 Original Article
 Open access
 Published:
EnergySaving and Punctuality Combined Velocity Planning for the AutonomousRail Rapid Tram with Enhanced Pseudospectral Method
Chinese Journal of Mechanical Engineering volume 36, Article number: 82 (2023)
Abstract
Autonomousrail rapid transit (ART) is a new mediumcapacity rapid transportation system with punctuality, comfort and convenience, but lowcost construction. Combined velocity planning is a critical approach to meet the requirements of energysaving and punctuality. An ART velocity preplanning and replanning strategy based on the combination of punctuality dynamic programming (PDP) and pseudospectral (PS) method is proposed in this paper. Firstly, the longitudinal dynamics model of ART is established by a multiparticle model. Secondly, the PDP algorithm with global optimal characteristics is adopted as the preplanning strategy. A model for determining the number of collocation points of the realtime PS method is proposed to improve the energysaving effect while ensuring computation efficiency. Then the enhanced PS method is utilized to design the velocity replanning strategy. Finally, simulations are conducted in the typical scenario with sloping roads, traffic lights, and intrusion of the pedestrian. The simulation results indicate that the ART with the proposed velocity trajectory optimization strategy can meet the punctuality requirement, and obtain better economy efficiency compared with the punctuality green light optimal speed advisory (PGLOSA).
1 Introduction
Urban commuters are confronted with the growing problem of traffic congestion [1]. Urban rail transit systems are very important for public transportation in large and mediumsized cities because of their advantages in transportation capacity, efficiency, punctuality, safety, and environmental friendliness [2, 3]. Due to the different factors such as population, economy, terrain, and traffic conditions in different cities, demands on the rail transit systems are also diverse. Thus, some innovative transportation products with new technologies and modes have emerged, such as autonomousrail rapid transit (ART) [4, 5].
Powered by electric motors, ART combines the features of urban rail transit and urban surface public transportation, which requires ART to meet the requirements of economy and punctuality simultaneously. After solving the problem of guaranteeing accurate tracking of a fixed line [6], various driving strategies can be adopted for ART to travel from the departure to the destination. In the operation of urban rail transit systems, the traction energy consumption of trains accounts for more than half of the total energy consumption. And the rest of energy consumption is mostly caused by infrastructure, such as ventilation systems, lighting systems, drainage systems, elevator equipment, etc. [7]. Therefore, the energyefficient driving method can achieve a better energysaving effect without changing the stations, lines, and operation diagram, which is an effective way to reduce the overall energy consumption and operation cost of the existing urban rail transit [8].
On the other hand, different from traditional rail transit, ART is used for surface public transportation, which is not subject to the physical constraints of the traditional rails and has a relatively open driving environment including traffic lights. Besides, the public traffic participant, such as pedestrians and general vehicles, may intrude into the virtual rail. To satisfy the demands of energysaving and punctuality in such a situation, a realtime optimization algorithm is needed for velocity planning. To sum up, the velocity profile of the ART between adjacent stations should be planned to ensure the performances in terms of economy and punctuality, by considering the constraints of vehicle dynamics, road conditions, traffic lights, and collision avoidance with other traffic participants.
The task of velocity planning is to calculate a reasonable and safe speed distribution on the future path, according to the states of the ART, the expected path, and environmental information. Some researches have been conducted on velocity planning to achieve the economy and punctuality of vehicles and trains. Pontryagin’s maximum principle (PMP) was used to solve the problem of optimal train speed profile planning for different driving conditions in Refs. [9, 10]. A realtime strategy for speed planning based on PMP and Lagrange multiplier technique was proposed in Ref. [11] to improve the performance of punctuality and energy efficiency of the highspeed train. However, the complex road environments for ART such as traffic lights, and other traffic participants must be considered in velocity planning, and the difficulty of solving the problem based on PMP will be greatly increased. Dynamic programming (DP) is a globally optimal algorithm that is usually utilized to design energyefficient driving strategies. The energy consumption and the arrival time of the train were taken as the optimization objectives to develop control strategies for the energysaving operation in Refs. [12, 13]. Haahr et al. [14] proposed a DP algorithm to optimize the economy of trains, taking into account the road altitude variation, the velocity, and the time constraints. Hellström et al. [15] developed a DPbased forwardlooking control strategy for heavyduty diesel trucks by comprehensively considering the energy consumption and travel time. The energydriving strategy based on DP is proposed in Refs. [16, 17], considering the road slope for fuel vehicles and electric vehicles, respectively. Asadi et al. [18] predicted the optimal speed trajectory to improve the driving economy of vehicles, considering the phase and time information of traffic lights. As a result, the vehicles passed within the green time, and the extra travel time and fuel consumption are reduced. Although the global optimum can be obtained by DP, the solving efficiency will be greatly reduced in the phenomenon of “dimension disaster” when the problem becomes complex and the optimization variables increase.
