Machine Learning-Based Load Forecasting for Nanogrid Peak Load Cost Reduction

Increased focus on sustainability and energy decentralization has positively impacted the adoption of nanogrids. With tremendous growth, load forecasting become crucial for their daily operation. Since loads nanogrids have large variations with sudden usage household electrical appliances, existing models, majorly focused lower volatile loads, may not work well. Moreover, abrupt operation appliances in a nanogrid, even shorter durations, especially “Peak Hours”, raises cost substantially. In this paper, an ANN model dynamic feature selection is developed to predict hour-ahead based meteorological data lag 1 h (t-1). addition, by thresholding predicted against average previous hours, peak time indices are accurately identified. Numerical testing results show that can Mean Square Error (MSE) 0.03 KW, Absolute Percentage (MAPE) 9%, coefficient variation (CV) 11.9% 20% savings shifting off-peak hours.