Highway on-ramp merging control method considering human driving uncertainty

Abstract

The present application relates to a kind of considering the merging control method of highway on-ramp of human driving uncertainty, belong to intelligent traffic control technical field.The method includes: S1: construct soft set model, the uncertain factor in the driving characteristics of human driving is quantified and characterized;S2: according to NGSIM data set, obtain the vehicle trajectory data of highway on-ramp merging area, the parameter in soft set model is calibrated by vehicle trajectory data;S3: using " virtual rotation " method, the merging control problem is converted into follow car control problem, and follow car control model is established;S4: based on soft set model, the consideration of human driving uncertainty factor is introduced into control model, and the merging control strategy considering human driving uncertainty is proposed;S5: according to the proposed merging control strategy, by SUMO traffic simulation software, set up highway on-ramp merging scene, the automatic driving vehicle in scene is controlled and model parameter is adjusted.

Description

Technical Field

[0001] This invention belongs to the field of intelligent traffic control technology and relates to a method for merging control of highway entrance ramps that takes into account the uncertainty of human driving. Background Technology

[0002] Highway entrance ramps are crucial connecting hubs between urban traffic and the main highway, and their merging control efficiency directly impacts segment capacity, traffic safety, and energy consumption. With the development of intelligent transportation systems (ITS), traditional control methods have gradually revealed their inadequacy in adapting to dynamic traffic environments and uncertainties. Early ramp control primarily relied on fixed traffic lights or timed adjustment strategies based on historical traffic flow data, such as periodically releasing ramp vehicles to match the theoretical capacity of the main road traffic flow. However, such static control models struggle to cope with nonlinear fluctuations in traffic flow, such as sudden congestion during peak hours or sudden changes in traffic flow caused by accidents. Studies show that when the main road traffic density exceeds a critical value, fixed-cycle control leads to a surge in ramp queue lengths, even triggering cascading congestion on urban roads. Although subsequent improved inductive dynamic control methods (such as adjusting ramp release frequency based on real-time main road traffic flow) have improved flexibility to some extent, their reliance on linear threshold models still cannot effectively handle complex factors such as driver behavior variations, sensor noise interference, and changes in weather conditions.

[0003] In recent years, advancements in vehicle-road cooperative technologies have driven the development of merging control methods based on vehicle trajectory planning, such as using Model Predictive Control (MPC) to optimize the coordinated movements of ramp vehicles and main road traffic. These methods acquire vehicle state information through V2X communication and generate precise merging gap and speed guidance strategies based on deterministic assumptions. However, in practical applications, issues such as communication latency, positioning drift, and data packet loss lead to significant differences between theoretical models and real-world scenarios. Some patents attempt to quantify single uncertainties using probabilistic models (such as Gaussian processes or Monte Carlo simulations), but fail to establish a mathematical representation of the multi-factor coupling propagation, thus limiting the robustness of the control strategy. For example, some methods only perform stochastic optimization for vehicle arrival time distributions without incorporating the dynamic errors of the sensing system into the decision-making framework, potentially causing the generated merging commands to fail in extreme scenarios.

[0004] On the other hand, merging control is essentially a multi-objective optimization problem, requiring a dynamic balance among dimensions such as traffic efficiency, safety, and fuel economy. Traditional methods often use weighted summation to transform multiple objectives into a single-objective function, such as minimizing average vehicle delay as the core indicator. However, such simplified methods are difficult to adapt to the dynamic changes in the weights of each objective under uncertain scenarios. For example, when the traffic density on the main road increases sharply, the priority of the safety objective needs to be significantly increased, while efficiency optimization may dominate in low-density scenarios. Existing patents using rule-based or fixed weight coefficient strategies lack adaptive adjustment mechanisms, resulting in large fluctuations in control performance under different operating conditions. Furthermore, although end-to-end control methods based on deep learning can implicitly learn traffic flow characteristics, their black-box nature leads to a lack of interpretability in the decision-making logic, making it difficult to verify reliability in edge scenarios such as sensor anomalies or missing data. For example, some neural network control models may output high-risk merging commands under extreme weather conditions with insufficient training data coverage, and safety verification cannot be performed using traditional formal methods.