Nonlinear model predictive control (NMPC) was also applied to generate fuelsaving speed trajectories considering a short speed preview of the preceding vehicle [19]. However, NMPC is dissatisfactory in optimizing economy and comfort with its limitation of prediction horizon [20,21,22]. On the other hand, the pseudospectral (PS) method was also adopted in Ref. [23] to plan the optimal trajectory with consideration of the indicators such as running comfort, punctuality, and economy. Xiao et al. [24] proposed a speed optimization model combining energy management with operation control strategies for highspeed train systems and used the PS method for achieving energysaving and efficient operation. Xu et al. [25] proposed a method to compute the optimal speed profiles for an ecodriving system based on PS, which has the advantages of high accuracy and fast convergence. In addition, in Refs. [26, 27], the economic driving strategy for vehicles at signalized intersections with traffic lights was designed based on the PS method, which improved the energy efficiency of vehicles. In these studies, the optimality of the solution is closely related to the distribution of collocation points of the PS method. Hence, the same distribution of collocation points does not always appropriate for multiple traffic scenarios. In conclusion, none of the above methods can simultaneously satisfy the realtime and optimality of velocity trajectory planning. Therefore, there is an urgent need to study the realtime optimization algorithm of velocity planning to ensure that the driving objectives of ART such as safety, punctuality, and economy can be achieved under complex driving conditions.
In this paper, a realtime optimization strategy for ART velocity trajectory planning is proposed to achieve a punctual and energysaving cruise between two stations. The main contributions of this paper are as follows. Firstly, the global preplanning velocity trajectory is derived with PDP considering the dynamics model of ART, real road information, punctuality constraints, and traffic lights. Secondly, a model for determining the number of collocation points is established to balance the energysaving effect and computation efficiency for the proposed PSbased velocity planning method. Thirdly, the enhanced PSbased realtime method is applied to replan the velocity trajectory of the remaining route when the operation of ART is intruded on by other traffic participants.
In this paper, the main contents of the longitudinal velocity planning of ART are as follows. In Section 2, the longitudinal dynamics model of ART is established. In Section 3, the PDP and PSbased strategy of velocity trajectory planning for ART is proposed. In addition, the model to determine the number of collocation points for the PS method is established. In Section 4, the PreScan/Trucksim/Simulink cosimulation environment is designed, and the simulation results are analyzed to demonstrate the effectiveness of the proposed strategy. Section 5 is the conclusion of this paper.
2 Framework of EnergySaving and Punctuality Optimization Velocity Planning for ART
2.1 Framework of the Velocity Planning Strategy
The framework of the velocity planning strategy for ART is shown in Figure 1, including traffic information, the ART dynamics model, the velocity preplanning module, and the velocity replanning module. The traffic information from the real road maps provides emergencies, road slope, and signal phase and timing (SPaT) for ART, which are considered constraints in velocity planning. The objective function of the optimization problem reflecting the energy consumption is determined by the dynamics model of ART. The PDP and PS methods are adopted for velocity preplanning and replanning, respectively.
Under normal situations, ART follows the velocity trajectory preplanned by PDP to improve punctuality and economic performance. Nevertheless, ART switches to follow the velocity trajectory replanned by the PS method when the route of ART is intruded on by pedestrians, lowspeed vehicles, or other traffic participants. Considering the signalized intersection, the whole journey is divided into several segments, and the velocity trajectory is planned sequentially.
2.2 Longitudinal Dynamics Model of ART
The singleparticle model [28,29,30] and the multiparticle train model [31, 32] can be used to model the dynamics for train operation control. With the multiparticle model, the forces of each carriage can be analyzed separately, and the slope change of the road can be considered [33, 34], thus the control accuracy can be improved.
In this paper, the multiparticle dynamics model is selected to compensate for the shortage of singleparticle models in simple modeling and incomplete analysis. ART is considered a “particle chain”, and each carriage is treated as a particle. Then, the forces of each carriage can be analyzed separately, which makes the dynamics model more accurate. On the other hand, the gradient resistance based on the multiparticle dynamics model changes gradually at different slope sections, therefore the planned velocity of the ART can be smoothed.
To establish a practical and computable dynamics model for the ART, it is necessary to make the following assumptions: The forces generated by the articulation are not considered, and the mass of each carriage is identical. To ensure the continuity of velocity, the longitudinal acceleration of each carriage is regarded as identical. In addition, the lateral motion and the vertical vibration of ART are ignored during the longitudinal velocity planning.
Based on the assumptions, the multiparticle dynamics model of the ART is demonstrated as the schematic diagram in Figure 2, where F_{t} is the driving force produced by the six identical motors. The resistance is composed of air resistance F_{w}, rolling resistance F_{f,i} and gradient resistance F_{g,i}. G_{i} and θ_{i} denote the gravity and slope angle of carriage i (i = 1, 2, 3), respectively. The interaction force of adjacent carriages is equal, where F_{1,2} = F_{2,1} and F_{2,3} = F_{3,2}. According to the force analysis, the longitudinal dynamics model of ART for velocity planning is shown as [35]:
where C_{d} denotes the air resistance coefficient; A_{F} denotes the front area of the first carriage; v denotes the longitudinal velocity of ART; f_{r} is the rolling resistance coefficient; M denotes the total mass of ART; g denotes gravitational acceleration; T_{m} denotes the torque of one motor; R_{t} denotes the radius of tire.
