Intelligent networked automobile lane changing trajectory planning method considering personalized riding requirements of passengers

Abstract

The invention relates to the technical field of automatic driving, and discloses an intelligent networked automobile lane changing track planning method considering personalized riding requirements of passengers. The method mainly can be divided into three parts, namely a passenger personalized demand extraction layer, a decision-making layer and a planning layer. In the passenger personalized demand extraction layer, a passenger demand factor mapping model based on an analytic hierarchy process and a fuzzy logic system is introduced to be used as a subsequent personalized quantitative benchmark. In the decision-making layer, a lane changing decision-making model considering the personalized riding demand of the passenger is designed based on the artificial potential energy field theory and the pure strategy game method, and a personalized lane changing decision can be generated based on the driving environment information and the passenger demand factor. In the planning layer, a lane changing time length prediction model is designed and used for restraining the lane changing time length, the problem is converted into a quadratic programming problem, and therefore the personalized lane changing track meeting the requirements of passengers is obtained through solving.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a method for planning lane-changing trajectories of an intelligent networked vehicle that takes into account personalized riding needs of passengers. Background Art

[0002] At present, the rapid development of autonomous driving technology, especially the breakthrough in lane change trajectory planning technology, has become the core of intelligent connected vehicles integrating into complex traffic environments. In driving behavior, lane change trajectory planning is not only related to the safety of the vehicle itself, but also directly affects the overall traffic smoothness. Therefore, it is necessary to accurately analyze the operating status of surrounding vehicles to ensure that the lane change process is both fast and safe. Passenger needs often directly affect the driving direction of autonomous driving vehicles. Autonomous driving trajectory planning combined with passenger needs can lead the traffic system to a higher level of intelligence. At present, there are many methods for lane change trajectory planning for autonomous driving vehicles, such as optimization-based and learning-based methods. The optimization-based method essentially achieves the purpose of trajectory planning by abstracting one or more goals and designing corresponding constraints; the learning-based method generally models the trajectory planning problem as a Markov process, and realizes the training of the agent for different tasks by designing the interaction mode between the agent and the environment.

[0003] Although existing methods have achieved certain results in common lane-changing trajectory planning tasks, existing methods do not adequately study the characteristics of passengers' riding preferences, and few studies have incorporated passengers' riding preferences into the consideration of lane-changing trajectory planning for autonomous vehicles. In addition, there are also deficiencies in the current research that combines the study of passengers' driving preferences with lane-changing trajectory planning. On the one hand, existing methods rely on learning about the environment and require a large amount of passenger preference data; on the other hand, existing methods all consider lane-changing trajectories from the driver's perspective without considering the real needs of passengers. Therefore, how to conveniently and quickly characterize the specific needs of passengers has become a technical difficulty. In addition, how to deeply integrate the real needs of passengers into the entire lane-changing decision-making and trajectory planning process is also a current technical difficulty. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a lane-changing trajectory planning method for an intelligent connected vehicle that takes into account the personalized riding needs of passengers;

[0005] To implement the above technical solution, the details are as follows:

[0006] S1. Convert the passenger's riding preference into a judgment matrix, and map the judgment matrix into a passenger preference weight vector;

[0007] The passenger's riding preference includes: riding comfort preference, safety preference and efficiency preference;

[0008] The preference for ride comfort, safety and efficiency is formed into a judgment matrix A in the form of a 3×3 matrix; the element A in the i-th (i=1, 2, 3) row and the j-th (j=1, 2, 3) column is ij It represents the interaction between passengers’ perception of ride comfort, safety and efficiency;

[0009] Passenger preference weight vector P = [w c w s w e ], respectively representing the importance of passengers to the interaction of comfort, safety and efficiency; the solution method is hierarchical analysis method, and the specific steps include:

[0010] S1.1. Calculate the sum of each column of the judgment matrix; the expression is as follows:

[0011]

[0012] In the formula, S j represents the sum of the elements in the jth column of the judgment matrix A; A ij represents the element in the i-th row and j-th column of the judgment matrix A. It means to sum the elements in the jth column;

[0013] S1.2, standardize each element in the judgment matrix A; the expression is as follows:

[0014]

[0015] In the formula, A' ij Represents the standardized judgment matrix elements;

[0016] S1.3. Calculate each weight value in the passenger preference weight vector; the expression is as follows:

[0017]

[0018] Where P represents the passenger preference weight vector; P k represents the kth element in the passenger preference weight vector;

[0019] S1.4. Ensure the rationality of the judgment matrix by performing consistency check on the customer preference weight vector. The specific steps are as follows:

[0020] S1.4.1. Calculate the maximum eigenvalue; the expression is as follows:

[0021]

[0022] In the formula, λ max is the maximum eigenvalue of the judgment matrix A, (A·P)k represents the kth element of the product of the judgment matrix A and the passenger preference weight vector P, which represents the comprehensive score of each factor;

[0023] S1.4.2. Calculate the consistency index; the expression is as follows:

[0024]

[0025] In the formula, CI represents the consistency index value, and n represents the matrix dimension;

[0026] S2, establish a secondary mapping method to obtain the passenger preference factor f through the passenger preference weight vector;

[0027] The secondary mapping methods include: triangular membership function and Gaussian membership function;

[0028] The specific process is shown as follows;

[0029] S2.1. Define the triangular membership function:

[0030]

[0031] Where a represents the left endpoint of the triangle in the triangular membership function, b represents the right endpoint of the triangle in the triangular membership function, and c represents the peak point of the triangle in the triangular membership function. x represents the input of the triangular membership function, i.e., the weight values ​​in the passenger preference weight vector P. μ T (x) represents the triangular membership function with input x;

[0032] S2.2, define the Gaussian membership function to define fuzziness:

[0033]

[0034] In the formula, μ and σ represent the mean and standard deviation of Gaussian distribution respectively, y represents the output value of triangular membership function, μ G (y) represents the Gaussian membership function with input y;

[0035] S2.3. After inputting each weight value in the passenger preference weight vector P into the triangular membership function and then passing through the Gaussian membership function, the final passenger preference factor f can be obtained by summing up; the expression is as follows:

[0036]

[0037] Wherein, the passenger preference factor f∈[-1,1] directly reflects the type of passengers, including: radical, ordinary and conservative, corresponding to passenger preference factors f=-1, f=0 and f=1 respectively;

[0038] The above steps realize the self-supervised mapping mode of passenger preference weight factors, overcoming the technical difficulty of supervised learning such as deep learning that requires a large amount of calibration data;

[0039] S3. Through the three-layer personalized lane-changing decision model, a mapping relationship between the passenger's acceptable collision time and the passenger demand factor is established to generate the lane-changing trajectory planning decision of the vehicle;

[0040] The three-layer personalized lane-changing decision model includes: risk assessment layer, global benefit assessment layer and personalized assessment layer;

[0041] The establishment process is as follows:

[0042] S3.1. In the risk assessment layer, the artificial potential field theory is introduced to calculate the risk value of each lane. The calculation formula is as follows:

[0043]

[0044] In the formula, represents the potential energy field size of the oth lane; r represents the number of all obstacle vehicles on the oth lane; t e Indicates the duration of lane change; η obs is the lane weight; A μ is the longitudinal cross-sectional area of ​​the μth obstacle vehicle, which is used to characterize the size of the vehicle, μ∈(1,+∞); is the Euclidean distance between the ego vehicle and the μth obstacle vehicle at the tth moment; Q represents the influence range of the vehicle's potential energy field, that is, when the Euclidean distance between an obstacle vehicle and the ego vehicle exceeds Q, the potential energy of the vehicle in the current lane is not calculated; represents the relative speed between the ego vehicle and the μth obstacle vehicle at the tth moment. Due to the limitation of the 10 Hz planning frequency, the interval between t and t+1 is 0.1 s. In addition, o and r are both integers within the interval range. The above process is valid only if When the difference is greater than the threshold ξ, the ego vehicle has the intention to change lanes;

