A multi-objective vehicle speed planning method for intelligent vehicles oriented to driving comfort and traffic efficiency

Abstract

The present application belongs to the technical field of automatic driving test, and in particular to a multi-target vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency. The method comprises the following steps: step one, establishing a quarter vehicle suspension dynamics equation and a continuous time state space equation, and discretizing the state space equation; step two, the intelligent vehicle uses the global road information obtained through the Internet of Vehicles to divide the road section and generate a vehicle speed adjustment position; step three, a cost function is established according to the planning target and the generated vertical cost, and a non-dominated beetle multi-target vehicle speed sequence generation method is designed, and finally an optimal vehicle speed sequence is generated combined with the preprocessed forward-looking information. The present application can significantly improve the vertical performance, longitudinal performance and traffic performance of the vehicle, and provides a new solution for intelligent vehicle speed planning aiming at driving comfort and traffic efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automatic driving test, and particularly relates to a multi-target vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency. BACKGROUND

[0002] Intelligent vehicles are the frontier of future vehicle development, and the suspension system as an important component part plays an important role in ensuring the driving comfort of vehicles. However, for some rough road sections, the suspension system alone cannot filter the impact of the road well. The vehicle needs to rely on reducing the speed to pass smoothly. Therefore, in order to balance the driving comfort and traffic efficiency of the vehicle, it is feasible and necessary to plan the speed of the vehicle according to the road bumping degree.

[0003] Intelligent vehicle speed planning can generally be divided into two parts: pre-processing of forward-looking information and generation of vehicle speed sequence. The rapidly developing vehicle networking technology can provide comprehensive forward-looking information for intelligent vehicles, such as traffic light signals, vehicle positions and road elevation information, and the obtained forward-looking information is pre-processed according to the planning target. By using the pre-processed forward-looking information, the intelligent vehicle can generate a vehicle speed sequence for the entire road.

[0004] However, on the one hand, most of the current vehicle speed sequence generation methods are oriented towards fuel economy and safety of the vehicle, and little attention is paid to the driving comfort and traffic efficiency of the vehicle. On the other hand, the vehicle often needs to consider multiple contradictory targets when generating a vehicle speed sequence. In pursuit of traffic efficiency, the vehicle often sacrifices part of the driving comfort, and vice versa. The traditional method converts multiple targets into a single target problem by artificially assigning weights to multiple targets, and then solves the problem. However, due to the contradiction between different targets, it is difficult to find the best balance point between multiple targets by adjusting the weights. SUMMARY

[0005] To solve the above problems, the application provides a multi-target vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency, which can significantly improve the vertical performance, longitudinal performance and traffic performance of the vehicle, and provides a new solution for intelligent vehicle speed planning aiming at driving comfort and traffic efficiency.

[0006] The technical scheme of the application is described below in combination with the drawings:

[0007] A multi-target vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency, comprising the following steps:

[0008] Step one, establish a quarter vehicle suspension dynamics equation and a continuous time state space equation, and perform state space equation discretization;

[0009] Step two, intelligent car uses global road information obtained by Internet of vehicles to divide road sections and generate speed adjustment points;

[0010] Step three, cost function is established according to planning target and generated vertical cost, and non-dominated beetle multi-objective speed sequence generation method is designed, and finally the optimal speed sequence is generated combined with the pre-processed look-ahead information.

[0011] Further, the specific method of step one is as follows:

[0012] 11) According to the quarter vehicle suspension model and the principle of vehicle dynamics, the quarter vehicle suspension dynamics equation is obtained, as shown below:

[0013]

[0014] In the formula, m s is the sprung mass, m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k t is the vertical equivalent stiffness of the tire; z s is the vertical displacement of the sprung mass; z u is the vertical displacement of the unsprung mass; z r is the road height;

[0015] 12) Select the sprung mass displacement z s (t), the unsprung mass displacement z u (t), the sprung mass displacement velocity The unsprung mass displacement velocity as state variables; select the sprung mass acceleration The suspension dynamic stroke z u (t)-z s (t), the tire dynamic deflection z r (t)-z u (t) as the output variables of the system; and select the road height z r and its rate of change as the disturbance of the system, and the quarter vehicle suspension dynamics equation (1) is rewritten as the state space equation of the system under continuous time, as shown below:

[0016]

[0017] Wherein,

[0018]

[0019] In the formula, x(t) is the state variable, is the derivative of the state variable; y(t) is the output variable; ω(t) is the disturbance variable; A is the suspension system matrix; C is the suspension output matrix; E is the state disturbance matrix; L is the output disturbance matrix; z s is the sprung mass displacement; z u is the unsprung mass displacement; is the sprung mass displacement velocity; is the unsprung mass displacement velocity; z r is the road height; m s is the sprung mass; m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k t is the vertical equivalent stiffness of the tire;

[0020] 13) The continuous-time matrix in equation (2) is discretized using the zero-order hold discretization method, and the discrete system state space equation is obtained as follows:

[0021]

[0022] wherein A is the suspension system matrix; E is the state disturbance matrix; C is the suspension output matrix; L is the output disturbance matrix; T s is the sampling period of the suspension system; is the discrete suspension system matrix; is the discrete state disturbance matrix; x(k) is the state variable of the kth period; y(k) is the output variable of the kth period; ω(k) is the disturbance variable of the kth period.

[0023] Further, the specific method of step two is as follows:

[0024] 21) The intelligent vehicle obtains the forward-looking information, i.e., the global road elevation information, through the Internet of Vehicles, and divides the entire road to be traveled into bumpy road sections and flat road sections using the global road elevation information;

[0025] 22) The entire road to be traveled is divided into a plurality of vehicle speed adjustment positions; each vehicle speed adjustment position is provided with a plurality of candidate vehicle speeds at intervals of δV, from the lowest vehicle speed V min to the highest vehicle speed V max .

