Forecasting Solar Irradiance with Weather Classification and Chaotic Gravitational Search Algorithm Based Wavelet Kernel Extreme Learning Machine

In this work an improved KELM based forecasting model is being proposed, which attains a specific level of prediction of solar irradiance affecting PV power management. The new method is known as Wavelet KELM. A new optimization technique known as Gravitational Search Algorithm (GSA) is implemented to optimize various parameters of the kernel function. The novelty of this work is that, it focuses on a KELM learning algorithm with parameter optimization for exact solar irradiance forecasting. The exact prediction of irradiance is highly essential for system level stability and future large scale PV installations. The GSA based optimized KELM (GSA-KELM) is implemented for short term solar irradiance forecasting based on various weather conditions.  A hit and trial method was used for the selection of kernel function parameters which affects the forecasting model performance. The optimized kernel structure not only minimizes the arbitrariness of the variables but also makes the process fast by extenuating the choice of parameter on the basis of the user which has to experience the repeated trial method for various kernel functions. Thus OKELM impart more accuracy in prediction within less time and outperforms the basic KELM. In this work, data collected from a photovoltaic power plant in India (the capacity being 1 MW) has been considered for forecasting model validation.