AI BASED APPROACH TO PREDICT THE SOLAR ENERGY IN SMART GRID ENVIRONMENT

Abstract

Due to various environmental problems power generation using renewable energy resources is gaining popularity. Among all the renewable energy resources solar energy is most popular as it is readily available. Because of this electricity generation using solar photovoltaic system is becoming popular as it requires less maintenance and it is environment friendly. However variation in meteorological parameters and solar irradiance causes variation in power generated by solar photovoltaic system. This uncertainty can be tackled by using energy storage system, integrating energy management system. However for proper designing of energy management system forecasting becomes important. Forecasting helps in reducing the uncertainties in solar power generation. Therefore for proper designing and development of solar photovoltaic system forecasting of solar energy becomes very important. The main aim here is to minimize the impact of solar irradiance variation on the power developed. In this dissertation short term solar energy forecasting is done using three methods namely fuzzy logic, artificial neural network, Particle Swarm Optimized artificial neural network. These three methods are used for forecasting purpose in various other fields. Fuzzy logic is based on human decision making. It takes decision based on vague and imprecise data. Artificial neural network is based on machine learning. However output provided by ANN is fluctuating in nature and is less accurate. So to increase its accuracy, optimization technique is used. PSO algorithm is used for improving the ANN and to get more accurate results. MATLAB is used for implementing these three technique for short term forecasting. Mean absolute percentage error (MAPE) is calculated for all the three methods. For fuzzy logic MAPE comes out to be 4.02%, for ANN MAPE comes out to be 4.12% and for PSO-ANN MAPE comes out to be 3.23%