[0026] Further, the specific determination method of the vehicle speed adjustment position is as follows:

[0027] (1) Assuming that the vehicle speed is constant on the bumpy road section, each bumpy road section is regarded as a vehicle speed adjustment position;

[0028] (2) Set 3 distance l k speed adjustment points before and after each vehicle speed change event to change the speed in time;

[0029] (3) When the vehicle is driving on a flat road section between two vehicle speed change events, no speed adjustment points are set between the two vehicle speed change events in addition to the speed adjustment points set in (2), to avoid frequent acceleration and deceleration of the vehicle.

[0030] Further, the specific method of step three is as follows:

[0031] 31) Determine the planning target;

[0032] The planning target includes vertical performance target, longitudinal performance target and traffic performance target;

[0033] The vertical performance includes sprung mass acceleration, tire dynamic deflection and suspension dynamic travel; the global road elevation information obtained by the Internet of Vehicles is injected into the passive suspension model, i.e. the quarter vehicle suspension model is established; the output variable y(k) at each sampling period is obtained through the discrete system state space equation (3), and the vertical performance of the corresponding rough road section is calculated as follows:

[0034]

[0035] In the formula, c i is the vertical performance of the i-th rough road section; N i is the suspension system sampling number of the i-th rough road section; y1(k) is the sprung mass acceleration at the k-th sampling period; y2(k) is the suspension dynamic travel at the k-th sampling period; y3(k) is the tire dynamic deflection at the k-th sampling period; σ1 is the weight coefficient of the sprung mass acceleration; σ2 is the weight coefficient of the suspension dynamic travel; σ3 is the weight coefficient of the tire dynamic deflection; l i is the length of the i-th rough road section; V i is the vehicle speed through the i-th rough road section; T s is the suspension system sampling time;

[0036] The cost function of the vertical performance target is established as follows:

[0037]

[0038] In the formula, V i is the vehicle speed through the i-th rough road section; Jver(Vi) is the vertical cost of the vehicle through the whole road; N b is the number of rough road sections; c i is the vertical performance of the i-th rough road section;

[0039] The cost function for defining the longitudinal performance target is the sum of the absolute values of the longitudinal acceleration of the vehicle in the whole road, reflecting the acceleration and deceleration of the vehicle, as shown below:

[0040]

[0041] wherein V i is the vehicle speed through the i-th speed adjustment point; J acc (V i ) is the longitudinal cost of the vehicle through the whole road; N ad is the number of speed adjustment points; Ls i is the distance from the start point of the i-th speed adjustment point to the start point of the whole road; Le i is the distance from the end point of the i-th speed adjustment point to the start point of the whole road.

[0042] The passing performance target represents the total time for the vehicle to pass through the whole road, and the cost function is shown as follows:

[0043]

[0044] wherein V i is the vehicle speed through the i-th speed adjustment point; J t (V i ) is the longitudinal cost of the vehicle through the whole road; N ad is the number of speed adjustment points; Ls i is the distance from the start point of the i-th speed adjustment point to the start point of the whole road, Le i is the distance from the end point of the i-th speed adjustment point to the start point of the whole road.

[0045] 32) Design a non-dominated beetle multi-objective speed sequence generation method according to the cost function of the established vertical performance target, the cost function of the longitudinal performance target and the cost function of the passing performance target to generate the optimal speed sequence; first, generate an initial speed sequence Z t ; then, iterate the speed sequence to generate an iterated speed sequence K t ; then, perform Pareto non-dominated sorting on the speed sequences before and after iteration Z t and K t , and divide them into n non-dominated sets F(1)-F(n); finally, select k speed sequences with smaller cost functions in order as the initial speed sequence Z t:1 for the next iteration.

[0046] Further, the specific method of step 32) is as follows:

[0047] 321) Generate a speed sequence;

[0048] Randomly generate k initial speed sequences Zt ; wherein each vehicle speed sequence is a one-dimensional array with a length of N s , the i-th element of which represents the speed of the vehicle at the i-th speed adjustment bit;

[0049] 322) iterating the vehicle speed sequence;

[0050] The vehicle speed sequence is divided into global exploration and local optimization when iterating;

[0051] In the global exploration, a global exploration sequence is used to search for the optimal vehicle speed sequence by moving away from the global worst sequence, i.e., the vehicle speed sequence with the highest cost. The iteration method of the global exploration sequence is as follows:

[0052] Xsi(t+1)＝Xsi(t)+ α ×k×Xsi(t -1)+β ×ΔX(t) (10)

[0053]

[0054] ΔX(t)=|Xs i (t)-X w (t)| (12)

[0055] In the formula, t is the current iteration number; Xs i (t) is the i-th search sequence at the t-th iteration; X w (t) is the global worst vehicle speed sequence at the t-th iteration; α is the variation coefficient; k is the inertia coefficient; β is the deflection coefficient; η is a random number between zero and one; and λ is the mutation probability;

[0056] In the local optimization, a local search sequence, a global search sequence, and a hybrid search sequence are used for detailed local optimization. Among them, the local optimal exploration sequence only searches in the local optimal region, while the global search sequence and the hybrid exploration sequence search in the global optimal region. Among them, the local optimal region and the global optimal region will change with the iteration number:

[0057]

[0058] In the formula, Lb * is the lower bound of the optimal region; Ub * is the upper bound of the local optimal region; Lb a is the lower bound of the global optimal region; Ub a is the upper bound of the global optimal region; t is the current iteration number; b * (t) is the local optimal vehicle speed sequence at the t-th iteration; b a (t) is the global optimal vehicle speed sequence at the t-th iteration; T max is the maximum iteration number; and Vmin is the lowest vehicle speed; V max is the highest vehicle speed;

[0059] The three sequences (local search sequence, global search sequence, and hybrid search sequence) will iterate in different ways; among them, the iteration way of the local search sequence is as follows:

