A comparative study of RSM, ANN and ANFIS models for predicting circuit Thevenin voltage

Abstract

Manual reduction of complex circuits using the conventional Thevenin’s theorem is a time consuming, laborious, and prone-to-mistakes task. Computational intelligence-based techniques have been successfully used in the prediction of process variables, albeit in fields other than circuit reduction. It is therefore necessary to test the suitability of these computational intelligence techniques in circuit analysis, for the elimination of the highlighted challenges of the Thevenin’s theorem. This research paper presents a comparative study of the Response Surface Methodology (RSM), Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System models for predicting the circuit Thevenin voltage. The Taguchi orthogonal array has been utilized in designing the experiment with three levels for each of the three control resistor variables. The Levenberg-Marquardt training algorithm has been implemented in the ANN modelling. Based on the Mean Absolute Percentage Error (MAPE), the ANN and RSM predicted values have been compared to each other. Research results show that the RSM, ANN and ANFIS models have prediction accuracies of 90.13%, 93.35% and 95.33%
respectively, in predicting circuit Thevenin voltage. The results show that ANFIS has a higher prediction accuracy of 99.86% when
using training data set. Based on the Student’s t-test, the research revealed that the mean values of RSM and ANN predicted Thevenin
voltages are not significantly different at p < 0.05. The results are a clear exhibition of the superiority of ANN and ANFIS over the RSM
model in circuit Thevenin voltage estimation. Based on the outcome, it is concluded that computational intelligence-based techniques
can be reliably used in circuit reduction.