INVERTED PENDULUM CONTROL USING NEURAL NETWORKS

Abstract

The inverted pendulum has been used as a useful laboratory idealization of unstable mechanical systems. As it is a highly non linear unstable system, it is required to have a controller to stabilize the Inverted Pendulum system. The goal here is to balance the pendulum and to reach the desired position. The task has to be achieved with optimised time specifications. The main aim of this thesis is to develop an intelligent controller for the control of an Inverted Pendulum, which reduces the uncertainties and effect of load disturbances while maintaining the performance with respect to angle and position. In order to achieve this level, this thesis investigates how to unify and hybridize many soft computing techniques including fuzzy logic, neuro-computing, genetic algorithms and Particle Swarm Optimization (PSO). First tuning capabilities of Genetic algorithm and PSO technique for a Proportional Integral Derivative (PID) are explored. Secondly an Adaptive Linear Neural Network controller is proposed for the control of the parameters of the Inverted Pendulum with the help of a PID controller. The results obtained with the neural controller shows how the neural networks are capable of handling uncertainties and for optimised specifications. Finally, the Fuzzy logic systems are developed through knowledge-based and then Neuro-fuzzy approach through the combination of both knowledge-based as well as human intelligence. These applications are analyzed on the Inverted Pendulum using both simulation as well as real time systems. The result of the application shows the capabilities of the proposed algorithms, i.e., the Fuzzy logic and ANFIS based controllers are the successful controllers. With complex uncertain problems and robust, simple viable and visible solutions offered by soft computing techniques for such nonlineariti