[0005] In summary, existing highway entrance ramp merging control methods still suffer from significant technical bottlenecks in uncertainty modeling, multi-factor coupling analysis, and dynamic multi-objective optimization. Specifically, these bottlenecks manifest as insufficient joint anti-interference capabilities against traffic flow abrupt changes, perception errors, and communication disturbances, as well as a lack of adaptive decision-making mechanisms for complex scenarios. Therefore, researching a novel highway entrance ramp merging control method to improve merging efficiency and safety is of paramount importance. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a highway entrance ramp merging control method that considers the uncertainties of human driving, constructs a control framework that can quantify the multi-source uncertainty propagation process, realize multi-objective dynamic trade-offs and has high real-time performance, and improves the merging efficiency and safety of entrance ramps.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for merging traffic at highway entrance ramps that considers the uncertainties of human driving includes the following steps:

[0009] S1: Construct a soft set model to quantitatively characterize the uncertainties in human driving characteristics;

[0010] S2: Obtain vehicle trajectory data of the merging zone of the highway entrance ramp based on the NGSIM dataset, and calibrate the parameters in the soft set model using the vehicle trajectory data;

[0011] S3: The merging control problem is transformed into a following control problem using the "virtual rotation" method, and a following control model is established;

[0012] S4: Based on the soft set model, the uncertainty of human driving is introduced into the control model, and a merging control strategy that takes into account the uncertainty of human driving is proposed.

[0013] S5: Based on the proposed merging control strategy, a merging scenario at the entrance ramp of a highway is built using SUMO traffic simulation software. Control is applied to the autonomous vehicles in the scenario and the model parameters are adjusted.

[0014] Furthermore, step S1 specifically includes: selecting the driver's judgment on the distance to the vehicle in front, the vehicle type, and the distance to the end of the merging zone as key feature parameters, and constructing a soft set model; each parameter is fuzzified using a trapezoidal membership function.

[0015] Furthermore, in step S1, a soft set model is constructed, specifically a parameterized soft set (F,E) is constructed, where F:E→P(U) is a mapping function that maps each parameter e∈E to the power set of the membership degree of the corresponding interval; the mathematical representation of the soft set is realized through a decision matrix, where the row vectors correspond to the intervals of the actual parameters and the column vectors correspond to the evaluation parameters, i.e., the language values.

[0016] Furthermore, step S2 specifically includes: using relevant data from the merging area of ​​highway entrance ramps in the NGSIM dataset and employing a genetic algorithm to calibrate the data.

[0017] Furthermore, in step S2, the calibration method for the distance data specifically includes the following steps:

[0018] S21: The collected distance data is divided into three categories. Considering that the membership degree of the three types of actions, namely acceleration, maintenance and deceleration, within each distance interval must meet the normalization constraint, the genetic algorithm makes a special design for parameter encoding: the mapping parameters are divided into three groups, intermediate variables α, β, γ are introduced for each group of parameters, and automatic normalization is achieved through the Softmax function.

[0019] S22: The fitness function is designed to minimize the difference between the model output and the actual observations. For action classification, the cross-entropy loss function is used as the optimization metric. For each data point i, its membership prediction vector is p. i =(p acc ,p keep ,p dec The true label corresponding to this data point is y. i The cross-entropy loss is calculated as follows:

[0020]

[0021] Among them, L i The cross-entropy loss represents data point i; The true label representing data point i; ε represents the membership value of state k; ε represents the smoothing term to prevent taking the logarithm for zero probability.

[0022] S23: Calculate the total loss of all N data points as the fitness value. The calculation formula is:

[0023] Furthermore, step S3 specifically includes: using a "virtual rotation" method to map the spatial distribution of vehicles on the main road and ramps to a unified reference frame, constructing a virtual axis to achieve a normalized representation of multi-dimensional traffic elements; defining the main road vehicle set I. M ={1 M ,2 M ,…,N M} and ramp vehicle collection I R ={1 R ,2 R ,…,N R}, where N M and N R Let X represent the total number of vehicles on the main road and the ramps, respectively; each set contains a set of vehicle positions and a set of vehicle speeds, denoted as X. M ={x1,x2,…,x m}, X R ={x1,x2,…,x r}, V M ={v1,v2,…,v m} and V R ={v1,v2,…,v r}, where x m The position of the m-th car on the main road, x r Let v be the position of the r-th vehicle on the ramp. m The speed of the m-th car on the main road, v r Let r be the speed of the r-th vehicle on the ramp;