The change in battery power reflects the energy consumption of each motor. The power model of each motor in driving and breaking condition is calculated by
where P_{m} denotes the power of motor; ω_{m} denotes the rotational speed. η_{m} and η_{e} denote the output efficiency and generation efficiency of the motor, respectively. In some researches about the ecological driving of electric vehicles, the battery model is simplified by setting the electrochemical conversion efficiency as a constant or ignoring electrochemical reaction [36, 37]. Therefore, the battery power equals P_{m}, and the energy consumption can be obtained as
where ∆Q represents the electricity consumption.
3 Velocity Trajectory Optimization Strategy Combined with Preplanning and Replanning Method
3.1 Preplanning of the Global Optimal Velocity
DP is an optimization algorithm based on the Bellman optimization principle, which is an effective mathematical method to solve the optimal control problem (OCP). With DP, the complex multivariable optimization problems are solved in two steps. First, the optimal solution of each stage is obtained, and second, the algorithm of the recursive call is used to obtain the optimal solution for complex problems. DP divides the optimal solution into discrete stages and transforms the whole optimization problem into N subproblems. Then, the state variables and the control decisions of each subproblem are described.
The longitudinal speed planning problem in this paper is an Nlevel decisionmaking process, in which the whole operation process of ART can be divided into N stages. Then the optimal control torque of each stage can be solved, so that the speed can be transferred continuously between each stage. For each substage, the electricity consumption and driving time of the initial state in the transfer process should be calculated accordingly. In summary, the optimal control variables are calculated by reverse solving, and then the optimal control set u_{k} (k = 0, 1, …, N−1) and the state set v_{k} are found by forward optimization.
The electricity consumption and travel time of the train between adjacent stations are selected as objective functions when constructing the optimal control problem of ART. In order to eliminate the magnitude difference between time and energy index, a weight coefficient α is set for the travel time. Therefore, the OCP of the system can be obtained as Eq. (4), where T_{min} and T_{max} represent the torque limits of the motor; v_{min} and v_{max} represent the limits of the cruising velocity; the state transition function f is derived from Eq. (1). t_{k} is the travel time of stage k. j_{max} denotes the limitation of jerk. v_{nmax} and v_{nmax} denote the minimum and maximum velocity when ART arrives at the next station. t_{arrival} denotes the arrival time. α denotes the weight coefficient of time in the objective function.
When the value of α is small, the weight of travel time in the objective function is small, resulting in longer travel time accordingly, and vice versa. The dichotomy method can be adopted to adjust the weight coefficient of time optimization to meet the requirement of punctuality, and thus realizes the punctuality dynamic programming (PDP). The procedures of the proposed PDP based on dichotomy method are summarized in Figure 3.
3.2 Replanning of Velocity Based on PS Method
During the operation of ART, unexpected emergencies such as pedestrian intrusion and traffic jams are inevitable. Therefore, a realtime velocity planning method is needed to replan the longitudinal velocity trajectory. PDP is a global optimization algorithm, but it cannot be adopted as a realtime strategy for replanning because of the low computational efficiency and poor realtime performance.
Remark 1: If the ART cruise is not disturbed by emergencies, ART will follow the preplanning velocity all the time to achieve the most energysaving cruise.
The OCP for velocity replanning can be solved with the PS method, which belongs to a special kind of direct method. The PS method has the advantages of fast convergence, large radius of convergence, high precision of solution, and low sensitivity of initial value. Its numerical solution has been proved theoretically to satisfy optimality [38]. Based on the PS method, the optimal control model of the longitudinal velocity planning problem is reconstructed and transformed into nonlinear programming (NLP), which can be solved efficiently.
The objective function is designed by the energy consumption function. In addition, the constraints of OCP are set to meet the requirements of ART in safety and punctuality. The constrains include velocity range v_{min} ≤ v_{k} ≤ v_{max}, torque range T_{min} ≤ T_{k} ≤ T_{max}, and the target cruising time t_{f} = t_{target}. To sum up, the OCP for velocity replanning of ART is defined as follows:
where distance s(t) and velocity v(t) are state variables; t_{0} denotes the starting time; t_{f} denotes the terminal time of OCP. The torque of the motor T_{m}(t) is taken as control variable. S_{0} represents the beginning position, and S_{f} represents the terminal position. v_{0} represents the initial velocity of ART.
Different types of PS methods have different advantages. The OCP of ART contains nonfree boundary conditions, such as the initial velocity, initial position, arrival time, and limits of the cruising velocity of ART. The Legendre PS method is adopted in this paper because of its better performance for solving the problem of ART with nonfree boundary conditions, while the Gauss PS method and Radau PS method may not converge in these conditions [25].
The state and control variables are scattered on Legendre Gauss Lobatto (LGL) collocation points, and approximated by Lagrange interpolating polynomials. The differential dynamic equation constraints are transformed into algebraic constraints on the interpolating points. The integral part in the objective function is calculated by the LGL integral. The detailed transformation process is described as follows.
