New intelligent direct power control of DFIG-based wind conversion system by using machine learning under variations of all operating and compensation modes

Abstract

Wind Turbine (WT)-based Doubly-Fed Induction Generator (DFIG) is a nonlinear system, in which the wind has variable behavior, and the local reactive power depends on the random demand of the AC grid; all these make conventional controls insufficient for an adequate power output. Thus, a new direct power control (DPC) strategy is proposed in this paper, to force the system to track the desired dynamics with higher performance and less drawbacks. This new approach is based on machine learning, Neural-Network- and Neuro-Fuzzy-DPC (NN- and NF-DPC), both are designed to overcome the problems related to power control, wind and local reactive power variations. All operating modes (sub-synchronous, super-synchronous and synchronous modes), with the possibility of local reactive power compensation are considered in this paper. Both NN and NF networks have been trained under the MATLAB interface. The trained NF showed better efficiency than the NN; the learning time is reduced to a few seconds with less computational effort, and less complexity. The effectiveness of both controls has been confirmed through simulation tests using MATLAB software. The obtained results have shown that the NF-DPC strategy presents better performances than NN-DPC and conventional control, with fast response, robustness, less power ripples and reduced Total Harmonic Distortion (THD) in the generated currents.