[0024] The core of the "virtual rotation" method lies in simplifying the complex merging problem into a car-following problem through geometric transformation. Its mathematical implementation can be expressed as follows: by rotating the entrance ramp coordinate system (y2 axis) to the principal coordinate system (y1 axis), and constructing a virtual axis Z as a unified reference system, the spatial distribution of multi-source traffic flows is mapped to a single-dimensional space while maintaining the physical distance between vehicles and the merging point. Based on this, the union of vehicle index, relative position, and speed is set on the virtual axis Z, as shown below:

[0025] X Z =X M ∪X R

[0026] V Z =VM ∪V R

[0027] I Z =I M ∪I R

[0028] Among them, X Z V represents the set of relative positions of vehicles on the virtual axis Z; Z Represents the set of vehicle velocities along the virtual axis Z; I Z Represents the set of vehicle indices on the virtual axis Z;

[0029] Based on the index of the autonomous vehicle and its position on the Z-axis, the X-axis is adjusted using the function g. Z To quickly obtain the virtual train-following sequence on the Z-axis, perform monotonically descending sorting:

[0030] (X' Z ,V' Z ,I' Z ,F Z ,T Z )=g(X Z V Z ,I Z )

[0031] Where X' Z V' represents the set of relative positions on the Z-axis after sorting; Z Represents the set of velocities on the Z-axis after sorting; I' Z I' represents the set of train following order on the Z-axis after sorting. Z ={1,2,…,N total Sort in ascending order, N total =M+R;F Z This represents the main road or ramp markers on the Z-axis after sorting. Specifically, for i∈I Z If the i-th vehicle is actually on the main road, then f Z,i =1, otherwise f Z,i =0;T Z This represents the vehicle type label on the Z-axis after sorting. When the i-th vehicle is an autonomous vehicle, t Z,i =1, otherwise t Z,i =0;

[0032] After transforming the main road and ramps to virtual axes through "virtual rotation," the highway entrance ramp merging problem is reduced to a car-following driving problem; a longitudinal control strategy is adopted, consisting of N... i The feedback term consisting of the distance and speed of the vehicle in front and N i-kThe feedforward term, composed of the accelerations of the autonomous vehicle, is used as the control input, and its expression is:

[0033]

[0034] Among them, U i (t) represents the control input of the autonomous vehicle; ω e,i ω represents the feedback gain that deviates from the equilibrium spacing. v,i The feedback gain representing the deviation from velocity balance; e i (t) represents the deviation in the distance between the front ends of the vehicles; Δv i,e (t) represents the deviation of the equilibrium speed; α represents the acceleration of the i-th vehicle ahead; f,k Weighting coefficients representing acceleration feedforward information.

[0035] Furthermore, step S4 specifically includes the following steps:

[0036] S41: Based on the analysis of uncertainties in human driving, the nonlinear mapping relationship between multi-dimensional uncertainties and control decisions is parametrically modeled. In the universe of discourse U, each driving scenario is uniquely identified by a triple, forming an object set, where the triple includes distance, merging point distance, and vehicle type. The parameter set E is extended to conform to the parameter cluster E = E1 × E2 × E3. For example, for the parameter of the driver's judgment on following distance, E1 = {safe, normal, dangerous} represents the distance assessment parameter. The soft set (F, E) establishes the membership relationship between the combination of conditional attributes and the control input decision set D = {D1, D2, D3} by defining the mapping F: E → P(D), where P(D) is the power set of D, allowing the superposition of decision membership degrees under the combined action of single or multiple parameters. In the control input decision set D = {D1, D2, D3}, D1, D2, and D3 represent the additional control inputs that the autonomous vehicle will take to adjust its speed to cooperate with the human driver in merging the flow. The parameters in the control input decision set will be calibrated in the simulation experiment section.