[0060] Xli(t+1)＝b * (t)+f1×(Xli(t) -Lb * )+f2×(Xli(t) -Ub * ) (14)

[0061] In the formula, t is the current iteration number; Xl i (t) is the i-th local search sequence at the t-th iteration; b * (t) is the local optimal vehicle speed sequence at the t-th iteration; f1, f2 are 1×D random vectors; Lb * is the lower bound of the local optimal region; Ub * is the upper bound of the local optimal region;

[0062] The iteration way of the global search sequence is as follows:

[0063] Xgi(t+1)＝Xgi(t)+c×(Xgi(t) -Lb a )+d×(Xgi(t) -Ub a ) (15)

[0064] In the formula, t is the current iteration number; Xg i (t) is the i-th global search sequence at the t-th iteration; c is a 1×D vector subject to normal distribution; d is a 1×D random vector; Lb a is the lower bound of the global optimal region; Ub a is the upper bound of the global optimal region;

[0065] The iteration way of the hybrid search sequence is as follows:

[0066] Xc i (t+1)=b a (t)+g×(|Xc i (t)-b * (t)|+|Xc i (t)-b a (t)|) (16)

[0067] In the formula, t is the current iteration number; Xc i (t) is the i-th hybrid search sequence at the t-th iteration; b *(t) is the local optimal speed sequence at the tth iteration; b a (t) is the global optimal speed sequence at the tth iteration, g is a random vector of 1 x D;

[0068] k initial speed sequences Z t After iteration by formula (10), formula (14), formula (15) and formula (16), k new speed sequences K t are generated; the 2k speed sequences before and after iteration are denoted as X t , wherein the ith speed sequence is X t (i);

[0069] 323) Perform a Pareto non-dominated sorting;

[0070] Select k speed sequences with smaller cost functions from all speed sequences before and after iteration for the next iteration; therefore, a Pareto non-dominated sorting needs to be performed on all speed sequences, as follows:

[0071] First, calculate the vertical performance, longitudinal performance and traffic performance of all speed sequences by the established cost function; then, select all sequences that are not dominated by other speed sequences and store them in the non-dominated set F(1); wherein, if a speed sequence X t (a) is not worse than another speed sequence X t (b) in terms of any defined planning objective, and X t (a) is better than X t (b) in terms of at least one planning objective, then X t (b) is dominated by X t (a); then, remove the sequences contained in F(1) from all speed sequences, i.e. X t , and select sequences that are not dominated by other speed sequences from the remaining speed sequences and store them in the non-dominated set F(2); in this way, until all speed sequences are sorted and stored in the non-dominated sets F(1)-F(n); according to the definition and properties of domination, it can be seen that the speed sequences in the non-dominated set F(1) are optimal, the speed sequences in F(2) are second, and the speed sequences in F(n) are the worst;

[0072] 324) Select a speed sequence;

[0073] First, store F(i) in the initial speed sequence Z t:1 for the next iteration in order from small to large according to the order of the non-dominated set, until F(p) is added, Z t:1the number of sequences in F (p) is greater than the initial number of vehicle speed sequences k; then, the non-dominated set F (p+1)-F (n) is eliminated, and the sequences in F (p) are sorted by using the concept of crowding distance; the crowding distance represents the density of the surrounding sequences of the selected vehicle speed sequence, and the crowding distance of sequence X t (n) is the volume of the cuboid Q t (n) that contains X t (n) itself but does not contain other sequences; finally, the vehicle speed sequence with a smaller crowding distance is selected for the next iteration; if the non-dominated set F (1)-F (p-1) contains m vehicle speed sequences, the first k-m vehicle speed sequences with smaller crowding distance in F (p) are selected, that is, F1 (p), and are stored in Z t:1 to make the number of sequences contained in Z t:1 equal to the initial number of vehicle speed sequences k, and the remaining vehicle speed sequences F2 (p) are eliminated, and then the vehicle speed sequences in Z are used for the next iteration.

[0074] The beneficial effects of the present application are as follows:

[0075] 1) The present application provides an intelligent vehicle multi-objective vehicle speed planning method for driving comfort and traffic efficiency, which can comprehensively consider the vertical performance, longitudinal performance and traffic performance of the vehicle during vehicle speed planning, and improve the driving comfort and traffic efficiency of the vehicle.

[0076] 2) The present application designs a forward-looking information preprocessing method based on vehicle networking, which obtains global road information through vehicle networking, and performs road section division and vehicle speed change event determination, and then generates a vehicle speed adjustment position, which can provide comprehensive information for the optimal vehicle speed sequence generation part, so that the vehicle speed can be adjusted in time to adapt to different road surfaces with different bumping degrees.

[0077] 3) The present application uses the scarab optimizer as the basis for generating the optimal vehicle speed sequence. The scarab optimizer is used to generate the optimal vehicle speed sequence from a variety of different speed sequences, and has higher convergence accuracy than other intelligent group optimization algorithms, and is easier to generate the optimal vehicle speed sequence.

[0078] 4) The present application designs a multi-objective vehicle speed sequence generation method, and designs a non-dominated scarab multi-objective vehicle speed sequence generation method based on the Pareto optimality theory, which can balance the mutually contradictory vertical performance, longitudinal performance and traffic performance, and generate the optimal vehicle speed sequence. BRIEF DESCRIPTION OF DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0080] Figure 1 Fig. 1 is a schematic diagram of a multi-objective vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency according to the present application;

[0081] Figure 2 Fig. 2 is a schematic diagram of a quarter vehicle suspension model;

[0082] Figure 3 Fig. 3 is a schematic diagram of road segment division;

[0083] Figure 4 Fig. 4 is a schematic diagram of look-ahead information processing;

[0084] Figure 5 Fig. 5 is a schematic diagram of a non-dominated multi-objective vehicle speed sequence generation method;

[0085] Figure 6 Fig. 6 is a schematic diagram of vehicle speed sequence congestion degree;

[0086] Figure 7 Fig. 7 is a schematic diagram of vehicle speed sequence comparison. DETAILED DESCRIPTION

[0087] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely intended for the purpose of interpretation of the present application and are not limiting of the present application. In addition, it should be noted that only the parts related to the present application are shown in the accompanying drawings for the purpose of description.