[0045] The calculation process of the gap threshold ξ is as follows:

[0046] Get the minimum braking distance for the vehicle to avoid collision with the obstacle vehicle; the expression is as follows:

[0047]

[0048] In the formula, a bmax represents the maximum braking deceleration of the vehicle, and s represents the minimum braking distance without collision between the vehicle and the obstacle vehicle. At this time, we can get:

[0049]

[0050] That is, the potential energy field size of the oth lane can be regarded as a function of the minimum braking distance s, where s∈(0,+∞). This function is a convex function in the domain of definition. Considering safety, the present invention calculates the value of the minimum braking distance s that can make the potential energy field of the oth lane Minimum, that is, solving the following equation:

[0051]

[0052] In the formula, represents the differential of the potential energy field of the oth lane with respect to the minimum braking distance s for the ego vehicle to avoid collision with the obstacle vehicle. The solution of this equation is regarded as the gap threshold ξ;

[0053] S3.2. In the global evaluation layer, a set of optimal strategies is generated between the ego vehicle and the vehicle behind it in the target lane through a pure strategy game method, which can minimize the impact on the overall traffic operation. The steps to obtain the optimal strategy are as follows:

[0054] S3.2.1. Generate a set of strategies with the highest benefits between the ego vehicle and the vehicle behind in the target lane by using a pure strategy game method, and make a lane change decision;

[0055]

[0056] Where U VC (t) and U VT (t) represents the benefit value of the current lane and the target lane of the ego vehicle at the tth moment, respectively; U(t) is the total benefit of the local traffic system at the tth moment; Γ represents the number of all obstacle vehicles in front of the current lane of the ego vehicle; p and q are the number of all obstacle vehicles in front of and behind the target lane from the perspective of the ego vehicle, respectively; ω t is the time weight value at the tth moment; V(t), and represents the vehicle, the γth obstacle vehicle in front of the current lane, the ρth obstacle vehicle in front of the target lane, and the ρth obstacle vehicle behind the target lane. The speed of the obstacle vehicle at the tth moment; where Γ∈[0,+∞), p∈[0,+∞), q∈[0,+∞), γ∈[0,+∞), ρ∈[0,+∞), The specific value depends on the perception limit of the vehicle, so the values ​​of Γ, p and q are a dynamic process;

[0057] All strategy combinations include: "the ego vehicle changes lanes and the vehicle behind the target lane accepts" and "the ego vehicle does not change lanes and the vehicle behind the target lane accepts". The global benefit value of each strategy combination is calculated based on the above global benefit function, and the evaluation is completed by finding the optimal benefit combination. It should be noted that the global benefit evaluation will continue with the subsequent evaluation only when the ego vehicle strategy is to change lanes, otherwise the decision of not changing lanes is directly output.

[0058] S3.3, if both layers have made lane change strategies, the passenger's needs are further evaluated by comparing the collision time between the vehicle and the vehicle behind in the target lane to determine whether the passenger's personalized needs are met;

[0059] Specifically, in the personalized evaluation layer, a mapping relationship between the minimum acceptable collision time of passengers and the passenger demand factor f is established, and the establishment process is as follows:

[0060] S3.3.1. Predicted collision duration; the expression for predicted collision duration is as follows:

[0061]

[0062] Where, T A Indicates the predicted collision duration; D TF is the Euclidean distance between the ego vehicle and the vehicle behind in the target lane; |vv TF | represents the absolute difference in speed between the ego vehicle and the vehicle behind in the target lane; v represents the speed of the ego vehicle; v TF is the speed of the vehicle behind in the target lane;

[0063] S3.3.2. Establish the mapping relationship between the passenger's acceptable collision time and the passenger demand factor f, expressed as follows:

[0064] Where, T P Indicates the acceptable collision time for passengers; T S is a time constant; η T Indicates the scaling weight, which is set to 0.3 based on experience;

[0065] S3.3.3, the final output is obtained by predicting the collision duration and mapping relationship;

[0066] Specifically, based on the above analysis, when T A >T P When , it indicates that the predicted collision time with the vehicle behind in the target lane meets the needs of the passenger, that is, the final output allows the decision to change lanes; otherwise, it indicates that the passenger rejects the decision to change lanes, and the decision to change lanes will not be executed;

[0067] S4. By establishing a trajectory generation model based on multidimensional coordinate space, the lane-changing trajectory planning decision problem of the ego vehicle is transformed into a multivariate linear equation system for solving the lane-changing trajectory; the expression is as follows:

[0068]

[0069] in, as well as They are respectively represented as the initial time t of the lane change of the vehicle s The projected distance on the centerline of the target lane, the initial speed, the initial acceleration, the initial plane abscissa, the initial plane ordinate, and the slope of the path point at the initial time; as well as They represent the projection distance of the vehicle on the center line of the target lane at the end of the lane change, the speed at the end, the acceleration at the end, the plane horizontal coordinate at the end, the plane vertical coordinate at the end, and the end time t e The slope of the path point; for the convenience of calculation, the above state quantities are combined into the initial state vector and the final state vector When the above two state vectors are determined, the trajectory polynomial coefficient vectors α and β can also be determined by solving a multivariate linear equation system. Since all state quantities in the initial state vector can be obtained by the vehicle-mounted sensors, it is necessary to further calculate the state quantities in the final state vector. As noted, the target lane centerline is actually a sequence of discrete points. It is only necessary to search for the point that matches the target lane centerline through the Euclidean distance. The nearest point, and through linear interpolation between its previous point and the next point, can solve all except the end time t e , the speed at the end And acceleration All state quantities of ;

[0070] S5. Optimizing and solving unknown quantities of the multivariate linear equations for the lane-changing trajectory;

[0071] That is, by setting the cost function and constraints that take into account the passenger preferences, the problem of solving the unknowns of the lane-changing trajectory multivariate linear equations is transformed into a nonlinear optimization problem. The specific steps are as follows:

[0072] S5.1. Establish a multi-objective cost function considering safety, comfort and efficiency based on passenger preferences, including the following steps:

[0073] S5.1.1, introduce lane-level potential energy field as a safety constraint, taking into account relative velocity, acceleration and longitudinal cross-sectional area of ​​the vehicle along the y direction;

[0074] The lane-level potential energy field calculation expression is as follows:

[0075]

[0076] In the formula, as well as They respectively represent the projection distance values ​​of the ego vehicle, the obstacle vehicle in the target lane behind the ego vehicle, the obstacle vehicle in the target lane in front of the ego vehicle, and the obstacle vehicle in the current lane in front of the ego vehicle on the lane change target lane at the tth moment; as well as They represent the speed of the ego vehicle and the obstacle vehicle in the target lane behind the ego vehicle at time t respectively; and is the maximum braking deceleration of the ego vehicle and the obstacle vehicle behind the ego vehicle in the target lane; τ is the planning time step of the ego vehicle; Q is the potential energy influence range of each vehicle;

[0077] S5.1.2. The safety cost function is expanded by a second-order Taylor expansion at each trajectory point to satisfy the quadratic form; the expression is as follows:

[0078]

[0079]

[0080] In the formula, Js represents the security cost function; X represents the form of high-dimensional unknowns; X0 is the input of each trajectory point; is the gradient calculation; H represents the Hessian matrix; represents partial differential calculation;

[0081] S5.1.3. In the comfort and efficiency cost functions, the acceleration and jerk (acceleration change rate) of each trajectory point of the vehicle are introduced to characterize the cost function; the expression is as follows:

[0082]

[0083] In the formula, J c and J e They represent the comfort cost function and the efficiency cost function respectively; j(t) represents the jerk of the vehicle at time t; t e Indicates the total lane change time;