[0037] S42: By incorporating the uncertainties in human driving behavior into the longitudinal driving strategy of autonomous vehicles, and integrating the autonomous vehicle control input obtained from the soft set feature information table into the formula, the improved autonomous vehicle control input F(t) can be expressed as:

[0038]

[0039] Among them, U i (t) represents the original control input, D iλ1 represents the membership degree of the control input obtained by the autonomous vehicle based on the driving characteristics information of the human driver; λ2 represents the weight coefficient of the original control input; and λ3 represents the weight coefficient of the control input of uncertain factors.

[0040] Furthermore, step S5 specifically includes the following steps:

[0041] S51: Build a scenario of a merging area at a highway entrance ramp using SUMO traffic simulation software, and conduct simulation experiments on the rightmost lane of the main road and the ramp in the merging area; take the area 1000m before the end of the merging area as the control area, and the autonomous vehicle in the control area will be controlled longitudinally by the merging control strategy proposed in step 4 that takes into account the uncertainty of human driving.

[0042] S52: Add several human-driven and autonomous vehicles to conduct micro-vehicle simulation experiments. Adjust the model parameters based on the experimental results to optimize the vehicle control effect.

[0043] S53: Define different levels of road traffic demand and different demand ratios for main roads and ramps, conduct macroscopic traffic flow simulation experiments, and further adjust model parameters based on experimental results to improve road traffic efficiency and safety.

[0044] The beneficial effects of this invention are as follows:

[0045] (1) Technological innovation in quantifying driving behavior uncertainty: Based on soft set theory, a trapezoidal membership function model is constructed, and for the first time, the triplet feature parameters of "spacing-merging point distance-vehicle type" are integrated to achieve a fuzzy representation of the complex decision-making logic of human drivers. Through NGSIM data-driven and genetic algorithm constraint optimization, the calibration error of the model parameters is reduced, and the accuracy of driving behavior prediction is significantly improved.

[0046] (2) Breakthrough in Dimensional Reconstruction of Merging Control: A "virtual rotation" coordinate system transformation method is proposed, which reduces the two-dimensional merging problem to a one-dimensional car-following control problem through geometric mapping. The virtual axis normalization representation technology is adopted to reduce the complexity of the control model and improve the real-time computing efficiency, laying the foundation for realizing large-scale vehicle cooperative control.

[0047] (3) Collaborative control mechanism for mixed traffic flow: The established vehicle-following control model is introduced into the IDM model to more realistically depict human driving behavior in simulation and optimize the longitudinal control strategy for autonomous vehicles. The proposed control strategy can effectively improve the road capacity of the merging zone, and improve the average traffic flow and average speed under the conditions of low autonomous vehicle penetration and high traffic demand.

[0048] (4) Data-model dual-driven optimization system: construct a genetic algorithm-Softmax joint calibration method to transform constrained optimization into an unconstrained problem.

[0049] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0050] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0051] Figure 1 This is a flowchart illustrating the overall process of the highway entrance ramp merging control method that takes into account the uncertainties of human driving in this invention.

[0052] Figure 2 This is a schematic diagram of a highway entrance ramp merging scenario built using SUMO simulation software in the embodiment.

[0053] Figure 3 This is a framework diagram of the vertical control strategy. Detailed Implementation

[0054] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0055] Please see Figures 1-3 This invention provides a merging control method for highway entrance ramps that considers the uncertainties of human driving, specifically including the following steps:

[0056] Step 1: Use soft set theory to quantitatively represent and model the uncertainties in human driving characteristics. This includes the following sub-steps:

[0057] Step 1.1: Analyze the uncertainties in human driving characteristics, such as the driver's judgment of the distance to the vehicle in front. Set a threshold to divide the distance between vehicles into several intervals, and obtain the membership function of the driver's judgment result. For example, the membership function of the driver's judgment that the following distance is safe can be described as:

[0058]

[0059] In the formula, u safe A represents the membership function that the driver determines is at a safe following distance. i1 d represents the membership value corresponding to each vehicle spacing interval; d represents the current following distance; d1 represents the distance threshold between the dangerous distance interval and the medium distance interval; d2 represents the distance threshold between the medium distance interval and the safe distance interval.