[0088] Embodiment One

[0089] Referring to Figure 1 , the present embodiment provides a multi-objective vehicle speed planning method for intelligent vehicles aiming at driving comfort and traffic efficiency, comprising the following steps:

[0090] Step One, referring to Figure 2 , a quarter vehicle suspension dynamics equation and a continuous time state space equation are established, and the state space equation is discretized, specifically as follows:

[0091] 11) The quarter vehicle suspension dynamics equation is obtained according to the quarter vehicle suspension model and vehicle dynamics principle, as shown below:

[0092]

[0093] In the formula, m s is the sprung mass, m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k tis the vertical equivalent stiffness of the tire; z s is the vertical displacement of the unsprung mass; z u is the vertical displacement of the sprung mass; z r is the road height;

[0094] 12) the sprung mass displacement z s (t) is selected as the state variable; the unsprung mass displacement z u (t) is selected as the state variable; the sprung mass displacement velocity the unsprung mass displacement velocity is selected as the state variable; the sprung mass acceleration the suspension dynamic travel z u (t) is selected as the state variable; the tire dynamic deflection z s (t) is selected as the state variable; the tire dynamic deflection z r (t) is selected as the state variable; the tire dynamic deflection z u (t) is selected as the state variable; the road height z r and its rate of change is the disturbance variable of the system, the quarter-car suspension dynamics equation (1) is rewritten as the system state space equation in continuous time, as follows:

[0095]

[0096] wherein,

[0097]

[0098]

[0099] wherein, x(t) is the state variable, is the derivative of the state variable; y(t) is the output variable; ω(t) is the disturbance variable; A is the suspension system matrix; C is the suspension output matrix; E is the state disturbance matrix; L is the output disturbance matrix; z s (t) is the sprung mass displacement; z u (t) is the unsprung mass displacement; is the sprung mass displacement velocity; is the unsprung mass displacement velocity; z r is the road height; m s is the sprung mass; m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k t is the vertical equivalent stiffness of the tire;

[0100] 13) The sensor in the suspension system is sampled in discrete time, and the state quantity obtained is discrete rather than continuous, so further, we need to discretize the system state space equation in continuous time. The zero-order hold discretization method is used to discretize the continuous-time matrix in equation (2), and the discrete system state space equation is obtained as follows:

[0101]

[0102] In the formula, A is the suspension system matrix, E is the state disturbance matrix, C is the suspension output matrix, L is the output disturbance matrix, T s is the sampling period of the suspension system; is the discrete suspension system matrix; is the discrete state disturbance matrix, x(k) is the state variable of the kth period, y(k) is the output variable of the kth period, and ω(k) is the disturbance variable of the kth period.

[0103] Step two, the intelligent vehicle uses the global road information obtained by the Internet of vehicles to perform road section division and vehicle speed adjustment bit generation, as follows:

[0104] 21) The intelligent vehicle obtains the forward-looking information, i.e., the global road elevation information, through the Internet of vehicles, and divides the entire road to be traveled into a bumpy road section and a flat road section using the global road elevation information, as shown in Figure 3 ;

[0105] The forward-looking information includes traffic light information, intersection information, global road elevation information, etc., but only the global road elevation information is used in the present application;

[0106] 22) In order to facilitate the adjustment of the vehicle speed, the entire road contains several vehicle speed adjustment bits. For each vehicle speed adjustment bit, there are multiple candidate vehicle speeds with a δV interval, from the lowest vehicle speed V min to the highest vehicle speed V max . Taking Figure 4 as an example, the combination of the vehicle speed adjustment bit and the preprocessed forward-looking information is introduced.

[0107] In the figure, the orange part is the bumpy road section, and the white part is the flat road section. Different colored dots represent different types of candidate vehicle speeds. The blue dot is the ordinary candidate vehicle speed, the red dot is the highest vehicle speed V max , and the green dot is the lowest vehicle speed V min . The entire road contains N s speed adjustment bits, so the speed adjustment bit at the end is denoted as Compared with normal driving, the vehicle changes the speed more frequently at the start, the bumpy road section and the approach to the end, so the start, the end and the bumpy road section are called "speed change events". In order to more clearly describe the determination method of the speed adjustment position, the part of the road after the start and before the end is omitted, as shown by the dashed line road part in the figure. The specific determination method of the speed adjustment position is as follows.

[0108] (1) Normally, the vehicle can only slowly drive on the bumpy road section, and it is difficult to produce acceleration and deceleration behavior. Therefore, the present application assumes that the vehicle speed is constant on the bumpy road section, and then each bumpy road section can be regarded as a speed adjustment position, that is, P b , P b:1 and P b:8 in the figure.

[0109] (2) In order to adapt to the frequent change of speed when approaching the speed change event, 3 speed adjustment positions with a distance of l k are set before and after each speed change event, respectively, for timely changing the speed, that is, P 2～4 , P b；3～b；1 , P b:2～b:4 , P b:5～b:7 , P b:9～b:11 and

[0110] (3) When the vehicle drives on the flat road section between two speed change events, it rarely performs large acceleration and deceleration to avoid too frequent speed adjustment. Therefore, in addition to the speed adjustment positions set in (2), no other speed adjustment positions are provided between the two speed change events to avoid frequent acceleration and deceleration of the vehicle.

[0111] So far, the pre-processing of the forward-looking information has been completed, and N s speed adjustment positions in the entire road have been generated.

