A PID Controller Architecture Inspired Enhancement to PSO Algorithm and its Application

Abstract

This thesis analyzes the connection between the Proportional Integral Derivative (PID) controller and the Particle Swarm Optimization (PSO) Algorithm and proposes two novel methods, PBSv2 and PAA, to enhance the performance of the PSO algorithm and its variants. PBSv2 modifies the approximation of the Derivative component of a previous work PBS and enables better diversity; The PAA method fully exploits the PID concept and introduces two adjustable parameters $(k_{i}, k_{d})$ to balance searching ability and convergence speed. Compared with the original PSO variants and PBS enhanced ones, the PSO variants assisted by the two newly proposed approaches achieve better optimization accuracy, which is demonstrated by applying our method on the CEC2014 Benchmark Suite. The result of these numerical experiments, which are included, shows that the PAA method significantly accelerates the convergence speed of the optimization process. This merit inspired us to employ PAA-PSO for auto-tuning of PID controller.

Using swarm-based algorithms to tune PID controllers is time-consuming since large amount of Fitness Evaluation (FE) operations are required and thus cannot be implemented in online tuning or real-time tuning for machines working in dynamic environments. Considering PAA-PSO possesses faster convergence speed and higher optimization accuracy compared with the original PSO, we propose that PAA-PSO is much more suitable for online tuning of PID controller. To validate the efficiency of this new application, we compare its tuning performances with 7 other state-of-art swarm-based optimization algorithms on 4 different kinds of systems with the requirement that the time usage is less than 1 second. The experimental results show that the new algorithm achieves the best trade-off between accuracy and time consumption and could be employed as a potential candidate for online or real-time tuning of PID controllers in applications such as driverless cars or robot manipulators, where fast decision making is critical.