3.2.1 Approximation of State and Control Variables
τ_{k} (k = 0, 1, … N) denotes the LGL collocation point. It is the root of polynomial \((1  \tau^{2} )\dot{P}_{N} (\tau )\) as well. P_{N} is the Legendre polynomial of order N, which is described as
The collocation points τ_{k} ∈ [−1, 1]. Therefore, the time domain t_{k} needs to be transformed into
where t_{f} equals to the target cruising time t_{target}. Then the state variable x(τ_{k}) and control variable u(τ_{k}) are approximated as:
where X_{i} and U_{i} are the interpolating points of state variables and control variables, respectively. L_{i} (i = 0, 1, …, N) denotes the Lagrange interpolating polynomial of order N, with the expression:
3.2.2 Transformation of Differential Dynamic Equations
By differentiating the state variables in Eq. (8), the dynamic equations can be transformed into
Specifically, the dynamic equations are expressed as:
where S, V, and T are the interpolating points of position, velocity and torque. D_{i}(τ_{k}) is the differential matrix [39].
3.2.3 Transformation of Objective Function
The integral part of objective function in the OCP can be approximated by the LGL integral, which is defined as
where ω_{k} is the integral coefficient. It is expressed as
After transformation, the mathematical model of the NLP is obtained as:
where S, V, and T are regarded as the variables to be optimized. For the N+1 collocation points obtained by the PS method, 2×(N+1) interpolation points for the state variables and (N+1) interpolation points for the control variables can be obtained discretely. Therefore, there are 3×(N+1) parameters to be optimized in this NLP.
3.3 Model to Determine the Number of Collocation Points
The number of collocation points of the PS method will affect the approximation accuracy of the discrete model to the original model and thus affect the economy of replanning velocity. However, with the number of collocation points increasing, the computation efficiency will decline. Energy optimality and computation efficiency cannot be guaranteed at the same time. To find the optimal number, the preplanning results of PDP are taken as an optimal reference to select the optimal number of collocation points for the replanning method based on PS. Firstly, the PDP approach is adopted to plan the velocity based on the road intercepted from the real road map. Secondly, the PS methods with different numbers of collocation points are taken to replan the velocity under the same conditions, and the results are compared with those based on the PDP method.
To determine the appropriate number of collocation points, both the computation time of algorithms and the planning results are considered in the model. For the planning results, the rootmeansquare error (RMSE) is used to describe the differences between the velocity trajectories based on the PS method and those based on PDP. By comparing the sums of the normalized value of the computation time and the RMSE for the different numbers of collocation points, the optimal number of collocation points can be found.
The RMSE is calculated by
where N is the number of sampling points, which equals to the number of discrete points based on PDP; v_{PS,k} and v_{PDP,k} are values sampled from the velocity curves planned by the PS and PDP methods, respectively.
For each collocation number k_{i} in set K = {k_{1}, k_{2}, …, k_{n}}, the computation time of PS method and RMSE are normalized by
where t_{i} is the computation time of algorithm with k_{i} collocation points, and k_{i} ∈{20, 30, 40, …, 200}. t_{max} and t_{min} are the maximum and minimum values of computation time; Γ_{i} is the normalized value of t_{i}. Ψ_{i} is the normalized value of E_{i}. E_{max} and E_{min} are the maximum and minimum values of RMSE. The final evaluation value is calculated by summing the normalized values of the computation time and the RMSE, and then the optimal number k_{m} of collocation points is chosen by Eqs. (20) and (21):
3.4 Target Cruising Time at Signalized Intersection in Preplanning and Replanning Program
Due to the heavy load and longitudinal size of ART, the frequent start and stop when passing the signalized intersections will not only increases the travel time, but also increases the energy consumption, which greatly affects the punctuality and economy of ART. Therefore, when passing the signalized intersections, ART should avoid stopping and waiting for red lights. In order to facilitate the solution of the problem, the assumptions are made: 1) The communication delay during the information interaction between the road equipment and the onboard equipment is ignored; 2) The ART can pass the signalized intersections without being affected by other traffic participants.
Based on the assumptions above, the intersections can be regarded as destinations and the whole route is divided into several segments by the intersections. As long as the target cruising time of each segment is in the green light window, ART can pass the intersection without stopping. In order to reduce the fluctuation of velocity, the average velocity of each segment is planned to be close to the average velocity of the whole route. Meanwhile, the target cruising time should stay away from the redlight window. As a result, the target cruising time is derived from the expected cruising time and redlight window. To find the appropriate target cruising time in each segment, a potential fieldbased model is designed to describe the effects of expected cruising time and red light window, and the time with the lowest potential is selected as the target cruising time.
The expected cruising time during Segment i is calculated by the length of the segment and the expected average velocity of the whole route:
where l_{seg,k} is the length of segment k; v_{avg} and t_{total} are the expected average velocity and required arrival time of the whole route between two adjacent stations, respectively. t_{seg,i} is the expected cruising time during segment i. The target cruising time t_{target,i} needs to approach the expected cruising time and stay away from the red light window. As shown in Figure 4, by comparing the summation of expected cruising time potential and red light potential, the time with the lowest potential is selected as the target cruising time t_{target,i}. When t_{seg,i} is in the n_{th} cycle of the traffic light, the potential field function is defined by a Gausstype function, and the potential filed based model is given as:
where U is the potential of time. t_{r} and t_{g} are the time of red light and green light, respectively. u_{e} and u_{r} are the potentials generated by expected cruising time and red light window, respectively. σ_{e} and σ_{r} are the convergence coefficients of the potential u_{e} and u_{r}, respectively. u_{r} is related to the starting time t_{0,i}, which aims to avoid the too short target cruising time in replanning program. The time interval is set to be the next cycle if t_{target,i} equals to nt_{cycle}.
