Artificial neural networks to control an inverter in a harmonic distortion compensation scheme

In this paper, two efficient and reliable neural approaches to control an inverter are developed. The objective is to improve the compensation performance of a conventional active power filter (APF) with a homogeneous neural structure allowing an efficient hardware implementation. The first control approach is based on a neural PI regulator. This technique uses an Adaline to determine the PI parameters. The second control approach is a direct inverse control method. It uses two multilayer neural networks with the backpropagation learning in order to identify the Jacobian of the process and to control the inverter. The originality lies in the error signal used for the weight adaption in the first approach, and in the choice of the inputs of the neural networks in the second approach. The performance of the two methods is evaluated through simulation and experimental results and demonstrates the effectiveness of the proposed neural approaches.