Optimized Dataset Generation and Disturbance Rejection Methodology for Neural Network Controllers in Second-Order Systems

Performance of a Neural Networks is highly dependent on the dataset provided it to learn. Traditionally, Neural Networks were used as tools to aid existing controller’s performance or they were trained using existing controllers such as Proportional-Integral (PI) or Sliding Model Controller’s data hence limiting its performance and accuracy to that of the controller itself. In this paper, an ideal dataset that gives the best desirable response is derived theoretically and used to train the neural controllers thus enhancing its performance much better than traditional PI controllers. The paper also proposes a control methodology by using multiple neural networks to reject the disturbance coming into the system and mitigate it easily much faster than the PI controllers with no overshoots. Detailed theoretical derivations are provided for the dataset generation and proposed methodology. The viability of the proposed neural controller was simulated and trained in MATLAB/Simulink environment and its response to variation in reference input has been inferred. The performance comparisons are made with the PI controller data and presented.