4 Simulation Results and Analysis
Simulations are carried out to verify the effect of the proposed velocity planning method in the cosimulation environment including PreScan, TruckSim, and MATLAB/ Simulink. The driving environmental scenes of the vehicle are established in PreScan, in which the virtual sensors (such as camera, radar, GPS, etc.) mounted on the vehicle send signals to Simulink in realtime. The perception information is processed in MATLAB/Simulink through the algorithms to obtain the environmental input for controller design [40]. The control output for the actuator is calculated by the control algorithm and sent to the vehicle dynamics system in TruckSim. The realtime motion states of the dynamics system are sent to PreScan for updating the virtual scene.
The parameters of ART used in simulations are demonstrated in Table 1. In this section, the proposed ART velocity trajectory optimization algorithm is verified by simulation in the scenario with traffic lights, varying sloping and invasion of the pedestrian.
4.1 Selection of the Collocation Points Number
According to the road data of the operating ART, the average station spacing is no more than 2 km. For a 15 km route collected from real traffic scenarios, which is shown in Figure 5, three representative road conditions of this route are analyzed. Road 1 is simplified from the mostly flat roads. Road 2 represents the less fluctuating roads, and Road 3 represents the most fluctuating roads. The results for collocation points derived from the three representative road conditions can be applied to the whole route.
The number of collocation points is set to be distributed in K = {20, 30, 40, …, 200}. The influences of collocation points on the realtime performance and planning results of the PS method are shown in Figure 6.
It can be seen in Figure 6(a)–(c) that when the number of the collocation points increases, the similarity of the planned velocity curves with the two methods is improved, and the computation time of the program increases. Figure 6(d) shows that the appropriate number of collocation points for a distance of 2000 m is between 60 and 100. To be concluded, the number of collocation points is chosen by the ratio of 80 points for a distance of 2000 m. This conclusion is applied to the following simulations in Section 4.
In order to verify the economy and punctuality of the proposed method, the velocity trajectories planned by the PDP method, the PS method, and the regular punctuality velocity (RPV) planning are compared. The target velocity of RPV is calculated by the dichotomy in the velocity range [v_{min}, v_{max}]. The target cruising time t_{target} is set as 290 s. The initial velocity and the terminal velocity of these three trajectories are 15 km/h. The target cruising time error t_{err} in the dichotomy program is set as 0.5 s. The parameters of travel time, energy consumption and planning time, are compared in Table 2. Figure 7 shows the velocity planning results in Road 3.
It can be seen from Figure 7 that the velocity curves obtained by the PS method and the PDP algorithm have similar trends. The PDPbased strategy can meet punctuality when α equals −3477. The velocity trajectory planned by the PS can also meet the requirements of the punctual and safe arrival of ART. Compared with the RPV, the energy consumption of the ART with the PDP method and the PS method reduces by 11.56% and 11.39%, respectively. The PS method can provide an almost equivalent energysaving effect for the ART compared with the PDP method. From the perspective of algorithmic computation time, the PDP algorithm takes 37.6 s for the whole process. The PS method takes only 0.23 s, saving quite a lot of computation time compared with the PDP algorithm, which greatly improves planning efficiency. It is obvious that the realtime performance of the PS method is better than the PDP.
4.2 Cosimulation in Complex Traffic Scenario with Multi Traffic Lights and Random Obstacles
Generally speaking, the number of signalized intersections is less than 2 between adjacent stations. Road 3 in Section 3 is chosen to establish the scenario in the simulation. As shown in Figure 8, two traffic lights are located at 800 m and 1600 m, respectively, and the whole route is divided into 3 segments, denoted by Segment 1 to Segment 3. The timetable of the traffic light within a cycle is shown in Table 3. The altitude information and traffic light signals can be collected in advance.
The velocity range is set within [5, 40] km/h. Considering the comfort of passengers and punctuality of ART, the terminal velocity interval, maximum jerk, and required arrival time are set as v_{ter} = [15, 20] km/h, j_{max} = 0.5 m/s^{3}, and t_{total} = 290 ± 5 s, respectively.
The virtual rail of ART is constructed on the existing transport network, and the rail might be invaded by pedestrians or other lowspeed vehicles, which are considered obstacles [41]. These obstacles can make ART deviate from the preplanned velocity trajectory based on PDP. To ensure the punctuality and economy of ART, the remaining routes are replanned with the PS method. To simulate an emergency caused by the obstacles, a pedestrian is assumed to appear at 200 m.
During Segment 1 of the route, t_{seg,1} is 116 s, which is out of the green light window. In the preplanning program, ART starts the cruise at t_{0,1} = 0. ART can detect the pedestrian and take the braking strategy immediately when encountering the pedestrian at a distance of 200 m. Then the remaining route is replanned based on the PS method after the pedestrian leaves the virtual rail at 48.8 s. In the replanning program, ART starts the cruise at t_{0,1} = 48.8 s. By calculating the total potential with Eq. (23), the target cruising time of Segment 1 in the preplanning program and replanning program can be obtained as 135 s when t_{0,1} = 0, and 140 s when t_{0,1} = 48.8 s, respectively. The target cruising time of Segment 2 can be obtained as t_{target,2} = t_{seg,2} = 232 s, because t_{seg,2} is located in the green light window. The target cruising time of Segment 3 t_{target,3} = 290 s.
