INTELLIGENT CONTROLLER FOR INVERTED PENDULUM SYSTEM

Abstract

In this dissertation work Artificial Intelligence (Al) Technique: Artificial Neural Network (ANN) is used to predict the controller model of the Inverted Pendulum (IP) System ANN has been proved a very useful and successful tool in predicting highly nonlinear and complex processes with an impressive accuracy. The reason why ANN attracted researchers is due to its capability of quickly learn the dynamics of the process, apart from that ANN also capable of handling the uncertain and noisy data. A supervised online and offline training has been done using ANN, with three layered feed-forward network architecture with appropriate parameter adjustments and optimization technique in order to minimize Mean Square Error between reference and predicted output. Training is done by `Trainlm' network function that updates weight and bias states according to Levenberg-Marquardt (LM) back-propagation algorithm. The study is focused on how to solve the problem of "parameter variations". The trained controller, both offline trained and online adaptive (ADALINE ANN & GRBF ANN) is applied to the reference IP model is then compared with reference controller i.e. Feedback control law and also tested with different data sets to check the controllability. The results are shown and discussed in detail.