[0060] Step 1.2: Construct a parameterized soft set (F, E), where F:E→P(U) is the mapping function that maps each parameter e∈E to the power set of the membership degree of the corresponding interval. The mathematical representation of the soft set is achieved through a decision matrix, where the row vectors correspond to the intervals of the actual parameters, and the column vectors correspond to the evaluation parameters, i.e., the linguistic values. For example, the decision matrix corresponding to the uncertainty of a driver's judgment of the distance to the vehicle in front is:

[0061]

[0062] In the formula, A ij This represents the membership value of the actual vehicle spacing interval to the evaluation parameter.

[0063] Step 2: Based on the NGSIM dataset, obtain vehicle trajectory data from the merging zone of highway entrance ramps, and use this data to calibrate the parameters in the soft set model. This includes the following sub-steps:

[0064] Step 2.1: Use relevant data from the merging areas of highway entrance ramps in the NGSIM dataset and employ a genetic algorithm to calibrate the data. Taking the distance membership function as an example, the collected distance data are divided into three categories: distance less than d1 = 50m; distance between d1 = 50m and d2 = 80m; and distance greater than d2 = 80m.

[0065] Step 2.2: Considering that the membership degrees of acceleration, hold, and deceleration actions within each distance interval must satisfy the normalization constraint, the algorithm employs a special design for parameter encoding: the mapping parameters are divided into three groups, and intermediate variables α, β, and γ are introduced for each group of parameters, with automatic normalization achieved through the Softmax function. For example, parameter A for intervals with distances less than 50m... 11 A 21 A 31 It can be represented as:

[0066]

[0067]

[0068]

[0069] In the formula, α represents the intermediate variable when the evaluation parameter is safe; β represents the intermediate variable when the evaluation parameter is normal; and γ represents the intermediate variable when the evaluation parameter is dangerous.

[0070] The parameters for the remaining intervals are handled in the same way. This method transforms the constrained optimization problem into an unconstrained optimization problem, significantly simplifying the algorithm's implementation complexity.

[0071] Step 2.3: The fitness function is designed to minimize the difference between the model output and the actual observations. For action classification, the cross-entropy loss function is used as the optimization metric. For each data point i, its membership prediction vector is p. i =(p acc ,p keep ,p dec The true label corresponding to this data point is y. i The cross-entropy loss is calculated as follows:

[0072]

[0073] In the formula, L i The cross-entropy loss represents data point i; The true label representing data point i; ε represents the membership value of state k; ε represents the smoothing term to prevent taking the logarithm for zero probability.

[0074] Step 2.4: Calculate the total loss of all N data points as the fitness value:

[0075]

[0076] Step 3: Employ the "virtual rotation" method to transform the merging control problem into a following control problem, and establish a following control model. This includes the following sub-steps:

[0077] Step 3.1: Using the "virtual rotation" theory, the spatial distribution of vehicles on the main road and ramps is mapped to a unified reference frame, constructing a virtual axis to achieve a normalized representation of multi-dimensional traffic elements. Define the main road vehicle set I. M ={1 M ,2 M ,…,N M} and ramp vehicle collection I R ={1 R ,2R ,…,N R}, where N M and N R Let X be the total number of vehicles on the main road and the ramps, respectively. Each set contains a set of vehicle positions and a set of vehicle speeds, denoted as X. M ={x1,x2,…,x m}, X R ={x1,x2,…,x r}, V M ={v1,v2,…,v m} and V R ={v1,v2,…,v r}

[0078] Step 3.2: By rotating and transforming the entrance ramp coordinate system y2 axis to the principal coordinate system y1 axis, and constructing a virtual axis Z as a unified reference system, the spatial distribution of multi-source traffic flow is mapped to a single-dimensional space while keeping the physical distance between vehicles and merging points unchanged. Based on this, we can set the union of vehicle index, relative position, and speed on the virtual axis Z, as shown below:

[0079] X Z =X M ∪X R

[0080] V Z =V M ∪V R

[0081] I Z =I M ∪I R

[0082] In the formula, X Z V represents the set of relative positions of vehicles on the virtual axis Z; Z Represents the set of vehicle velocities along the virtual axis Z; I Z This represents the set of vehicle indices on the virtual axis Z.

