Energy demand forecasting using machine learning perspective Bangladesh

Abstract

Bangladesh is a largely populated country with a total area of 1, 47,570 square km and per capita electricity generation of 182kWh, which is one of the world's lowest. Supplying an uninterrupted power supply to this huge population becomes a challenge for the govt. of Bangladesh. Therefore it becomes necessary to use modern energy management tools like machine learning-based load forecasting techniques to make the decision-making action more efficient. Due to the chaotic nature of electric load demand, an artificial neural network (ANN) is preferred for electrical load forecasting purposes. In this study, we explored several machine learning algorithms like Long Short-Term Memory Network (LSTM), Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors (SARIMAX), and Fbprophet on 11 years of power generation data (2003 to 2014) of Bangladesh to forecast the load demand. The findings of this study reveal that LSTM methods outperformed SARIMAX and Fbprophet methods with the least RMSE and MAPE error 150.26 and 0.4821%. The findings of this study will help the government in policy making and the individual consumer to tackle the energy challenges in the near future.