the main road to avoid the impact of merging vehicles changing lanes. Therefore, the cellular automation model of the weaving area restricts straight vehicles on the main road from changing lanes in one direction. For straight vehicles on the middle lane of the main road, if the driving conditions in the left lane are better than those in the middle lane and safety conditions are met, the vehicle is allowed to change lanes to the left; straight vehicles on the left lane of the main road are prohibited from changing lanes to the right in the weaving area; S505: The traffic volume of vehicles going straight on the ramp accounts for a very small proportion. These vehicles only move forward on the ramp in the weaving area and do not perform lane changing operations.

Images