Neural Network Predictive Control for Mobile Robot Using PSO with Controllable Random Exploration Velocity

Being a kind of intelligent method, Neural Network Predictive Control (NNPC) has been used widely to control nonlinear systems. However if traditional gradient decent algorithm (GDA) is employed to generate control signals, the computational cost is relatively too large, so that NNPC seems not to be acceptable for systems with rapid dynamics. To make NNPC be more efficient to general control signals for rapid dynamics, such as mobile robots, this paper proposes an improved optimization technique, particle swarm optimization with controllable random exploration velocity (PSOCREV), to take place of GDA in NNPC. Moreover from theoretical analysis, it is observed that PSO-CREV has higher search ability than the conventional PSO. Therefore for one time of optimization, PSOCREV needs small iterations than GDA, and less population size than conventional PSO. Hence the computational cost of NNPC is reduced by using PSO-CREV, therefore NNPC using PSO-CREV is more feasible for control of rapid processes. NNPC for trajectory tracking of mobile robots is chosen as a test to compare performance of PSOCREV with other algorithms to show its advantages, especially in computational time.