photovoltaic power prediction using evolutionary computing based machine learning techniques for reliable grid energy management

Abstract

newline In the present scenario, the Renewable Energy Sources plays an important role to newlineanswer the rapid power demand growth. In particular, the solar energy is one of the most newlinepromising and eco-friendly RES and it s demand is increasing rapidly due to its own newlineadvantages and significant contributions which reduces energy cost. However solar energy newlinegeneration is influenced by metrological factors like solar radiation, cloud coverage, newlinehumidity and temperature. The large-scale integration of solar energy into conventional newlinegrid is affected by these metrological factors. newlineAccurate forecasting of solar power generation or solar irradiance is highly essential newlinefor the successful integration of solar energy into electric grid. There are so many existing newlineforecasting techniques like, Artificial Neural Network (ANN), Auto Regressive Integrated newlineMoving Average (ARIMA), Radial Basis Function Neural Network (RBFNN), Least newlineSquares Support Vector Machine (LS-SVM) model and they have been used in forecasting newlinesolar power. But the investigation of these models is associated with some draw backs like newlineslow convergence, inaccurate prediction and long computational time. newlineThis thesis is focused on short term forecasting of solar power and irradiance using a newlinenew forecasting method known as Extreme Learning Machine (ELM) techniques and newlineoptimized ELM like PSO-EMD-ELM and RPSO-EMD-ELM is proposed to forecast solar newlinepower output variations over the time horizons of 5min, 15min, 30min and 60 minutes newlineahead. In order to reduce the forecasting errors and to reduce nonlinearity of the solar newlinepower, Empirical Mode Decomposition based ELM is studied in the second part of this newlinethesis. The next part of the contribution of the thesis includes investigation of another newlineimproved forecasting model known as couple based PSO based Pruned Extreme Learning newlineMachine (P-ELM) with Empirical Mode Decomposition (EMD). To reduce further newlineforecasting error the last part of the study is focuses on Chaotic Gravitational Search based newlineKernel Extreme Learning Machine. newline