Performance analysis of neural network based MRAC

Analysis of effects and outcomes of applying model reference adaptive technique to processes with different control challenges has been discussed in paper. Results have been plotted in each case and qualitative discussion has been presented to support same. To make base of analysis stronger two case studies have been considered. Comparison has been done about applying MIT rule and using neural network for model reference adaptive control. For case study simulation of two bench-mark process control applications, level control in coupled tank and level control in quadruple tank has been discussed. Systems have unique feature like nonminimum phase behavior for certain parameter range, nonlinearity in differential equation, scope to add uncertainty etc. Initially MIT rule technique has been applied for control in both cases. Data of adjustment parameter θ has been saved and used to train NN for applying NN technique. Using NN based algorithm, new values of adjustment parameter θ have been generated. For the system having uncertainty, results shows that as compared to conventional MRAC, θ generated using NN leads to better result and robustness in specific range. Development of MRAC theory and related literature survey has been presented.