The proposed algorithm of velocity planning is verified in the cosimulation environment including PreScan, TruckSim, and Matlab/Simulink. In this Section, the simulation takes the traffic lights into consideration. To make the comparison more convincing, the variate, i.e. arrival time of different strategies should be the same. The commonly used speed planning method, i.e. green light optimal speed advisory (GLOSA), can ensure that ART drives into the intersection on a green signal [42, 43]. However, punctuality cannot be guaranteed if GLOSA only is adopted. A punctuality GLOSA (PGLOSA) is designed as a comparison, where the target velocity is calculated in two steps. Step 1: the target velocity range [v_{target_min}, v_{target_max}] is obtained by GLOSA in Eq. (24). Step 2: the target velocity is calculated by the dichotomy in velocity range [v_{target_min}, v_{target_max}] to meet the punctuality requirement.
where L is the distance to the traffic light. The simulation results and comparisons are as follows.
Figure 9(a) shows that the ART reaches the first signalized intersection at 153.8 s in the red light window if ART continues to track the preplanning velocity without replanning after the pedestrian leaves. On the other hand, in Figure 9(b)–(c), ART passes the first signalized intersection at 143.5 s and 144.5 s with the replanning velocity trajectory based on PGLOSA and PS method, respectively, which are both in the green light windows.
As shown in Figure 10(a), the replanning velocity based on the PS method varies with the change of slope. The comparison of the output torque of the two replanning strategies is shown in Figure 10(b). It can be seen from the figure that the motor output torque of the PS methodbased replanning strategy is smoother than that of the PGLOSAbased replanning strategy. Furthermore, Figure 10(c) demonstrates that the energy consumption of the PS methodbased replanning strategy is lower than the PGLOSAbased replanning strategy, which is 6.77 kWh for the PS method and 7.46 kWh for the PGLOSA method.
Simulation results demonstrate that, after replanning the velocity based on the PS method, the ART arrives at the next station on time because the ART does not stop and restart while passing through the intersection. As shown in Table 4, the average velocities of the ART with two replanning strategies are 25.0 km/h and 24.5 km/h, and the corresponding arrival times are 290.9 s and 293.4 s, respectively, both meeting the requirements of punctuality. Moreover, the energy consumption of the PSbased velocity replanning strategy has reduced by 9.19% compared with the PGLOSAbased strategy. To better understand how energy saving is achieved, the MAPefficiency profile is given in Figure 11(a), where the yellow triangle symbols and blue cross symbols represent the working points when ART follows the target velocity replanned by PS method and PGLOSA, respectively. The frequencies of motor efficiency in different ranges during the simulation are counted in Figure 11(b). The statistical histogram in Figure 11(b) suggests that the replanning speed based on the PS method makes the motor work in the highefficiency region as much as possible, which leads to better energy efficiency.
In conclusion, ART can consume less energy and reach the next station on time by combining the PDP and PSbased velocity planning strategy.
5 Conclusions
In this paper, the PDP algorithm and PS method are combined to preplan and replan the velocity of ART considering its economy and punctuality. The main conclusions are summarized as follows.

(1)
Compared with the velocity planning method based on PGLOSA, results of PDPbased planning show lower energy consumption but a higher computation burden. By applying PDP to determine the optimal number of the collocation points for the realtime PS method and optimize the energysaving effect and computation efficiency, the enhanced PS method reaches a similar planning result compared with PDP but shows far better realtime performance.

(2)
The effectiveness of the proposed strategy for planning velocity trajectory combined with these two algorithms is verified in a typical traffic scenario including sloping roads, SPaT and invasion of the pedestrian. The result shows that the economic efficiency of ART is improved with a 9.19% reduction in energy consumption compared with PGLOSA and the punctuality is also adequate.
In summary, the proposed strategy has good application prospects for public transport under typical traffic scenarios. For future work, the proposed velocity preplanning and replanning strategies will be verified by real vehicle experiments on ART.
References
M Saberi, H Hamedmoghadam, M Ashfaq, et al. A simple contagion process describes spreading of traffic jams in urban networks. Nature Communications, 2020, 11: 1616.
Z Gao, L Yang. Energysaving operation approaches for urban rail transit systems. Frontiers of Engineering Management, 2019, 6(2): 139151.
W Li, Q Peng, C Wen, et al. Integrated optimization on energy saving and quality of service of urban rail transit system. Journal of Advanced Transportation, 2020: 122.
L Xiao, S Y Guo, X W Yuan, et al. Analysis of vision based automatic steering control for an articulated all wheel steered vehicle. 2016 IEEE Vehicle Power and Propulsion Conference (VPPC), 2016: 15.
X Yuan, R Huang, X Zhang, et al. Development and validation of an automatic steering control system for rubbertire transit revenue service. 2019 IEEE Vehicle Power and Propulsion Conference (VPPC), 2019: 16.
