A Short Term Forecasting of PhotoVoltaic Power Generation Using Coupled Based Particle Swarm Optimization Pruned Extreme Learning Machine.

Abstract

Presently, there is a high growth of power demand, and to minimize the air pollution by the conventional power plants, we enforce the renewable power to the existing conventional grids. The non linearity and erratic nature of PV power generation creates a great challenge in the interconnection and grid management. To install further solar plants of large capacity, the prediction of PV power is extremely required. To overcome this challenge, many machine learning techniques has been successfully implemented for short/long term forecasting of the PV power. Among all the methods, Extreme learning machine (ELM) is one among the few victorious methodologies’ in machine learning approaches. One of the key potencies, of ELM, for which it is appreciated, is  low computational effort required for training recent data because the hidden and output layer nodes are arbitrarily selected and rationally decided. Further the conventional ELM does not perform satisfactory in some complicated problems or in case of big data. Thus in this work we have investigated the pruned-ELM (P-ELM) approach as a methodical and programmed approach for developing a forecasting model to predict PV power generation on short term horizon. P-ELM shows statistical ways to compute the significance of inner nodes. Starting from an initial large number of inner nodes, inappropriate nodes are then pruned by taking into account the appropriateness to the forecasting problem. In order to increase the performance of the P-ELM the weights of input layer are optimized by coupled based PSO algorithm.