[0112] Step three, according to the planning target and the generated vertical cost, a cost function is established, and a non-dominated scarab multi-objective speed sequence generation method is designed, and finally the optimal speed sequence is generated combined with the pre-processed forward-looking information, as follows:

[0113] 31) When generating the speed sequence, the planning target should be determined first, and then the corresponding cost function is established for generating the optimal speed sequence; in order to ensure the driving comfort and traffic efficiency of the vehicle, the planning target in the present application is as follows:

[0114] (1) Vertical performance target;

[0115] (2) Longitudinal performance target;

[0116] (3) Traffic performance target;

[0117] where the vertical performance includes sprung mass acceleration, tire deflection and suspension travel. To obtain the vertical performance of the vehicle passing the bump road section in the speed planning stage, the global road elevation information obtained by the Internet of Vehicles is injected into the passive suspension model. The output variable y(k) in each sampling period is obtained by the discrete system state space equation (3), and the vertical performance of the corresponding bump road section is calculated:

[0118]

[0119] where c i is the vertical performance of the i-th bump road section; N i is the suspension system sampling number of the i-th bump road section; y1(k) is the sprung mass acceleration in the k-th sampling period; y2(k) is the suspension travel in the k-th sampling period; y3(k) is the tire deflection in the k-th sampling period; σ1 is the weight coefficient of the sprung mass acceleration; σ2 is the weight coefficient of the suspension travel; σ3 is the weight coefficient of the tire deflection; l i is the length of the i-th bump road section; V i is the vehicle speed passing the i-th bump road section; T s is the suspension system sampling time;

[0120] Therefore, the cost function of the vertical performance target can be established as follows:

[0121]

[0122] where V i is the vehicle speed passing the i-th bump road section; Jver(Vi) is the vertical cost of the vehicle passing the whole road; N b is the number of bump road sections; c i is the vertical performance of the i-th bump road section;

[0123] The speed of the vehicle between adjacent speed adjustment positions is linearly changed, i.e. the acceleration is a constant. Therefore, the cost function of the longitudinal performance target is defined as the sum of the absolute values of the longitudinal acceleration of the vehicle in the whole road, reflecting the acceleration and deceleration of the vehicle, and is specifically as follows:

[0124]

[0125] where V i is the vehicle speed passing the i-th speed adjustment position; J acc (V i ) is the longitudinal cost of the vehicle passing the whole road; N ad is the number of speed adjustment positions; Ls i is the distance from the starting point of the i-th speed adjustment position to the starting point of the whole road; Le iDistance from the end of the ith speed adjustment position to the start of the whole road;

[0126] The passing performance objective represents the total time for a vehicle to pass through the whole road, and its cost function is expressed as follows:

[0127]

[0128] where V i is the vehicle speed through the ith speed adjustment position; J t is the cost of the vehicle passing through the ith speed adjustment position; N i is the number of speed adjustment positions; Ls ad is the distance from the start of the whole road to the start of the ith speed adjustment position; Le i is the distance from the start of the whole road to the end of the ith speed adjustment position; and L i is the distance from the end of the ith speed adjustment position to the start of the whole road.

[0129] 32) According to the established cost function, the non-dominated beetle multi-objective speed sequence generation method is designed to generate the optimal speed sequence, as shown in the following formula: Figure 5 First, the initial speed sequence Z t is generated; then, the speed sequence iteration is performed to generate the iterated speed sequence K t ; then, the speed sequences before and after iteration Z t and K t are subjected to Pareto non-dominated sorting, and are divided into n non-dominated sets F(1)-F(n); finally, the better k speed sequences are selected in order as the initial speed sequence Z t:1 for the next iteration. The specific steps are as follows:

[0130] 321) Speed sequence generation;

[0131] k initial speed sequences Z t are randomly generated; each speed sequence is a one-dimensional array with a length of N s , and the ith element in the array represents the speed of the vehicle at the ith speed adjustment position;

[0132] 322) Speed sequence iteration;

[0133] The speed sequence iteration can be divided into global exploration and local optimization.

[0134] In global exploration, the global exploration sequence is used to find the optimal speed sequence by moving away from the global worst sequence, i.e., the speed sequence with the highest cost. The iteration method of the global exploration sequence is as follows:

[0135] Xs i (t+1) = Xs i(t) + a x k x Xs i (t - 1) + β x ΔX(t) (10)

[0136]

[0137] ΔX(t) = |Xs i (t) - X w (t) | (12)

[0138] where t is the current iteration number; Xs i (t) is the i-th search sequence at the t-th iteration; X w (t) is the global worst speed sequence at the t-th iteration; a is the mutation coefficient; k is the inertia coefficient; β is the deflection coefficient; η is a random number between zero and one; λ is the mutation probability;

[0139] In local optimization, local search sequence, global search sequence and hybrid search sequence are used for detailed local optimization. Among them, the local optimal exploration sequence searches only in the local optimal region, while the global search sequence and the hybrid exploration sequence search in the global optimal region. Among them, the local optimal region and the global optimal region will change with the iteration number:

[0140]

[0141] where Lb * is the lower bound of the optimal region; Ub * is the upper bound of the local optimal region; Lb a is the lower bound of the global optimal region; Ub a is the upper bound of the global optimal region; t is the current iteration number; b * (t) is the local optimal speed sequence at the t-th iteration; b a (t) is the global optimal speed sequence at the t-th iteration; T max is the maximum iteration number; V min is the minimum speed; V max is the maximum speed;

[0142] The three sequences will iterate in different ways; among them, the iteration method of the local search sequence is as follows:

[0143] Xli(t+1)＝b * (t)+f1×(Xli(t) -Lb * )+f2×(Xli(t) -Ub * ) (14)

