Extreme learning machine approach to estimate hourly solar radiation on horizontal surface (PV) in Surabaya -East java

Abstract

Solar radiation is a source of alternative energy that is very influential on the photovoltaic performance in generating energy. The need for solar radiation estimation has become a significant feature in the design of photovoltaic (PV) systems. Recently, the most popular method used to estimate solar radiation is artificial neural network (ANN). However, a new approach, called the extreme learning machine (ELM) algorithm is a new learning method of feed forward neural network with one hidden layer or known as Single Hidden Layer Feed Forward Neural Network (SLFN). In this research, ELM and a multilayer feed-forward network with back propagation are implemented to estimate hourly solar radiation on horizontal surface in Surabaya. In contrast to previous researches, this study has emphasized the use of meteorological data such as temperature, humidity, wind speed, and direction of speed as inputs for ANN and ELM model in estimating solar radiation. The MSE and learning rate has been used to measure the performance of two methods. The simulation results showed that the ELM model built had best performance for 400 nodes in which MSE and learning rate achieved were 5,88e-14 and 0,0156 second, respectively. The values were much smaller compared with the results of ANN. Overall, the ELM provided a better performance.