New Correlation Technique for RE Power Forecasting using Neural Networks

Abstract

Ambiguity in present power system operation increases due to variable nature of climate and more penetration of Renewable Energy (RE). Therefore, for successful operation of power system network an accurate and efficient forecasting of RE power generation is essential. In this paper, multi-layer Feed Forward Artificial Neural Network (FF-ANN) model is used for training the datasets for short term forecasting of wind power. There are two steps involved in this work namely training and forecasting. During training, for optimizing the parameters of FF-ANN, Levenberg Marquadrt (LM) learning algorithm is used. For forecasting wind power, a new technique has been proposed and the method here is referred as Weighted Least Square Error Correlation method (WLSEC). The proposed method is implemented in C+ + platform. The performance of the model has been tested with practical data, in one of the southern states in India, considering one year historical data with hourly resolution. The Mean Absolute Percentage Error (MAPE) for forecasting wind power hourly is observed as7.32% with proposed method, where as it is 9% with Back Propagation Neural Network (BPNN). This comparison clearly shows the effectiveness of proposed model to fore cast short term (hourly) wind power.