Mathematical Modeling and Adaptive Control Strategies for Managing Unforeseen Nonlinear Load Variations

Abstract

Nonlinear loads that vary suddenly and rapidly create extreme harmonic distortion. This poses a risk to the DC link of the Shunt Active Power Filter (SAPF). This paper addresses such risks by proposing an adaptive Artificial Neural Network (ANN)-based method that does two things at once: (i) it detects harmonic components for the generation of the current reference, and (ii) it controls the DC-link voltage of the SAPF in active power filtering during sudden loading shifts. The proposed ANN controller relies on a multilayer feed-forward structure trained online with the Levenberg–Marquardt/backpropagation algorithm; its inputs consist of phase voltages and currents and its outputs correspond to the three compensating reference currents for the SAPF. Multiple step and ramp load changes have been tested through simulations in Matlab/Simulink and the results of the proposed adaptive ANN have been benchmarked against the conventional p–q theory and a fixed ANN. It is evident that adaptive ANNs are able to regulate the DC-link voltage more precisely while outperforming in supply current total harmonic distortion (THD) minimization for all tested scenarios. The analyses presented in the manuscript have now been supplemented with a sensitivity analysis (short-circuit ratio, switching frequency, measurement noise) and a statistical analysis (mean ± standard deviation over several executions) to validate robustness.