Design and Experimental Validation of Longitudinal Controller of Connected Vehicles using Model Predictive Control

Abstract

In this thesis, a vehicle longitudinal control algorithm based on model predictive control (MPC) is applied to compute the desired relative acceleration of the following vehicle in leader-follower systems. Kinematic equations are used to describe the dynamic relationship between the leading and following vehicles. Compared to the conventional model predictive control (CMPC), the control horizon is expressed using Laguerre functions. This makes the optimization problem easier to solve and available to be tuned. Appropriate parameters are investigated by comparing the different approximation results under different decay factors. The design of MPC based on Laguerre functions (LMPC) enables the system to be adjustable through the selection of the decay factors depending on the characteristics such as response time and overshoot of the closed-loop system. The effectiveness of the design approach was demonstrated using simulations and experiments. Control performance of the closed-loop system was investigated by selecting different parameters including the states weighting matrix, the input weighting matrix, and Laguerre coefficients. With constraints on the control variables and the difference of the control variables, the following vehicle can track the leading vehicle with a specific distance and at the same speed in the simulation. Experiments which illustrate the performance of the control system were performed on an experimental platform used by Federal Highway Administration (FHWA), followed by the experimental data showing the following vehicle can track the leading vehicle with a specific distance and at the same speed. However, there is overshoot of the distance and the relative speed is not zero. The reasons of the poor performance of the control system were explored, which include the absence of the acceleration of the leading vehicle, large constraints on the difference of the desired acceleration. Solutions such as decreasing the constraints on the incremental variation and enlarging the input weighting matrix are discussed.