Application of online trained Echo State Networks for harmonic compliance issues

Complexities with existing algorithms have thus far limited supervised training techniques for Recurrent Neural Networks (RNN) from widespread use. Echo State Networks (ESN) present a newer approach to training RNNs. However for certain applications, it is mandatory to learn and adjust the ESN parameters online, i.e. during the actual task that the ESN is supposed to perform. Certain properties of ESNs make online learning unsuitable. This paper proposes a modified version of the standard ESN wherein the output weights are trained online, as against computing it. The new algorithm is applied to determine the true harmonic current injection of a nonlinear load in a power distribution network. Experimental results presented in this paper confirm that attempting to predict the Total Harmonic Distortion (THD) of a load by simply measuring the load´s current may not be accurate and the results of the new algorithm are validated with actual system measurements. The method of predicting the true harmonic injection of a nonlinear load is referred to as load modeling [1] and is applicable for both single and three phase loads.