A Sorniotti, P Barber, S De Pinto. Path tracking for automated driving: a tutorial on control system formulations and ongoing research. Automated Driving, 2016: 71140.
X Yang, X Li, B Ning, et al. A survey on energyefficient train operation for urban rail transit. IEEE Transactions on Intelligent Transportation Systems, 2016, 17(1): 213.
S Su, T Tang, X Li. Driving strategy optimization for trains in subway systems. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 2016, 232(2): 369383.
K Bao, S Lu, F Xue, et al. Optimization for train speed trajectory based on Pontryagin's maximum principle. 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), 2017: 16.
S Uebel, N Murgovski, C Tempelhahn et al. Optimal energy management and velocity control of hybrid electric vehicles. IEEE Transactions on Vehicular Technology, 2018, 67(1): 327337.
Z Xiao, Q Wang, P Sun, et al. Realtime energyefficient driver advisory system for highspeed trains. IEEE Transactions on Transportation Electrification, 2021, 7(4): 31633172.
S Su, X Wang, Y Cao, et al. An energyefficient train operation approach by integrating the metro timetabling and ecodriving. IEEE Transactions on Intelligent Transportation Systems, 2020, 21(10): 42524268.
Y Huang, L Yang, T Tang, et al. Train speed profile optimization with onboard energy storage devices: A dynamic programming based approach. Computers & Industrial Engineering, 2018, 126: 149164.
J Haahr, D Pisinger, M Sabbaghian. A dynamic programming approach for optimizing train speed profiles with speed restrictions and passage points. Transportation Research Part B: Methodological, 2017, 99: 167182.
E Hellström, M Ivarsson, J Åslund, et al. Lookahead control for heavy trucks to minimize trip time and fuel consumption. Control Engineering Practice, 2009, 17(2): 245254.
M Zhou, H Jin, F Ding. Minimizing vehicle fuel consumption on hilly roads based on dynamic programming. Advances in Mechanical Engineering, 2017, 9(5): 168781401769411.
W Zhuang, L Qu, S Xu, et al. Integrated energyoriented cruising control of electric vehicle on highway with varying slopes considering battery aging. Science China Technological Sciences, 2019, 63(1): 155165.
B Asadi, A Vahidi. Predictive cruise control: utilizing upcoming traffic signal information for improving fuel economy and reducing trip time. IEEE Transactions on Control Systems Technology, 2011, 19(3): 707714.
M F Ozkan, Y Ma. A predictive control design with speed previewing information for vehicle fuel efficiency improvement. 2020 American Control Conference (ACC), 2020: 23122317.
S Xu, H Peng. Design and comparison of fuelsaving speed planning algorithms for automated vehicles. IEEE Access, 2018, 6: 90709080.
S E Li, Q Guo, S Xu, et al. Performance enhanced predictive control for adaptive cruise control system considering road elevation information. IEEE Transactions on Intelligent Vehicles, 2017, 2(3): 150–160.
Y Yan, J Wang, K Zhang, et al. Driver’s individual risk perceptionbased trajectory planning: A humanlike method. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 2041320428.
Y Wang, B De Schutter, T van den Boom, et al. Optimal trajectory planning for trains – A pseudospectral method and a mixed integer linear programming approach. Transportation Research Part C: Emerging Technologies, 2013, 29: 97114.
Z Xiao, Q Wang, P Sun, et al. Modeling and energyoptimal control for highspeed trains. IEEE Transactions on Transportation Electrification, 2020, 6(2): 797807.
S Xu, K Deng, S E Li, et al. Legendre pseudospectral computation of optimal speed profiles for vehicle ecodriving system. 2014 IEEE Intelligent Vehicles Symposium Proceedings, 2014: 11031108.
S E Li, S Xu, X Huang, et al. Ecodeparture of connected vehicles with V2X communication at signalized intersections. IEEE Transactions on Vehicular Technology, 2015, 64(12): 54395449.
Q Lin, S Li, X Du, et al. Minimize the fuel consumption of connected vehicles between two redsignalized intersections in urban traffic. IEEE Transactions on Vehicular Technology, 2018, 67(10): 90609072.
Y Liu, K Fan, Q Ouyang. Intelligent traction control method based on model predictive fuzzy PID control and online optimization for permanent magnetic maglev trains. IEEE Access, 2021, 9: 2903229046.
Y Cao, Z Wang, F Liu, et al. Bioinspired speed curve optimization and sliding mode tracking control for subway trains. IEEE Transactions on Vehicular Technology, 2019, 68(7): 63316342.
L Wang, X Wang, K Liu, et al. Multiobjective hybrid optimization algorithm using a comprehensive learning strategy for automatic train operation. Energies, 2019, 12(10): 1882.
H Tang, Q Wang, X Feng. Robust stochastic control for highspeed trains with nonlinearity, parametric uncertainty, and multiple timevarying delays. IEEE Transactions on Intelligent Transportation Systems, 2018, 19(4): 10271037.
Y Chen, D Huang, T Huang, et al. Tracking control via iterative learning for highspeed trains with distributed input constraints. IEEE Access, 2019, 7: 8459184601.