[0144] where t is the current iteration number; Xl i(t) is the i-th local search sequence at the t-th iteration; b * (t) is the local optimal speed sequence at the t-th iteration; f1, f2 are random vectors of 1 x D; Lb * is the lower bound of the local optimal region; Ub * is the upper bound of the local optimal region;

[0145] The iteration mode of the global search sequence is as follows:

[0146] Xgi(t+1)＝Xgi(t)+c×(Xgi(t) -Lb a )+d×(Xgi(t) -Ub a ) (15)

[0147] In the formula, t is the current iteration number; Xg i (t) is the i-th global search sequence at the t-th iteration; c is a vector of 1 x D subject to normal distribution; d is a random vector of 1 x D; Lb a is the lower bound of the global optimal region; Ub a is the upper bound of the global optimal region;

[0148] The iteration mode of the hybrid search sequence is as follows:

[0149] Xc i (t+1)=b a (t)+g×(|Xc i (t)-b * (t)|+|Xc i (t)-b a (t)|) (16)

[0150] In the formula, t is the current iteration number; Xc i (t) is the i-th hybrid search sequence at the t-th iteration; b * (t) is the local optimal speed sequence at the t-th iteration; b a (t) is the global optimal speed sequence at the t-th iteration; g is a random vector of 1 x D;

[0151] k initial speed sequences Z t After iteration by formula (10), formula (14), formula (15) and formula (16), k new speed sequences K t are generated; the 2k speed sequences before and after iteration are denoted as X t , wherein the i-th speed sequence is X t (i);

[0152] 323) Pareto non-dominated sorting;

[0153] To keep the number of speed profiles constant in each iteration, the k speed profiles with the smaller cost function are selected from all the speed profiles before and after the iteration to be used in the next iteration. Therefore, all the speed profiles need to be sorted in a non-dominated Pareto set, as shown in step three of Figure 5 .

[0154] First, the vertical performance, longitudinal performance and traffic performance of all the speed profiles are calculated by the cost function. Then, all the speed profiles that are not dominated by other speed profiles are selected and stored in the non-dominated set F(1). If a speed profile X t (a) is not better than another speed profile X t (b) in all the planning objectives, then X t (a) dominates X t (b). If X t (a) is better than X t (b) in at least one planning objective, then X t (b) is dominated by X t:1 (a). Then, the speed profiles in F(1) are removed from all the speed profiles, and the speed profiles that are not dominated by other speed profiles are selected from the remaining speed profiles and stored in the non-dominated set F(2). This process is repeated until all the speed profiles are sorted and stored in the non-dominated sets F(1)-F(n). According to the definition and properties of domination, the speed profiles in F(1) are the best, the speed profiles in F(2) are the second best, and the speed profiles in F(n) are the worst;

[0155] 324) Selection of speed profiles;

[0156] First, the non-dominated sets F(i) are stored in the initial speed profiles Z t:1 in order from small to large, as shown in step four of Figure 5 . When F(p) is added, the number of speed profiles in Z t:1 exceeds the initial number of speed profiles k. Then, the non-dominated sets F(p+1)-F(n) are eliminated, and the speed profiles in F(p) are sorted according to the concept of crowding distance. The crowding distance represents the density of the surrounding speed profiles of the selected speed profile, as shown by the green cuboid in Figure 6 . The volume of the cuboid Q t (n) that contains X t (n) itself but does not contain other speed profiles represents the crowding distance of X t (n). Finally, the speed profiles with small crowding distance are selected for the next iteration. If the non-dominated sets F(1)-F(p-1) contain m speed profiles, then the first k-m speed profiles with small crowding distance in F(p) are selected, i.e. F1(p), and stored in Z t:1The number of sequences contained therein reaches the initial vehicle speed sequence number k, and the remaining vehicle speed sequence F2(p) is eliminated. In turn, the vehicle speed sequence in Z t:1 can be used for the next iteration.

[0157] At this point, one iteration of the vehicle speed sequence is completed. After a maximum number of iterations T max , the optimal vehicle speed sequence V opt can be obtained.

[0158] Example Two

[0159] To verify the effectiveness and superiority of the method proposed in the present application, a road as shown in Figure 7 is established for simulation verification. The total length of the road is 1450m, of which 300m-500m, 400m-450m and 900m-1050m are rough road sections, and the rest are flat road sections. The traditional scarab optimizer is used for comparison with the method proposed in the present application, and the comparison of the obtained vehicle speed sequences is shown in the blue curve and the orange curve in Figure 7 .

[0160] It can be seen that the vehicle speed sequence planned by the multi-objective vehicle speed planning method proposed in the present application changes smoothly in the entire road, without dramatic acceleration and deceleration behavior, and most of the time is above the vehicle speed sequence planned by the traditional scarab optimizer. It shows that the vehicle speed planning method proposed in the present application can significantly improve the vertical performance, longitudinal performance and passing performance of the vehicle, and can improve the driving comfort and passing efficiency of the vehicle.