[0084] S5.1.4. Combining safety, comfort and efficiency, a multi-objective cost function is obtained; the expression is as follows:

[0085] J=w s ·J s +w c ·J c +w e ·J e

[0086] In the formula, w s 、w c and w e are the weight values ​​of the three indicators in the passenger preference weight vector P;

[0087] S5.2. Optimize the hyperparameters of the long short-term memory neural network through the whale optimization algorithm to obtain the upper and lower limits of the safety constraint interval;

[0088] The specific steps are as follows:

[0089] S5.2.1. Establish a long short-term memory neural network based on the whale optimization algorithm to obtain the estimated driving distance; the expression is as follows:

[0090]

[0091] In the formula, is the speed of the vehicle at the initial moment of lane change, is the estimated driving distance, and Respectively represent the speed of the ego vehicle and the obstacle vehicle in the target lane in front of the ego vehicle;

[0092] S5.2.2. Collect data sets from open source data sets, randomly select 70% of them as training sets and the remaining 30% as test sets; and use the whale optimization algorithm to optimize the hyperparameters of the long short-term memory neural network, including: learning rate and maximum number of iterations;

[0093] By calculating the loss value through the loss function, we can determine the value of the hyperparameter that the whale optimization algorithm needs to search for to minimize the loss value, and then update the model weights in reverse:

[0094]

[0095] In the formula, Λ represents the total number of training samples; T φ and They represent the actual lane changing time and the lane changing time predicted by the model respectively; F represents the predicted loss value of each iteration, that is, the fitness function value of the whale optimization algorithm;

[0096] S5.2.3. Introduce the passenger preference factor f and design the following lane change time constraint form:

[0097] t e ∈[T min ,T max ]

[0098]

[0099] Where, T min represents the lower bound of the lane change time, T maxrepresents the upper bound of the lane change duration, Represents the predicted lane change duration output by the model;

[0100] S5.2.4. Introduce collision constraints during vehicle driving to ensure that the vehicle does not collide;

[0101] Specifically, it is necessary to ensure that the vehicle does not collide during driving, so it is necessary to introduce collision constraints into the optimization problem; in order to improve the calculation efficiency of the model, the collision constraints of the present invention can be expressed in a linearized form;

[0102] The distance projected on the target lane by the midpoint of the rear axle of the vehicle at the initial moment of lane change is Then, three vertices, namely the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper, can be calculated through geometric calculation as follows:

[0103]

[0104] In the formula, w ego is the width of the vehicle; θ t is the heading angle of the vehicle at time t, θ t ∈(0,2π);lr ego Indicates the length from the midpoint of the vehicle's rear axle to the vehicle's rear bumper; and They represent the projection distance values ​​of the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper on the center line of the target lane at time t, respectively. and The value range of is [0,+∞); and Combined into a vector S Ω , then obviously S Ω is the projection value of the vehicle's distance on the center line of the target lane χ∈{A,B,C} is a linear function; in this case, the linearization constraint of the ego-vehicle safety can be expressed as:

[0105]

[0106] In the formula, f colli and r colli They represent the safety distance threshold in front of the vehicle and the safety distance threshold behind the vehicle respectively; s 1 ,t 、s 2,t and 3,t They respectively represent the distance projection value of the obstacle vehicle behind the target lane on the target lane, the distance projection value of the obstacle vehicle in front of the target lane on the target lane, and the distance projection value of the obstacle vehicle in front of the lane where the vehicle is currently located on the target lane;

[0107] S5.2.5. The planned trajectory needs to be subject to motion state constraints at the end of the lane change;

[0108] Specifically, in order to ensure that the ego vehicle can smoothly follow the traffic flow of the target lane at the end of the lane change, it is necessary to impose motion state constraints on the end point of the lane change trajectory planned by the ego vehicle;

[0109] The present invention introduces a vehicle following model to calculate the motion state of the lane change endpoint, and derives the acceleration constraint at the lane change endpoint as:

[0110]

[0111] In the formula, Represents the vector composed of the accelerations of the obstacle vehicles behind and in front of the vehicle on the target lane; Δs is the speed of the obstacle vehicle in front of the vehicle in the current lane at the end of the lane change; * is a distance constant with a value range of Δs * ∈[1,5]; ω1 and ω2 are the vehicle distance weight value and relative speed weight value respectively. The vehicle distance weight value ranges from ω1∈[0,1]; the relative speed weight value ranges from ω2∈[0,1]; l OV-3 and ego They represent the length of the obstacle vehicle in front of the current lane of the ego vehicle and the length of the ego vehicle respectively; therefore, the motion state constraints can be summarized as:

[0112]

[0113] In the formula, represents the projection distance of the vehicle on the center line of the target lane at time t; represents the speed of the vehicle at time t;

[0114] S5.3. The trajectory optimization problem described in the above steps S5.1 and S5.2 is organized into a nonlinear programming problem. The complete expression is as follows:

[0115]

[0116] Substitute the expression of the above nonlinear programming problem into the IPOPT sequential quadratic programming solver to obtain the personalized planning trajectory;

[0117] Beneficial effects of the present invention

[0118] The present invention can be mainly divided into three parts, namely the passenger personalized demand extraction layer, the decision layer and the planning layer. In the passenger personalized demand extraction layer, the present invention introduces a passenger demand factor mapping model based on the hierarchical analysis method and the fuzzy logic system, which can map the language information on the passenger side into a passenger demand weight vector and a passenger demand factor for subsequent personalized quantitative benchmarks. In the decision layer, the present invention designs a lane change decision model that takes into account the personalized riding needs of passengers based on the artificial potential field theory and the pure strategy game method, which can generate personalized lane change decisions based on driving environment information and passenger demand factors. In the planning layer, the present invention constructs a weighted cost function through the passenger demand weight vector and transforms the lane change trajectory planning problem into a nonlinear optimization problem. In order to further improve the computational efficiency, the present invention designs a constrained linearization method for safety constraints, and introduces a long short-term memory neural network model optimized by an optimization algorithm for predicting the lane change duration of the vehicle. The prediction result will be used as a priori condition for trajectory planning. Finally, based on the quadratic planning, the lane change trajectory planning problem is solved to plan a personalized lane change trajectory that meets the passenger needs, thereby improving the problem that the personalized needs of passengers of autonomous driving vehicles are difficult to meet and optimizing the operation of the overall traffic flow.

[0119] Compared with the prior art, firstly, the present invention does not rely on calibration data that requires high costs, and can realize the mapping logic of passengers' fuzzy demands to preference weight values ​​in a fixed range, and can directly characterize passengers' specific demands conveniently and quickly, and the computational complexity is superior to that of the prior art; secondly, the decision layer of the present invention greatly reduces the computational complexity compared with the existing decision-making methods based on learning or optimization, and has great potential for online application; in addition, compared with the prior art, the present invention also considers the impact of the lane changing behavior of the vehicle on the global traffic operation state, and optimizes the operation of the overall traffic flow to a certain extent; furthermore, the present invention models the personalized lane changing trajectory planning problem of passengers as a sequential quadratic programming problem, and designs a linearization method of complex constraints, which can have higher computational efficiency than various existing optimization-based trajectory planning methods; finally, the present invention uses the passenger's real needs through the passenger preference weight vector and the passenger preference factor to act on the lane changing decision and planning process of the vehicle, which can reflect the passenger's real needs more deeply and systematically than the existing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 It is a flow chart of the present invention;

[0121] Figure 2 It is a schematic diagram of the present invention;

[0122] Figure 3 This is a schematic diagram of the passenger preference extraction model of the present invention;

[0123] Figure 4 It is the process of mapping the passenger preference weight vector to the passenger preference factor of the present invention;

[0124] Figure 5 It is a schematic diagram of the potential energy field range of the present invention;

[0125] Figure 6 This is a schematic diagram of the safety constraint linearization of the present invention;

[0126] Figure 7 This is a schematic diagram of the lane change duration prediction effect of the present invention;

[0127] Figure 8 Schematic diagram of the lane changing trajectory planning results of the present invention in different test scenarios, wherein part (a) is a schematic diagram of the motion planning curves of three different passengers; part (b) is a schematic diagram of the lane changing paths of different speeds and accelerations corresponding to three different passengers. DETAILED DESCRIPTION

[0128] The present invention will be further described in detail below in conjunction with specific embodiments.