X Yan, B Cai, B Ning, et al. Moving horizon optimization of dynamic trajectory planning for highspeed train operation. IEEE Transactions on Intelligent Transportation Systems, 2016, 17(5): 12581270.
W Wang, X Zeng, T Shen, et al. Energyefficient speed profile optimization for urban rail transit with considerations on train length. 2018 21st International Conference on Intelligent Transportation Systems (ITSC), 2018: 15851591.
M Wu, J Wang, Y Yan, et al. Velocity trajectory planning of the autonomousrail rapid tram considering terrain and traffic lights. 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC), 2020: 16.
J Han, A Sciarretta, L L Ojeda et al. Safe and ecodriving control for connected and automated electric vehicles using analytical stateconstrained optimal solution. IEEE Transactions on Intelligent Vehicles, 2018, 3(2): 163172.
A Sciarretta, G De Nunzio, L L Ojeda. Optimal ecodriving control: energyefficient driving of road vehicles as an optimal control problem. IEEE Control Systems Magazine, 2015, 35(5): 7190.
G Elnagar, M A Kazemi, M Razzaghi. The pseudospectral Legendre method for discretizing optimal control problems. IEEE Transactions on Automatic Control, 1995, 40(10): 17931796.
F Fahroo, I Ross. Advances in pseudospectral methods for optimal control. AIAA Guidance, Navigation and Control Conference and Exhibit, 2008.
J Wang, Y Yan, K Zhang, et al, Path planning on large curvature roads using drivervehicleroad system based on the kinematic vehicle model, IEEE Transactions on Vehicular Technology, 2021, 71(1): 311325.
Y Yan, J Wang, Y Wang, et al. A cooperative trajectory planning system based on the passengers' individual preferences of aggressiveness, IEEE Transactions on Vehicular Technology, 2023, 72(1): 395406.
X Tang, Z Duan, X Hu, et al. Improving ride comfort and fuel economy of connected hybrid electric vehicles based on traffic signals and real road information. IEEE Transactions on Vehicular Technology, 2021, 70(4): 31013112.
H Dong, W Zhuang, B Chen, et al. Enhanced ecoapproach control of connected electric vehicles at signalized intersection with queue discharge prediction. IEEE Transactions on Vehicular Technology, 2021, 70(6): 54575469.
Acknowledgements
Not applicable.
Funding
Supported by National Natural Science Foundation of China (Grant Nos. 52072073 and 52025121) and National Key R&D Program of China (Grant No. 2018YFB1201602).
Author information
Authors and Affiliations
Contributions
JW was in charge of the whole trial, and carried out the research, and wrote the manuscript. DH was in charge of data analysis and wrote the manuscript. YY reviewed the manuscript. NL and NS assisted with formula analyses. GY provided theoretical guidance and guided journal submissions. All authors read and approved the final manuscript.
Authors’ Information
Jinxiang Wang, born in 1979, is currently an Associate Professor with School of Mechanical Engineering, Southeast University, China. He received the B.S. degree in mechanical engineering and automation and the PhD degree in vehicle engineering from Southeast University, China, in 2002 and 2010, respectively. From 2014 to 2015, he was a Visiting Research Scholar with Department of Mechanical and Aerospace Engineering, The Ohio State University, Columbus, OH, USA. His research interests include vehicle dynamics and control, assisteddriving system, and control on autonomous vehicles.
Dongming Han, born in 1998, is currently working toward M.S. degree at School of Mechanical Engineering, Southeast University, China. He received his B.S. degree at School of Mechanical Engineering, Southeast University, China, in 2020.
Yongjun Yan, born in 1995, is currently working toward the PhD degree in vehicle engineering from Southeast University, China. He received his B.S. degree in vehicle engineering from Nanjing Agricultural University, China, in 2017, and his M.S. degree in vehicle engineering from Southeast University, China, in 2020. His research interests include vehicle dynamics and control, automotive active safety control, intelligent vehicle decisionmaking and control, personalized driving.
Neng Liu, born in 1998, is currently working toward M.S. degree at School of Mechanical Engineering, Southeast University, China. He received his B.S. degree at School of Mechanical Engineering, Southeast University, China, in 2021.
Ning Sun, born in 1979, is currently a lecturer at College of Automobile and Traffic Engineering, Nanjing Forestry University, China. She received the PhD degree in vehicle engineering from Southeast University, China, in 2011.
Guodong Yin, born in 1976, received the Ph.D. degree in vehicle engineering from Southeast University, China, in 2007. From 2011 to 2012, he was a Visiting Research Scholar with Department of Mechanical and Aerospace Engineering, Ohio State University, Columbus, OH, USA. He is currently a Professor with School of Mechanical Engineering, Southeast University, China. His current research interests include vehicle dynamics and control, connected vehicles, and multiagent control.
Corresponding author
Ethics declarations
Competing Interests
The authors declare no competing financial interests.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Wang, J., Han, D., Yan, Y. et al. EnergySaving and Punctuality Combined Velocity Planning for the AutonomousRail Rapid Tram with Enhanced Pseudospectral Method. Chin. J. Mech. Eng. 36, 82 (2023). https://doi.org/10.1186/s10033023008919
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1186/s10033023008919