The comparison of ANN-BP and ANN-PSO as learning algorithm to track MPP in PVSystem

Abstract

The P-V curve of photovoltaic system exhibits multiple peaks under various conditions of function and changes in meteorological conditions which reduce the effectiveness of conventional maximum power point tracking (MPPT) methods. Artificial Neural Network (ANN) is one of soft computing used for learning, modeling, and analyzing a very complicated phenomenon. Furthermore, there is an algorithm based on meta-heuristic, which is usually used for some optimization problems. One of meta-heuristic algorithms used in this paper is Particle Swarm Optimization (PSO) algorithm. In this paper, a comparison between ANN using PSO and ANN used back propagation as a learning algorithm to track MPP in photovoltaic system. Each training model was conducted with different learning rate, but the number of neurons and activation functions used was similar in each training model. To evaluate both training models of ANN, Mean Square Error (MSE) was used. The result showed that ANN using PSO as a training algorithm require 17 epochs to convergent, but ANN using back propagation require 105 epochs to convergent. Furthermore, the average value of power generated from PV system, ANN using PSO as training algorithm for track MPP was 90.92 kW and ANN using back propagation as training algorithm for track MPP was 88.65 kW.