[0129] like Figure 1 , Figure 2 As shown, a lane-changing trajectory planning method for an intelligent connected vehicle considering the personalized riding needs of passengers includes the following steps:

[0130] S1. Convert the passenger's riding preference into a judgment matrix, and map the judgment matrix into a passenger preference weight vector;

[0131] like Figure 3 As shown, the passenger's riding preference is extracted through the passenger preference extraction model;

[0132] The passenger's riding preference includes: riding comfort preference, safety preference and efficiency preference;

[0133] The preference for ride comfort, safety and efficiency is formed into a judgment matrix A in the form of a 3×3 matrix; the element A in the i-th (i=1, 2, 3) row and the j-th (j=1, 2, 3) column is ij It represents the interaction between passengers’ perception of ride comfort, safety and efficiency;

[0134] Passenger preference weight vector P = [w c w s w e ], respectively representing the importance of passengers to the interaction of comfort, safety and efficiency; the solution method is hierarchical analysis method, and the specific steps include:

[0135] S1.1. Calculate the sum of each column of the judgment matrix; the expression is as follows:

[0136]

[0137] In the formula, S j represents the sum of the elements in the jth column of the judgment matrix A; A ij represents the element in the i-th row and j-th column of the judgment matrix A. It means to sum the elements in the jth column;

[0138] S1.2, standardize each element in the judgment matrix A; the expression is as follows:

[0139]

[0140] In the formula, A' ij Represents the standardized judgment matrix elements;

[0141] S1.3. Calculate each weight value in the passenger preference weight vector; the expression is as follows:

[0142]

[0143] Where P represents the passenger preference weight vector; P k represents the kth element in the passenger preference weight vector;

[0144] S1.4. Ensure the rationality of the judgment matrix by performing consistency check on the customer preference weight vector. The specific steps are as follows:

[0145] S1.4.1. Calculate the maximum eigenvalue; the expression is as follows:

[0146]

[0147] In the formula, λ max is the maximum eigenvalue of the judgment matrix A, (A·P) k represents the kth element of the product of the judgment matrix A and the passenger preference weight vector P, which represents the comprehensive score of each factor;

[0148] S1.4.2. Calculate the consistency index; the expression is as follows:

[0149]

[0150] In the formula, CI represents the consistency index value, and n represents the matrix dimension;

[0151] If the consistency ratio CR<0.1, the passenger preference weight vector calculated from step S1.1 to step S1.3 is considered acceptable. Otherwise, it indicates that the judgment matrix A has a high consistency, that is, there is a large conflict in the information input by the passenger. This is due to input errors. Therefore, the obtained passenger preference weight vector is not reasonable. At this time, the passenger needs to re-enter his riding preference and repeat steps S1.1 to step S1.6 until the consistency ratio CR<0.1;

[0152] S2, establish a secondary mapping method to obtain the passenger preference factor f through the passenger preference weight vector;

[0153] The process of mapping the passenger preference weight vector to the passenger preference factor is as follows: Figure 4 As shown; the details are as follows:

[0154] The secondary mapping method includes: triangular membership function and Gaussian membership function; that is, by establishing a fuzzy logic system coupled by triangular membership function and Gaussian membership function, the passenger preference weight vector P can be passed through the fuzzy logic system to obtain the passenger preference factor f, and the passenger preference weight vector with one-dimensional length can be further reduced to a numerical value. The specific process is shown as follows;

[0155] S2.1. Define the triangular membership function:

[0156]

[0157] Where a represents the left endpoint of the triangle in the triangular membership function, b represents the right endpoint of the triangle in the triangular membership function, and c represents the peak point of the triangle in the triangular membership function. x represents the input of the triangular membership function, i.e., the weight values ​​in the passenger preference weight vector P. μ T (x) represents the triangular membership function with input x;

[0158] S2.2, define the Gaussian membership function to define fuzziness:

[0159]

[0160] In the formula, μ and σ represent the mean and standard deviation of Gaussian distribution respectively, y represents the output value of triangular membership function, μ G (y) represents the Gaussian membership function with input y;

[0161] S2.3. After inputting each weight value in the passenger preference weight vector P into the triangular membership function and then passing through the Gaussian membership function, the final passenger preference factor f can be obtained by summing up; the expression is as follows:

[0162]

[0163] Wherein, the passenger preference factor f∈[-1,1] directly reflects the type of passengers, including: radical, ordinary and conservative, corresponding to passenger preference factors f=-1, f=0 and f=1 respectively;

[0164] The above steps realize the self-supervised mapping mode of passenger preference weight factors, overcoming the technical difficulty of supervised learning such as deep learning that requires a large amount of calibration data;

[0165] S3. Through the three-layer personalized lane-changing decision model, a mapping relationship between the passenger's acceptable collision time and the passenger demand factor is established to generate the lane-changing trajectory planning decision of the vehicle;

[0166] The three-layer personalized lane-changing decision model includes: risk assessment layer, global benefit assessment layer and personalized assessment layer;

[0167] The establishment process is as follows:

[0168] S3.1. In the risk assessment layer, the artificial potential field theory is introduced to calculate the risk value of each lane. The calculation formula is as follows:

[0169]

[0170] In the formula, represents the potential energy field size of the oth lane; r represents the number of all obstacle vehicles on the oth lane; t e Indicates the duration of lane change; η obs is the lane weight, and in this embodiment, η is taken obs =0.5; A μ is the longitudinal cross-sectional area of ​​the μth obstacle vehicle, which is used to characterize the size of the vehicle, μ∈(1,+∞); is the Euclidean distance between the ego vehicle and the μth obstacle vehicle at the tth moment; Q represents the influence range of the potential energy field of the vehicle. In this embodiment, Q=10m, that is, when the Euclidean distance between an obstacle vehicle and the ego vehicle exceeds Q, the potential energy of the vehicle in the current lane is not calculated; represents the relative speed between the ego vehicle and the μth obstacle vehicle at the tth moment. Due to the limitation of the 10 Hz planning frequency, the interval between t and t+1 is 0.1 s. It should be noted that the total number of lanes is limited to 2 in this embodiment, which can represent the basic model of lane change, that is, o∈[1,2]. The total number of perceived vehicles r is not limited and depends on the specific road conditions, so r∈[0,+∞) and μ∈[0,r]. Since the potential energy field needs to be calculated for each trajectory point in the process of lane change, t∈[0,t e ]; In addition, o and r are integers within the interval; the above process is valid only if When the difference is greater than the threshold value ξ, the vehicle has the intention to change lanes. After calculation, ξ=0.7 in this embodiment;

[0171] The gap threshold calculation process is as follows:

[0172] Get the minimum braking distance for the vehicle to avoid collision with the obstacle vehicle; the expression is as follows:

[0173]

[0174] In the formula, a bmax represents the maximum braking deceleration of the vehicle, and s represents the minimum braking distance without collision between the vehicle and the obstacle vehicle. At this time, we can get:

[0175]

[0176] That is, the potential energy field size of the oth lane can be regarded as a function of the minimum braking distance s, where s∈(0,+∞). This function is a convex function in the domain of definition. Considering safety, this embodiment calculates the value of the minimum braking distance s that can make the potential energy field of the oth lane Minimum, that is, solving the following equation:

[0177]

[0178] In the formula, represents the differential of the potential energy field of the oth lane with respect to the minimum braking distance s for the ego vehicle to avoid collision with the obstacle vehicle. The solution of this equation is regarded as the gap threshold ξ in this embodiment;

[0179] The potential energy field range is Figure 5 As shown;

[0180] It should be noted that after the lane-changing intention is generated, subsequent calculations are still required to determine whether a lane-changing decision is made;

[0181] S3.2. In the global evaluation layer, a set of optimal strategies is generated between the ego vehicle and the vehicle behind it in the target lane through a pure strategy game method, which can minimize the impact on the overall traffic operation. The steps to obtain the optimal strategy are as follows:

[0182] S3.2.1. Generate a set of strategies with the highest benefits between the ego vehicle and the vehicle behind in the target lane by using a pure strategy game method, and make a lane change decision;

[0183]

[0184] Where U VC (t) and U VT(t) represents the benefit value of the current lane and the target lane of the ego vehicle at the tth moment, respectively; U(t) is the total benefit of the local traffic system at the tth moment; Γ represents the number of all obstacle vehicles in front of the current lane of the ego vehicle; p and q are the number of all obstacle vehicles in front of and behind the target lane from the perspective of the ego vehicle, respectively; ω t is the time weight value of γ at time t; V(t), and represents the self-vehicle, the obstacle vehicle in front of the current lane, the obstacle vehicle in front of the target lane, and the obstacle vehicle behind the target lane. The speed of the obstacle vehicle at the tth moment; where Γ∈[0,+∞), p∈[0,+∞), q∈[0,+∞), γ∈[0,+∞), ρ∈[0,+∞), The specific value depends on the perception limit of the vehicle, so the values ​​of Γ, p and q are a dynamic process;

[0185] All strategy combinations include: "the ego vehicle changes lanes and the vehicle behind the target lane accepts" and "the ego vehicle does not change lanes and the vehicle behind the target lane accepts". The global benefit value of each strategy combination is calculated based on the above global benefit function, and the evaluation is completed by finding the optimal benefit combination. It should be noted that the global benefit evaluation will continue with the subsequent evaluation only when the ego vehicle strategy is to change lanes, otherwise the decision of not changing lanes is directly output.

[0186] S3.3, if both layers have made lane change strategies, the passenger's needs are further evaluated by comparing the collision time between the vehicle and the vehicle behind in the target lane to determine whether the passenger's personalized needs are met;

[0187] Specifically, in the personalized evaluation layer, a mapping relationship between the minimum acceptable collision time of passengers and the passenger demand factor f is established, and the establishment process is as follows:

[0188] S3.3.1. Predicted collision duration; the expression for predicted collision duration is as follows:

[0189]

[0190] Where, T A Indicates the predicted collision duration; D TF is the Euclidean distance between the ego vehicle and the vehicle behind in the target lane; |vv TF | represents the absolute difference in speed between the ego vehicle and the vehicle behind in the target lane; v represents the speed of the ego vehicle; v TF is the speed of the vehicle behind in the target lane;

[0191] S3.3.2. Establish the mapping relationship between the passenger's acceptable collision time and the passenger demand factor f, expressed as follows:

[0192] Where, T P Indicates the acceptable collision time for passengers; T S is a time constant, which is 3 seconds in this embodiment; T Indicates the scaling weight, which is set to 0.3 based on experience;

[0193] S3.3.3, the final output is obtained by predicting the collision duration and mapping relationship;

[0194] Specifically, based on the above analysis, when T A >T P When , it indicates that the predicted collision time with the vehicle behind in the target lane meets the needs of the passenger, that is, the final output allows the decision to change lanes; otherwise, it indicates that the passenger rejects the decision to change lanes, and the decision to change lanes will not be executed;

[0195] S4. By establishing a trajectory generation model based on multidimensional coordinate space, the lane-changing trajectory planning decision problem of the ego vehicle is transformed into a multivariate linear equation system for solving the lane-changing trajectory; the expression is as follows:

[0196]

[0197] in, as well as They are respectively represented by the initial time t of the lane change of the vehicle s The projected distance on the centerline of the target lane, the initial speed, the initial acceleration, the initial plane horizontal coordinate, the initial plane vertical coordinate, and the slope of the path point at the initial time. It should be noted that in this embodiment, t s =0; as well as They represent the projection distance of the vehicle on the center line of the target lane at the end of the lane change, the speed at the end, the acceleration at the end, the plane horizontal coordinate at the end, the plane vertical coordinate at the end, and the end time t e The slope of the path point; for the convenience of calculation, the above state quantities are combined into the initial state vector and the final state vector When the above two state vectors are determined, the trajectory polynomial coefficient vectors α and β can also be determined by solving a multivariate linear equation system. Since all state quantities in the initial state vector can be obtained by the vehicle-mounted sensors, it is necessary to further calculate the state quantities in the final state vector. As noted, the target lane centerline is actually a sequence of discrete points. It is only necessary to search for the point that matches the target lane centerline through the Euclidean distance. The nearest point, and through linear interpolation between its previous point and the next point, can solve all except the end time t e , the speed at the end And acceleration All state quantities of ;

[0198] S5. Optimizing and solving unknown quantities of the multivariate linear equations for the lane-changing trajectory;

[0199] That is, by setting the cost function and constraints that take into account the passenger preferences, the problem of solving the unknowns of the lane-changing trajectory multivariate linear equations is transformed into a nonlinear optimization problem. The specific steps are as follows:

[0200] S5.1. Establish a multi-objective cost function considering safety, comfort and efficiency based on passenger preferences, including the following steps:

[0201] S5.1.1, introduce lane-level potential energy field as a safety constraint, taking into account relative velocity, acceleration and longitudinal cross-sectional area of ​​the vehicle along the y direction;

[0202] The lane-level potential energy field calculation expression is as follows:

[0203]

[0204] In the formula, as well as They respectively represent the projection distance values ​​of the ego vehicle, the obstacle vehicle in the target lane behind the ego vehicle, the obstacle vehicle in the target lane in front of the ego vehicle, and the obstacle vehicle in the current lane in front of the ego vehicle on the lane change target lane at the tth moment; as well as They represent the speed of the ego vehicle and the obstacle vehicle in the target lane behind the ego vehicle at time t respectively; and is the maximum braking deceleration of the ego vehicle and the obstacle vehicle in the target lane behind the ego vehicle. τ is the planning time step of the vehicle, and in this embodiment, τ=0.1s; Q is the potential energy influence range of each vehicle, and in this embodiment, Q=10m;

[0205] S5.1.2. The safety cost function is expanded by a second-order Taylor expansion at each trajectory point to satisfy the quadratic form; the expression is as follows:

[0206]

[0207]

[0208] In the formula, Js represents the security cost function; X represents the form of high-dimensional unknowns; X0 is the input of each trajectory point; is the gradient calculation; H represents the Hessian matrix; represents partial differential calculation;

[0209] S5.1.3. In the comfort and efficiency cost functions, the acceleration and jerk (acceleration change rate) of each trajectory point of the vehicle are introduced to characterize the cost function; the expression is as follows:

[0210]

[0211] In the formula, J c and J e They represent the comfort cost function and the efficiency cost function respectively; j(t) represents the jerk of the vehicle at time t; t e Indicates the total lane change time;

[0212] S5.1.4. Combining safety, comfort and efficiency, a multi-objective cost function is obtained; the expression is as follows:

[0213] J=w s ·J s +w c ·J c +w e ·J e

[0214] In the formula, w s 、w c and w e are the weight values ​​of the three indicators in the passenger preference weight vector P;

[0215] S5.2. Optimize the hyperparameters of the long short-term memory neural network through the whale optimization algorithm to obtain the upper and lower limits of the safety constraint interval;