[0083] Step 3.3: Based on the index of the autonomous vehicle and its position on the Z-axis, we can use the function g to adjust the X... Z To quickly obtain the virtual train-following sequence on the Z-axis, perform monotonically descending sorting:

[0084] (X' Z ,V' Z ,I' Z ,F Z ,T Z )=g(X Z V Z ,I Z )

[0085] In the formula, X' Z V' represents the set of relative positions on the Z-axis after sorting; Z Represents the set of velocities on the Z-axis after sorting; I' Z F represents the set of train following order on the Z-axis after sorting; Z Represents the main road / ramp signs on the Z-axis after sorting; T Z This represents the vehicle type label on the Z-axis after sorting. I' Z ={1,2,…,N total Sort in ascending order, N total =M+R. Specifically, for i∈I Z If the i-th vehicle is actually on the main road, then f Z,i =1, otherwise f Z,i =0. When the i-th vehicle is an autonomous vehicle, t Z,i =1, otherwise t Z,i =0.

[0086] Step 3.4: After transforming the main road and ramps to virtual axes through "virtual rotation," the highway entrance ramp merging problem is reduced to a car-following driving problem. This invention employs a longitudinal control strategy, consisting of N... i The feedback term consisting of the distance and speed of the vehicle in front and N i-k The feedforward term, composed of the accelerations of the autonomous vehicle, is used as the control input, and its expression is:

[0087]

[0088] In the formula, U i (t) represents the control input of the autonomous vehicle; ω e,i ω represents the feedback gain that deviates from the equilibrium spacing. v,i The feedback gain representing the deviation from velocity balance; e i (t) represents the deviation in the distance between the front ends of the vehicles; Δv i,e (t) represents the deviation of the equilibrium speed; α represents the acceleration of the i-th vehicle ahead; f,k Weighting coefficients representing acceleration feedforward information.

[0089] Step 4: Based on the soft set model, the uncertainty of driver-driven operations is incorporated into the control model, and a merging control strategy considering this uncertainty is proposed. This includes the following sub-steps:

[0090] Step 4.1: Based on the analysis of uncertainties in human-driven driving in Step 1, the nonlinear mapping relationship between multi-dimensional uncertainties and control decisions was parametrically modeled. In the universe of discourse U, each driving scenario is uniquely identified by a triple (distance, merging point distance, vehicle type), forming an object set. The parameter set E is extended to conform to the parameter cluster E = E1 × E2 × E3. For example, for the parameter of the driver's judgment on following distance, E1 = {safe, normal, dangerous} represents the distance assessment parameter. The soft set (F, E) establishes the membership relationship between the combination of conditional attributes and the control input decision set D = {D1, D2, D3} by defining the mapping F: E → P(D), where P(D) is the power set of D, allowing the superposition of decision membership degrees under the combined action of single or multiple parameters. In the control input decision set D = {D1, D2, D3}, D1, D2, and D3 respectively represent the additional control inputs that the autonomous vehicle will take to adjust its speed to cooperate with the human driver in completing the merging. The parameters in the control input decision set will be calibrated in the simulation experiment section.

[0091] Step 4.2: Incorporate the uncertainties of human driving behavior into the longitudinal driving strategy of the autonomous vehicle. Integrate the autonomous vehicle control input obtained from the soft set feature information table into the formula. The improved autonomous vehicle control input can be expressed as:

[0092]

[0093] In the formula, U i (t) represents the original control input; D i λ1 represents the membership degree of the control input obtained by the autonomous vehicle based on the driving characteristics information of the human driver; λ2 represents the weight coefficient of the original control input; and λ3 represents the weight coefficient of the control input of uncertain factors.

[0094] Step 5: Based on the proposed merging control strategy, a merging scenario at a highway entrance ramp is constructed using SUMO simulation software. Control is applied to the autonomous vehicles in the scenario, and model parameters are adjusted. This includes the following sub-steps:

[0095] Step 5.1: As Figure 2 As shown, a scenario of a merging area at a highway entrance ramp is constructed using SUMO traffic simulation software. Simulation experiments are conducted on the rightmost lane of the main road and the ramp within the merging area. The area 1000m before the end of the merging area is designated as the control zone. Within the control zone, autonomous vehicles will be controlled longitudinally by the merging control strategy proposed in step 4, which considers the uncertainties of human driving.