[0161] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-objective speed planning method for intelligent vehicles oriented towards driving comfort and traffic efficiency, characterized by: The following steps are involved: Step 1: Establish the quarter vehicle suspension dynamics equation and continuous-time state-space equation, and discretize the state-space equation; Step 2: The smart car uses the global road information obtained by the Internet of Vehicles to divide the road sections and generate the speed adjustment position; Step 3: Based on the planning objectives and the generated vertical costs, a cost function is established, and a method for generating multi-objective speed sequences for non-dominated dung beetles is designed. Finally, the optimal speed sequence is generated by combining the pre-processed forward-looking information. The specific method of step three is as follows: 31) Determine planning objectives; The planning objectives include vertical performance objectives, longitudinal performance objectives and traffic performance objectives; The vertical performance includes sprung mass acceleration, tire deflection, and suspension travel. The global road elevation information obtained by the Internet of Vehicles is injected into the passive suspension model, i.e., a quarter-vehicle suspension model is established. The output variable y(k) at each sampling period is obtained through the discretized system state space equation, and the vertical performance of the corresponding bumpy road section is calculated as follows: Where c i is the vertical performance of the i-th bumpy road section; N i is the number of suspension system samples in the i-th bumpy road section; y1(k) is the sprung mass acceleration of the k-th sampling period; y2(k) is the suspension dynamic travel of the k-th sampling period; y3(k) is the tire dynamic deflection of the k-th sampling period; σ1 is the weight coefficient of sprung mass acceleration; σ2 is the weight coefficient of suspension dynamic travel; σ3 is the weight coefficient of tire dynamic deflection; l i is the length of the i-th bumpy road section; V i is the speed of the vehicle passing through the i-th bumpy road section; T s is the sampling time of the suspension system; The cost function of the vertical performance objective is established as follows: Where V i is the speed of the vehicle passing through the i-th bumpy road section; J ver (V i ) is the vertical cost of the vehicle passing through the entire road; N b is the number of bumpy road sections; c i is the vertical performance of the i-th bumpy road section; The cost function for defining the longitudinal performance objective is the sum of the absolute values ​​of the vehicle's longitudinal acceleration along the entire road, reflecting the vehicle's acceleration and deceleration, as shown below: Where V i is the speed of the vehicle passing through the i-th speed adjustment position; J acc (V i ) is the longitudinal cost for a vehicle to pass through the entire lane; N ad The number of vehicle speed adjustment positions; Ls i Le is the distance from the starting point of the i-th speed adjustment position to the starting point of the entire road; i is the distance from the end point of the i-th speed adjustment position to the starting point of the entire road; The traffic performance objective represents the total time it takes for a vehicle to pass through the entire road, and the cost function is expressed as follows: Where V i is the speed of the vehicle passing through the i-th speed adjustment position; J t (V i ) is the longitudinal cost for a vehicle to pass through the entire lane; N ad The number of vehicle speed adjustment positions; Ls i Le is the distance from the starting point of the i-th speed adjustment position to the starting point of the entire road, i is the distance from the end point of the i-th speed adjustment position to the starting point of the entire road; 32) Based on the established cost functions of vertical performance objectives, longitudinal performance objectives, and traffic performance objectives, a non-dominated dung beetle multi-objective vehicle speed sequence generation method is designed to generate the optimal vehicle speed sequence. First, the initial speed sequence Z is generated. t ; Then, the vehicle speed sequence is iterated to generate the iterated speed sequence K t ; Then, the speed sequence Z before and after the iteration t and K t Perform Pareto non-dominated sorting and divide it into n non-dominated sets F(1)-F(n); finally, select k vehicle speed sequences with smaller cost functions in order as the initial speed sequence Z for the next iteration. t+1 .

2. The multi-objective speed planning method for intelligent vehicles oriented towards driving comfort and traffic efficiency according to claim 1 is characterized in that: The specific method of step one is as follows: 11) Based on the quarter-car suspension model and vehicle dynamics principles, the quarter-car suspension dynamics equation is obtained as follows: Where m s is the sprung mass, m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k t is the vertical equivalent stiffness of the tire; z s is the vertical displacement of the sprung mass; z u is the vertical displacement of the unsprung mass; z r is the road surface height; 12) Select the sprung mass displacement z s (t), unsprung mass displacement z u (t), sprung mass displacement velocity Unsprung mass displacement velocity As the state variable; select sprung mass acceleration Suspension travel z u (t)-z s (t), tire dynamic deflection z r (t)-z u (t) is the output variable of the system; then the road height z is selected r and its rate of change is the disturbance quantity of the system, and the quarter vehicle suspension dynamics equation (1) is rewritten as the system state space equation under continuous time, as shown below: in, Where x(t) is the state variable, is the derivative of the state variable; y(t) is the output variable; ω(t) is the disturbance variable; A is the suspension system matrix; C is the suspension output matrix; E is the state disturbance matrix; L is the output disturbance matrix; z s (t) is the displacement of the sprung mass; z u (t) is the unsprung mass displacement; is the sprung mass displacement velocity; is the displacement velocity of the unsprung mass; z r is the road surface height; m s is the sprung mass; m u is the unsprung mass; k s is the spring stiffness; c s is the suspension damping coefficient; c t is the vertical equivalent damping of the tire; k t is the vertical equivalent stiffness of the tire; 13) The continuous time matrix in Equation (2) is discretized using the zero-order hold discretization method, and the discretized system state space equation is obtained as follows: Where A is the suspension system matrix; E is the state interference matrix; C is the suspension output matrix; L is the output interference matrix; T s is the sampling period of the suspension system; is the discrete suspension system matrix; is the discrete state interference matrix; x(k) is the state variable of the kth period; y(k) is the output variable of the kth period; ω(k) is the interference variable of the kth period.

3. The multi-objective speed planning method for intelligent vehicles oriented towards driving comfort and traffic efficiency according to claim 1 is characterized in that: The specific method of step 2 is as follows: 21) The intelligent vehicle obtains forward-looking information, namely global road elevation information, through the Internet of Vehicles, and uses this global road elevation information to divide the entire road to be driven into bumpy sections and flat sections; 22) The entire road to be driven is divided into several speed adjustment positions; each speed adjustment position is provided with a speed adjustment interval of δV, starting from the lowest speed V min To the maximum speed V max Multiple candidate vehicle speeds.

4. The multi-objective speed planning method for intelligent vehicles oriented towards driving comfort and traffic efficiency according to claim 3 is characterized in that: The specific method for determining the vehicle speed adjustment position is as follows: (1) Assuming that the vehicle's speed is constant on the bumpy road section, each bumpy road section is considered as a speed adjustment position; (2) Set three intervals of l before and after each speed change event k The speed adjustment position is used to change the speed in time; (3) When a vehicle passes through a flat road section between two speed change events, it is stipulated that except for the speed adjustment position set in (2), no other speed adjustment position shall be set between the two speed change events to avoid frequent acceleration and deceleration of the vehicle.

