Machine learning based source and load forecasting for efficient microgrid energy management system

Abstract

Globally microgrids are becoming more successful for uninterrupted power, power access in remote areas, better power quality and low carbon emissions due to integration of renewable energy sources. The major renewable energy source of microgrids are solar photovoltaic (PV) power generation. However, the solar PV power generation is stochastic in nature. This may create energy management challenge with microgrid in grid tied mode as well as islanded mode. The lower power generation may cause penalty to the generation company and high-power generation without appropriate storage will cause waste of energy. On the other side the load is also highly uncertain in nature. The appropriate source and load forecasting is thus necessary to avoid this situation. This research proposes two different algorithms for solar PV generation and load forecasting namely multiple linear regression and neural network. The entire data analysis path is explained in detail for accurate prediction of independent variables. The solar PV generation forecasting is done by considering weather conditions for three different seasons: winter, summer and monsoon. The load forecasting is done by considering the weather conditions and holidays. The actual and predicated results for both algorithms are tested by mean absolute percentage error. The accuracy of multiple linear regression is 92.67%, 93.79%, 89.3% and 94.05% for winter, summer, monsoon and load forecasting respectively. The accuracy of neural network is 96.6%, 97.1%, 94.9% and 97.8 for winter, summer, monsoon and load forecasting respectively.