[0096] Step 5.2: Add several human-driven and autonomous vehicles to conduct microscopic vehicle simulation experiments. Adjust the model parameters based on the experimental results to optimize the vehicle control effect.

[0097] Step 5.3: Define different levels of road traffic demand and different demand ratios for main roads and ramps, conduct macroscopic traffic flow simulation experiments, and further adjust the model parameters based on the experimental results to improve road traffic efficiency and safety.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for merging control of highway entrance ramps considering the uncertainties of driver and vehicle operation, characterized in that, The method specifically includes the following steps: S1: Construct a soft set model to quantitatively characterize the uncertainties in human driving characteristics; S2: Obtain vehicle trajectory data of the merging zone of the highway entrance ramp based on the NGSIM dataset, and calibrate the parameters in the soft set model using the vehicle trajectory data; S3: The merging control problem is transformed into a following control problem using the "virtual rotation" method, and a following control model is established. Specifically, this includes: mapping the spatial distribution of vehicles on the main road and ramps to a unified reference frame using the "virtual rotation" method, constructing a virtual axis to achieve a normalized representation of multi-dimensional traffic elements; and defining the set of vehicles on the main road. Gathering with ramp vehicles ,in and Let represent the total number of vehicles on the main road and the ramps, respectively; each set contains a set of vehicle positions and a set of vehicle speeds, denoted as . , , and ,in, First on the main road m The location of the car The first one on the ramp r The location of the car First on the main road m The speed of the car The first one on the ramp r The speed of the vehicle; The core of the "virtual rotation" method lies in simplifying the complex merging problem into a car-following problem through geometric transformation. Its mathematical implementation is expressed as follows: by changing the coordinate system of the entrance ramp... Axis rotation transformation to principal coordinate system Axis, and construct virtual axes As a unified reference frame, while keeping the physical distance between vehicles and merging points constant, the spatial distribution of multi-source traffic flows is mapped to a single-dimensional space; based on this, on the virtual axis The above sets the union of vehicle index, relative position, and speed, as shown below: in, Represents virtual axis The set of relative positions of vehicles on the road; Represents virtual axis The set of vehicle speeds on the road; Represents virtual axis The set of vehicle indexes on the website; Based on the index of autonomous vehicles and in Position on the axis, via a function right Perform monotonically descending sort to quickly obtain Virtual train following sequence on the axle: in, After sorting The set of relative positions on the axis; After sorting Velocity set on the axis; After sorting The set of train following order on the axle Sort in ascending order ; After sorting Main road or ramp signs on the axle Specifically, for If the first If the vehicle is actually on the main road, then Conversely ; After sorting Vehicle type markings on the axle When the first When the vehicle is an autonomous vehicle Conversely ; After transforming the main road and ramps to virtual axes through "virtual rotation," the highway entrance ramp merging problem is reduced to a car-following driving problem; a longitudinal control strategy is then adopted, from The feedback term consisting of the distance and speed of the vehicle in front and The feedforward term, composed of the accelerations of the autonomous vehicle, is used as the control input, and its expression is: in, Represents the control inputs for autonomous vehicles; The feedback gain represents the deviation from the equilibrium spacing; The feedback gain represents the deviation from the speed balance; This represents the deviation in the distance between the front of the vehicle; The deviation representing the equilibrium speed; Representing the front number The acceleration of a vehicle; Weighting coefficients representing acceleration feedforward information; S4: Based on the soft set model, the uncertainty of human driving is introduced into the control model, and a merging control strategy that takes into account the uncertainty of human driving is proposed. S5: Based on the proposed merging control strategy, a merging scenario at the entrance ramp of a highway is built using SUMO traffic simulation software. Control is applied to the autonomous vehicles in the scenario and the model parameters are adjusted.

2. The merging control method for highway entrance ramps according to claim 1, characterized in that, Step S1 specifically includes: selecting the driver's judgment on the distance to the vehicle in front, the vehicle type, and the distance to the end of the merging zone as key feature parameters, and constructing a soft set model; each parameter is fuzzified using a trapezoidal membership function.