5. The multi-objective speed planning method for intelligent vehicles oriented towards driving comfort and traffic efficiency according to claim 1 is characterized in that: The specific method of step 32) is as follows: 321) generating a vehicle speed sequence; Randomly generate k initial vehicle speed sequences Z t ; Each speed sequence is a length equal to the number of speed adjustment bits N s A one-dimensional array of , where the i-th element in the array represents the speed of the vehicle at the i-th speed adjustment position; 322) Iterate vehicle speed sequence; The speed sequence is divided into two parts during iteration: global exploration and local optimization; During global exploration, the global exploration sequence is used to find the optimal speed sequence by moving away from the global worst sequence, that is, the speed sequence with the highest cost. The iterative method of the global exploration sequence is as follows: Xs i (t+1)=Xs i (t)+α×k×Xs i (t-1)+β×△X(t) (10) △X(t)=|Xs i (t)-X w (t)| (12) Where t is the current iteration number; Xs i (t) is the i-th search sequence at the t-th iteration; X w (t) is the global worst speed sequence at the t-th iteration; α is the coefficient of variation; k is the inertia coefficient; β is the deflection coefficient; η is a random number between zero and one; λ is the probability of variation; During local optimization, a local search sequence, an overall search sequence, and a hybrid search sequence are used to perform detailed local optimization. The local optimal exploration sequence searches only in the local optimal region, while the overall search sequence and the hybrid exploration sequence search in the overall optimal region. The local optimal region and the overall optimal region change with the number of iterations: Where, Lb * is the lower bound of the optimal region; Ub * is the upper bound of the local optimal region; Lb a is the lower bound of the overall optimal region; Ub a is the upper bound of the overall optimal region; t is the current number of iterations; b * (t) is the local optimal speed sequence at the tth iteration; b a (t) is the global optimal vehicle speed sequence at the tth iteration; T max is the maximum number of iterations; V min is the minimum speed; V max is the maximum vehicle speed; The three sequences, local search sequence, global search sequence, and hybrid search sequence, are iterated in different ways. The iterative method of the local search sequence is as follows: Xl i (t+1)=b * (t)+f1×(Xl i (t)-Lb * )+f2×(Xl i (t)-Ub * ) (14) Where t is the current iteration number; Xl i (t) is the i-th local search sequence at the t-th iteration; b * (t) is the local optimal speed sequence at the tth iteration; f1 and f2 are 1×D random vectors; Lb * is the lower bound of the local optimal region; Ub * is the upper bound of the local optimal region; The overall search sequence is iterated as follows: Xg i (t+1)=Xg i (t)+c×(Xg i (t)-Lb a )+d×(Xg i (t)-Ub a ) (15) Where t is the current iteration number; Xg i (t) is the i-th overall search sequence at the t-th iteration; c is a 1×D vector that obeys the normal distribution; d is a 1×D random vector; Lb a is the lower bound of the overall optimal region; Ub a is the upper bound of the overall optimal region; The hybrid search sequence is iterated as follows: Xc i (t+1)=b a (t)+g×(|Xc i (t)-b * (t)|+|Xc i (t)-b a (t)|) (16) Where t is the current iteration number; Xc i (t) is the i-th mixed search sequence at the t-th iteration; b * (t) is the local optimal speed sequence at the tth iteration; b a (t) is the global optimal vehicle speed sequence at the tth iteration, g is a 1×D random vector; k initial vehicle speed sequences Z t After iterating through equations (10), (14), (15) and (16), k new vehicle speed sequences K are generated. t ; The 2k speed sequences before and after the iteration are recorded as X t , where the i-th vehicle speed sequence is X t (i); 323) Perform Pareto non-dominated sorting; From all the speed sequences before and after the iteration, k speed sequences with smaller cost functions are selected for the next iteration; therefore, all speed sequences need to be Pareto non-dominated sorted as follows: First, the vertical performance, longitudinal performance and traffic performance of all speed sequences are calculated by the established cost function; then, all sequences that are not dominated by other speed sequences are selected and stored in the non-dominated set F(1); among them, if a speed sequence X t (a) The performance of any defined planning objective is no better than that of another speed sequence X t (b) Difference, and X t (a) Outperform X on at least one planning objective t (b), then X t (b) Being X t (a) dominates; then, in all speed sequences, i.e., X t Remove the sequences included in F(1), and select the sequences that are not dominated by other speed sequences from the remaining speed sequences and store them in the non-dominated set F(2); and so on, until all speed sequences are sorted and stored in the non-dominated set F(1)-F(n); according to the definition and properties of domination, it can be seen that the speed sequence in the non-dominated set F(1) is the best, the speed sequence in F(2) is the second best, and the speed sequence in F(n) is the worst; 324) Select vehicle speed sequence; First, store F(i) in the initial velocity sequence Z of the next iteration in the order of the non-dominated set from small to large. t+1 Until F(p) is added, Z t+1 The number of sequences in F(p) exceeds the initial number of speed sequences k; then, the non-dominated set F(p+1)-F(n) is eliminated, and the sequences in F(p) are sorted using the concept of congestion; the congestion represents the density of the surrounding sequences of the selected speed sequence, and the sequence X is used. t (n) Surrounded by X t (n) A cuboid Q that contains no other sequences. t (n) volume representation; finally, select the speed sequence with the smallest congestion for the next iteration; if the non-dominated set F(1)-F(p-1) contains m speed sequences, then select the first km speed sequence with the smallest congestion in F(p), that is, F1(p), and store it in Z t+1 Make the number of sequences contained in it reach the initial speed sequence number k, and eliminate the remaining speed sequence F2(p), and then Z t+1 The vehicle speed sequence in is used for the next iteration.

Images