[0216] Specifically, the optimized neural network is trained through open source data sets to output the predicted lane change duration of the vehicle at the start of the lane change, and the lane change duration constraint is obtained by combining the passenger riding preference factor f; the safety constraint is linearized to improve the solution speed of the model;

[0217] The specific steps are as follows:

[0218] S5.2.1. Establish a long short-term memory neural network based on the whale optimization algorithm to obtain the estimated driving distance; the expression is as follows:

[0219]

[0220] In the formula, is the speed of the vehicle at the initial moment of lane change, is the estimated driving distance, and Respectively represent the speed of the ego vehicle and the obstacle vehicle in the target lane in front of the ego vehicle;

[0221] S5.2.2. This embodiment collects a total of 400 sets of data from the NGSIM open source data set, randomly selects 70% of them as training sets, and the remaining 30% as test sets; and uses the whale optimization algorithm to optimize the hyperparameters of the long short-term memory neural network, including: learning rate and maximum number of iterations;

[0222] By calculating the loss value through the loss function, we can determine the value of the hyperparameter that the whale optimization algorithm needs to search for to minimize the loss value, and then update the model weights in reverse:

[0223]

[0224] In the formula, Λ represents the total number of training samples, here Λ=350; T φ and They represent the actual lane changing time and the lane changing time predicted by the model respectively; F represents the predicted loss value of each iteration, that is, the fitness function value of the whale optimization algorithm;

[0225] The final training results are as follows Figure 6 As shown in the figure, it can be seen that the prediction result of the lane changing time of the ego vehicle by the long short-term memory neural network optimized by the whale optimization algorithm is closer to the true value than the prediction result of the lane changing time of the ego vehicle by the ordinary long short-term memory neural network; the fitness function value of the whale optimization algorithm reaches the minimum and converges when it iterates to the 10th generation;

[0226] S5.2.3. Introduce the passenger preference factor f and design the following lane change time constraint form:

[0227] t e ∈[T min ,T max ]

[0228]

[0229] Where, T min represents the lower bound of the lane change time, T max represents the upper bound of the lane change duration, Represents the predicted lane change duration output by the model;

[0230] S5.2.4. Introduce collision constraints during vehicle driving to ensure that the vehicle does not collide;

[0231] Specifically, it is necessary to ensure that the vehicle does not collide during driving, so it is necessary to introduce collision constraints into the optimization problem; in order to improve the calculation efficiency of the model, the collision constraints of this embodiment can be expressed in a linearized form;

[0232] Specifically, if Figure 7 As shown, the distance projected on the target lane by the midpoint of the rear axle of the vehicle at the initial moment of lane change is Then, three vertices, namely the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper, can be calculated through geometric calculation as follows:

[0233]

[0234] In the formula, w ego is the vehicle width of the vehicle, in this embodiment, w ego =2m;θ t is the heading angle of the vehicle at time t, θ t ∈(0,2π);lr ego It represents the length from the midpoint of the rear axle of the vehicle to the rear bumper of the vehicle. In this embodiment, lr ego =1m; and They represent the projection distance values ​​of the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper on the center line of the target lane at time t, respectively. and The value range of is [0,+∞); and Combined into a vector S Ω , then obviously S Ω is the projection value of the vehicle's distance on the center line of the target lane χ∈{A,B,C} is a linear function; in this case, the linearization constraint of the ego-vehicle safety can be expressed as:

[0235]

[0236] In the formula, f colli and r colli They represent the safety distance threshold in front of the vehicle and the safety distance threshold behind the vehicle respectively. In this embodiment, f colli =r colli =2m;s 1,t 、s 2,t and 3,t They respectively represent the distance projection value of the obstacle vehicle behind the target lane on the target lane, the distance projection value of the obstacle vehicle in front of the target lane on the target lane, and the distance projection value of the obstacle vehicle in front of the lane where the vehicle is currently located on the target lane;

[0237] S5.2.5. The planned trajectory needs to be subject to motion state constraints at the end of the lane change;

[0238] Specifically, in order to ensure that the ego vehicle can smoothly follow the traffic flow of the target lane at the end of the lane change, it is necessary to impose motion state constraints on the end point of the lane change trajectory planned by the ego vehicle;

[0239] This embodiment introduces a vehicle following model to calculate the motion state of the lane change endpoint, and derives the acceleration constraint at the lane change endpoint as:

[0240]

[0241] In the formula, Represents the vector composed of the accelerations of the obstacle vehicles behind and in front of the vehicle on the target lane; Δs is the speed of the obstacle vehicle in front of the vehicle in the current lane at the end of the lane change; * is a distance constant with a value range of Δs * ∈[1,5], in this embodiment, Δs * =2; ω1 and ω2 are the vehicle distance weight value and the relative speed weight value respectively. The vehicle distance weight value range is ω1∈[0,1], and the vehicle distance weight value in this embodiment is ω1=1; the relative speed weight value range is ω2∈[0,1], and the relative speed weight value in this embodiment is ω2=1; l OV-3 and ego They represent the length of the obstacle vehicle in front of the current lane of the ego vehicle and the length of the ego vehicle respectively; therefore, the motion state constraints can be summarized as:

[0242]

[0243] In the formula, represents the projection distance of the vehicle on the center line of the target lane at time t; represents the speed of the vehicle at time t;

[0244] S5.3. The trajectory optimization problem described in the above steps S5.1 and S5.2 is organized into a nonlinear programming problem. The complete expression is as follows:

[0245]

[0246] Substitute the expression of the above nonlinear programming problem into the IPOPT sequential quadratic programming solver to obtain the personalized planning trajectory;

[0247] S6. In order to verify the effect of the trajectory planning model, this embodiment designs three types of passengers with different preferences, namely conservative passenger 1, ordinary passenger 2 and radical passenger 3. This embodiment conducts a simulation experiment on the experimental parameter configuration shown in Table 1, and the simulation road is a set of randomly generated smooth discrete points:

[0248] Table 1 Specific parameters of simulation experiment

[0249]

[0250] The lane change trajectory planning results for different test scenarios are shown in Figure 2. Figure 8 As shown;

[0251] The motion planning curves of the lane-changing trajectory planning method designed in this embodiment under three different passengers. In this scenario, all three passengers prefer to change lanes by their own vehicle, such as Figure 8 As shown in part (a), it can be seen that the aggressive passenger plans a shorter lane change path and uses a higher speed and acceleration to achieve the minimum lane change duration, as shown in Figure 8 Therefore, it can be seen that the lane change trajectory planning method proposed in this embodiment realizes personalized lane change trajectory planning by emphasizing differentiated lane change decision sensitivity, different riding experience, and the balance between driving efficiency and comfort.

[0252] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lane-changing trajectory planning method for an intelligent connected vehicle considering the personalized riding needs of passengers, characterized in that: The following steps are involved: S1. Convert the passenger's riding preference into a judgment matrix, and map the judgment matrix into a passenger preference weight vector; The passenger's riding preference includes: riding comfort preference, safety preference and efficiency preference; The preference for ride comfort, safety and efficiency is formed into a judgment matrix A in the form of a 3×3 matrix; wherein the element A in the i-th (i=1, 2, 3) row and the j-th (j=1, 2, 3) column is ij It represents the interaction between passengers’ perception of ride comfort, safety and efficiency; S2, establish a secondary mapping method to obtain the passenger preference factor through the passenger preference weight vector; The secondary mapping method includes: triangular membership function and Gaussian membership function; S3. Through the three-layer personalized lane-changing decision model, a mapping relationship between the passenger's acceptable collision time and the passenger demand factor is established to generate the lane-changing trajectory planning decision of the vehicle; The three-layer personalized lane-changing decision model includes: a risk assessment layer, a global benefit assessment layer, and a personalized assessment layer; S4, by establishing a trajectory generation model based on multidimensional coordinate space, the lane-changing trajectory planning decision problem of the ego vehicle is transformed into a multivariate linear equation system for solving the lane-changing trajectory; S5. Optimize and solve the unknown variables of the multivariate linear equations of the lane-changing trajectory to complete the lane-changing of the vehicle.

2. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The method for solving the passenger preference weight vector is the hierarchical analysis method, and the specific steps include: S1.

1. Calculate the sum of each column of the judgment matrix; the expression is as follows: In the formula, S j represents the sum of the elements in the jth column of the judgment matrix A; A ij represents the element in the i-th row and j-th column of the judgment matrix A. It means to sum the elements in the jth column; S1.2, standardize each element in the judgment matrix A; the expression is as follows: In the formula, A′ ij Represents the standardized judgment matrix elements; S1.

3. Calculate each weight value in the passenger preference weight vector; the expression is as follows: Where P represents the passenger preference weight vector; P k represents the kth element in the passenger preference weight vector; S1.

4. Ensure the rationality of the judgment matrix by performing consistency check on the customer preference weight vector. The specific steps are as follows: S1.4.

1. Calculate the maximum eigenvalue; the expression is as follows: In the formula, λ max is the maximum eigenvalue of the judgment matrix A, (A·P) k represents the kth element of the product of the judgment matrix A and the passenger preference weight vector P, which represents the comprehensive score of each factor; S1.4.

2. Calculate the consistency index; the expression is as follows: In the formula, CI represents the consistency index value, and n represents the matrix dimension.

3. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The steps of establishing the secondary mapping method and obtaining the passenger preference factor through the passenger preference weight vector are as follows: S2.

1. Define the triangular membership function: Wherein, a represents the left endpoint of the triangle in the triangular membership function, b represents the right endpoint of the triangle in the triangular membership function, and c represents the peak point of the triangle in the triangular membership function. x represents the input of the triangular membership function, i.e., each weight value in the passenger preference weight vector P. μT(x) represents the triangular membership function with input x. S2.2, define the Gaussian membership function to define fuzziness: In the formula, μ and σ represent the mean and standard deviation of Gaussian distribution respectively, y represents the output value of triangular membership function, μ G (y) represents the Gaussian membership function with input y; S2.

3. After inputting each weight value in the passenger preference weight vector P into the triangular membership function and then passing through the Gaussian membership function, the final passenger preference factor f can be obtained by summing up; the expression is as follows: Wherein, the passenger preference factor f∈[-1, 1] directly reflects the type of passenger, including: radical, ordinary and conservative, corresponding to passenger preference factors f=-1, f=0 and f=1 respectively.

4. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The steps of establishing a mapping relationship between the passenger's acceptable collision time and the passenger demand factor through the three-layer personalized lane-changing decision model and generating the lane-changing trajectory planning decision of the vehicle are as follows: S3.

1. In the risk assessment layer, the artificial potential field theory is introduced to calculate the risk value of each lane. The calculation formula is as follows: In the formula, represents the potential energy field size of the oth lane; r represents the number of all obstacle vehicles on the oth lane; t e Indicates the duration of lane change; η obs is the lane weight; A μ is the longitudinal cross-sectional area of ​​the μth obstacle vehicle, which is used to characterize the size of the vehicle, μ∈(1,+∞); is the Euclidean distance between the ego vehicle and the μth obstacle vehicle at the tth moment; Q represents the influence range of the vehicle's potential energy field, that is, when the Euclidean distance between an obstacle vehicle and the ego vehicle exceeds Q, the potential energy of the vehicle in the current lane is not calculated; represents the relative speed between the ego vehicle and the μth obstacle vehicle at the tth moment. Due to the limitation of the 10 Hz planning frequency, the interval between t and t+1 is 0.1 s. In addition, o and r are both integers within the interval. The above process is valid only if When the difference is greater than the threshold ξ, the ego vehicle has the intention to change lanes; The calculation process of the gap threshold ξ is as follows: Get the minimum braking distance for the vehicle to avoid collision with the obstacle vehicle; the expression is as follows: In the formula, a bmax represents the maximum braking deceleration of the vehicle, and s represents the minimum braking distance without collision between the vehicle and the obstacle vehicle. At this time, we can get: That is, the potential energy field of the oth lane can be regarded as a function of the minimum braking distance s, where s∈(0,+∞). In the domain of definition, this function is a convex function. When the minimum braking distance s takes a value that can make the potential energy field of the oth lane Minimum, that is, solving the following equation: In the formula, represents the differential of the potential energy field of the oth lane with respect to the minimum braking distance s for the ego vehicle to avoid collision with the obstacle vehicle. The solution of the equation is the gap threshold ξ; S3.

2. In the global evaluation layer, a set of optimal strategies is generated between the ego vehicle and the vehicle behind it in the target lane through a pure strategy game method, which can minimize the impact on the overall traffic operation. The steps to obtain the optimal strategy are as follows: S3.2.

1. Generate a set of strategies with the highest benefits between the ego vehicle and the vehicle behind in the target lane by using a pure strategy game method, and make a lane change decision; The global benefit function expression is: Where U VC (t) and U VT (t) represents the benefit value of the current lane and the target lane of the ego vehicle at the tth moment, respectively; U(t) is the total benefit of the local traffic system at the tth moment; Γ represents the number of all obstacle vehicles in front of the current lane of the ego vehicle; p and q are the number of all obstacle vehicles in front of and behind the target lane from the perspective of the ego vehicle, respectively; ω t is the time weight value at the tth moment; V(t), and represents the vehicle, the γth obstacle vehicle in front of the current lane, the ρth obstacle vehicle in front of the target lane, and the ρth obstacle vehicle behind the target lane. The speed of the obstacle vehicle at the tth moment; where Γ∈[0,+∞), p∈[0,+∞), q∈[0,+∞), γ∈[0,+∞), ρ∈[0,+∞), The specific value depends on the perception limit of the vehicle, so the values ​​of Γ, p and q are a dynamic process; All strategy combinations include: "the ego vehicle changes lanes and the vehicle behind the target lane accepts" and "the ego vehicle does not change lanes and the vehicle behind the target lane accepts". The global benefit value of each strategy combination is calculated based on the above global benefit function, and the evaluation is completed by finding the optimal benefit combination. S3.

3. If both layers have made lane change strategies, the passenger's needs are further evaluated by comparing the collision time between the vehicle and the vehicle behind in the target lane to determine whether the passenger's personalized needs are met. The steps are as follows: S3.3.

1. Predicted collision duration; the expression for predicted collision duration is as follows: Where, TA 表 The predicted collision time is shown; D TF is the Euclidean distance between the ego vehicle and the vehicle behind in the target lane; |VV TF | represents the absolute difference in speed between the ego vehicle and the vehicle behind in the target lane; v represents the speed of the ego vehicle; V TF is the speed of the vehicle behind in the target lane; S3.3.

2. Establish the mapping relationship between the passenger's acceptable collision time and the passenger demand factor f, expressed as follows: T P =T S -η T ·f Where, T P Indicates the acceptable collision time for passengers; T S is a time constant, which is 3 seconds in the present invention; η T Indicates the scaling weight, which is set to 0.3 based on experience; S3.3.3, the final output is obtained by predicting the collision duration and mapping relationship; The final output is when T A >T P When the predicted collision time with the vehicle behind in the target lane meets the needs of the passenger, the final output is a decision to allow the lane change to be executed; otherwise, it indicates that the passenger rejects the lane change decision, and the lane change decision is not executed.

5. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The expression of the lane-changing trajectory multivariate linear equation group is as follows: in, as well as They are respectively represented by the initial time t of the lane change of the vehicle s The projected distance on the centerline of the target lane, the initial speed, the initial acceleration, the initial plane abscissa, the initial plane ordinate, and the slope of the path point at the initial time; as well as They represent the projection distance of the vehicle on the center line of the target lane at the end of the lane change, the speed at the end, the acceleration at the end, the plane horizontal coordinate at the end, the plane vertical coordinate at the end, and the end time t e The slope of the path point; for the convenience of calculation, the above state quantities are combined into the initial state vector and the final state vector When the above two state vectors are determined, the trajectory polynomial coefficient vectors α and β can also be determined by solving a multivariate linear equation system; Since all state quantities in the initial state vector can be obtained through the vehicle-mounted sensors, it is necessary to further calculate the state quantities in the final state vector; As we can see, the center line of the target lane is actually a sequence of discrete points. We only need to search for the point that matches the center line of the target lane by using the Euclidean distance. The nearest point, and through linear interpolation between its previous point and the next point, can solve all except the end time t e , the speed at the end And acceleration All state quantities.

6. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The steps of the optimization solution are as follows: S5.

1. Establish a multi-objective cost function considering safety, comfort and efficiency based on passenger preferences, including the following steps: S5.1.1, introduce lane-level potential energy field as a safety constraint, taking into account relative velocity, acceleration and longitudinal cross-sectional area of ​​the vehicle along the y direction; The lane-level potential energy field calculation expression is as follows: In the formula, as well as They respectively represent the projection distance values ​​of the ego vehicle, the obstacle vehicle in the target lane behind the ego vehicle, the obstacle vehicle in the target lane in front of the ego vehicle, and the obstacle vehicle in the current lane in front of the ego vehicle on the lane change target lane at the tth moment; as well as They represent the speed of the ego vehicle and the obstacle vehicle in the target lane behind the ego vehicle at time t respectively; and is the maximum braking deceleration of the ego vehicle and the obstacle vehicle behind the ego vehicle in the target lane; τ is the planning time step of the ego vehicle; Q is the potential energy influence range of each vehicle; S5.1.

2. The safety cost function is expanded by a second-order Taylor expansion at each trajectory point to satisfy the quadratic form; the expression is as follows: In the formula, J s represents the security cost function; X represents the form of high-dimensional unknown number; X0 is the input of each trajectory point; is the gradient calculation; H represents the Hessian matrix; represents partial differential calculation; S5.1.

3. In the comfort and efficiency cost functions, the acceleration and jerk of each trajectory point of the vehicle are introduced to characterize the cost; the expression is as follows: In the formula, J c and J e They represent the comfort cost function and the efficiency cost function respectively; j(t) represents the jerk of the vehicle at time t; t e Indicates the total lane change time; S5.1.

4. Combining safety, comfort and efficiency, a multi-objective cost function is obtained; the expression is as follows: J=w s ·J s +w c ·J c +w e ·J e In the formula, w s 、w c and w e are the weight values ​​of the three indicators in the passenger preference weight vector P; S5.

2. Optimize the hyperparameters of the long short-term memory neural network through the whale optimization algorithm to obtain the upper and lower limits of the safety constraint interval; The specific steps are as follows: S5.2.

1. Establish a long short-term memory neural network based on the whale optimization algorithm to obtain the estimated driving distance; the expression is as follows: In the formula, is the speed of the vehicle at the initial moment of lane change, is the estimated driving distance, and Respectively represent the speed of the ego vehicle and the obstacle vehicle in the target lane in front of the ego vehicle; S5.2.

2. Collect data sets from open source data sets, randomly select 70% of them as training sets and the remaining 30% as test sets; and use the whale optimization algorithm to optimize the hyperparameters of the long short-term memory neural network, including: learning rate and maximum number of iterations; By calculating the loss value through the loss function, we can determine the value of the hyperparameter that the whale optimization algorithm needs to search for to minimize the loss value, and then update the model weights in reverse: In the formula, Λ represents the total number of training samples; T φ and They represent the actual lane changing time and the lane changing time predicted by the model respectively; F represents the predicted loss value of each iteration, that is, the fitness function value of the whale optimization algorithm; S5.2.

3. Introduce the passenger preference factor f and design the following lane change time constraint form: t e ∈[T min ,T max ] Where, T min represents the lower bound of the lane change time, T max represents the upper bound of the lane change duration, Represents the predicted lane change duration output by the model; S5.2.

4. Introduce collision constraints during vehicle driving to ensure that the vehicle does not collide; Specifically, it is necessary to ensure that the vehicle does not collide during driving, so it is necessary to introduce collision constraints into the optimization problem; in order to improve the calculation efficiency of the model, the collision constraints of the present invention can be expressed in a linearized form; The distance projected on the target lane by the midpoint of the rear axle of the vehicle at the initial moment of lane change is Then, three vertices, namely the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper, can be calculated through geometric calculation as follows: In the formula, w ego is the width of the vehicle; θ t is the heading angle of the vehicle at time t, θ t ∈(0,2π);lr ego Indicates the length from the midpoint of the vehicle's rear axle to the vehicle's rear bumper; and They represent the projection distance values ​​of the left point of the rear bumper, the left point of the front bumper, and the right point of the front bumper on the center line of the target lane at time t, respectively. and The value range of is [0,+∞); and Combined into a vector S Ω , then obviously S Ω is the projection value of the vehicle's distance on the center line of the target lane A linear function of ; at this time, the linearization constraint of the vehicle safety can be expressed as: In the formula, f colli and r colli They represent the safety distance threshold in front of the vehicle and the safety distance threshold behind the vehicle respectively; s 1,t 、s 2,t and 3,t They respectively represent the distance projection value of the obstacle vehicle behind the target lane on the target lane, the distance projection value of the obstacle vehicle in front of the target lane on the target lane, and the distance projection value of the obstacle vehicle in front of the lane where the vehicle is currently located on the target lane; S5.2.5, The planned trajectory needs to impose motion state constraints at the end of the lane change; Specifically, in order to ensure that the ego vehicle can smoothly follow the traffic flow of the target lane at the end of the lane change, it is necessary to impose motion state constraints on the end point of the lane change trajectory planned by the ego vehicle; The present invention introduces a vehicle following model to calculate the motion state of the lane change endpoint, and derives the acceleration constraint at the lane change endpoint as: In the formula, Represents the vector composed of the accelerations of the obstacle vehicles behind and in front of the vehicle on the target lane; Δs is the speed of the obstacle vehicle in front of the vehicle in the current lane at the end of the lane change; * is a distance constant with a value range of Δs * ∈[1,5]; ω1 and ω2 are the vehicle distance weight value and relative speed weight value respectively. The vehicle distance weight value ranges from ω1∈[0,1]; the relative speed weight value ranges from ω2∈[0,1]; l OV-3 and ego They represent the length of the obstacle vehicle in front of the current lane of the ego vehicle and the length of the ego vehicle respectively; therefore, the motion state constraints can be summarized as: In the formula, represents the projection distance of the vehicle on the center line of the target lane at time t; represents the speed of the vehicle at time t; S5.

3. The trajectory optimization problem described in step S5.1 and step S5.2 is organized into a nonlinear programming problem. The complete expression is as follows: Substitute the expression of the above nonlinear programming problem into the IPOPT sequential quadratic programming solver to obtain the personalized planning trajectory.

7. The method for planning lane-changing trajectories of an intelligent connected vehicle taking into account the personalized riding needs of passengers according to claim 1, characterized in that: The method also includes: S6. In order to verify the effect of the trajectory planning model, three passengers with different preference types are designed, namely conservative passenger 1, ordinary passenger 2 and radical passenger 3, and simulation experiments are carried out to verify the effectiveness of a lane changing trajectory planning method for intelligent connected vehicles that takes into account the personalized riding needs of passengers.

Images