3. The merging control method for highway entrance ramps according to claim 2, characterized in that, In step S1, a soft set model is constructed, specifically a parameterized soft set model. ,in For the mapping function, each parameter The set is mapped to the power set of the membership degree of the corresponding interval; the mathematical representation of the soft set is realized through the decision matrix, where the row vectors correspond to the intervals of the actual parameters and the column vectors correspond to the evaluation parameters, i.e., the language values.

4. The merging control method for highway entrance ramps according to claim 1, characterized in that, Step S2 specifically includes: obtaining vehicle trajectory data of the merging area of ​​the highway entrance ramp based on the NGSIM dataset, and using a genetic algorithm to calibrate the parameters in the soft set model.

5. The merging control method for highway entrance ramps according to claim 4, characterized in that, Step S2 specifically includes the following steps: S21: Divide the collected distance data into three categories: distance less than the first threshold Distance at the first threshold With the second threshold Between, the distance is greater than the second threshold ,and ,in This represents the distance threshold between the dangerous distance range and the medium distance range. This represents the distance threshold between the medium distance interval and the safe distance interval. Considering that the membership degrees of acceleration, holding, and deceleration actions within each distance interval must satisfy normalization constraints, the genetic algorithm employs a special design for parameter encoding: dividing the mapping parameters into three groups and introducing intermediate variables for each group. Automatic normalization is achieved through the Softmax function. S22: The fitness function is designed to minimize the difference between the model output and the actual observations. For action classification, the cross-entropy loss function is used as the optimization metric. For each data point... Its membership prediction vector is The true label corresponding to this data point is The cross-entropy loss is calculated as follows: in, Representative data points Cross-entropy loss; Representative data points The true label; Representing state The membership degree value; This represents a smoothing term to prevent taking the logarithm of zero probability; S23: Calculate all The total loss for each data point is the fitness value, calculated as follows: .

6. The merging control method for highway entrance ramps according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41: Based on the analysis of uncertainties in human-driven driving, the nonlinear mapping relationship between multidimensional uncertainties and control decisions is parametrically modeled; in the universe of discourse... In this context, each driving scenario is uniquely identified by a triple, forming a set of objects. The triple includes the spacing, merging point distance, and vehicle type; the parameter set... Extended to conform to parameter family Regarding the parameter of the driver's judgment of following distance, Characterization interval evaluation parameters; soft set By defining mappings Establish conditional attribute combinations into the control input decision set The subordinate relationship, among which for The power set allows for the superposition of decision memberships under single-parameter or multi-parameter combined effects; in the control input decision set middle, , and These represent additional control inputs for the autonomous vehicle to accelerate, maintain, and decelerate to adjust its speed in coordination with the human driver's actions to complete the merging; the parameters in the control input decision set will be calibrated in the simulation experiment section. S42: The longitudinal driving strategy of autonomous vehicles incorporates considerations of uncertainties in human driving behavior. The control inputs of the autonomous vehicle obtained from the soft set feature information table are integrated into the formula, resulting in improved control inputs for the autonomous vehicle. Represented as: in, Represents the original control input, This represents the membership degree of the control input obtained by the autonomous vehicle based on the driving characteristics information of the human driver; Weighting coefficients representing the original control inputs; The weighting coefficient represents the input of uncertain factors.

7. The merging control method for highway entrance ramps according to claim 1, characterized in that, Step S5 specifically includes the following steps: S51: Build a scenario of a merging area at a highway entrance ramp using SUMO traffic simulation software, and conduct simulation experiments on the rightmost lane of the main road and the ramp in the merging area; take the area 1000m before the end of the merging area as the control area, and the autonomous vehicle in the control area will be controlled longitudinally by the merging control strategy proposed in step 4 that takes into account the uncertainty of human driving. S52: Add several human-driven and autonomous vehicles to conduct micro-vehicle simulation experiments. Adjust the model parameters based on the experimental results to optimize the vehicle control effect. S53: Define different levels of road traffic demand and different demand ratios for main roads and ramps, conduct macroscopic traffic flow simulation experiments, and further adjust model parameters based on experimental results to improve road traffic efficiency and safety.

Images