Echo State Networks Based Method for Harmonic Extraction in Shunt Active Power Filters

Abstract

With the wide use of power conversion devices, harmonic currents are being injected into the power grid. Shunt Active Power Filters (SAPF) is a power electronic device to compensate the harmonic currents caused by nonlinear loads. As the foundation of the harmonics recognition and compensation, harmonic extraction techniques are becoming more and more important. This paper proposes a new harmonic extraction method based on the Echo State Networks (ESN). ESN is a new type of Recurrent Neural Networks (RNN), which has much faster training speed than other types of RNN. To evaluate the dynamic system modeling capability of the ESN, the ESN with different dynamic reservoir size are discussed. The performance of the ESN based harmonic extraction method is compared with traditional methods and method based on multilayer perceptron networks (MLP). The ESN algorithm is trained and tested in MATLAB.