Energy consumption forecasting is vitally important for the deregulated electricity industry in the world. A large variety of mathematical models have been developed in the literature for energy forecasting. However, researchers are involved in developing novel methods to estimate closer values. In this paper, authors attempted to develop new models in minimizing the forecasting errors. In the present study, the economic indicators of the state including population, gross state domestic product, yearly peak demand, and per capita income were considered for forecasting the electricity consumption of a state in a developing country. Initially, a multiple linear regression model has been developed. Then, the coefficients of the regression model were optimized using two heuristic approaches namely genetic algorithm and simulated annealing. The mean absolute percentage error obtained for the three models were 2.00 for multiple linear regression model, 1.94 for genetic algorithm based linear regression and 1.86 for simulated annealing based linear regression.
Online joint-prediction of multi-forward-step battery SOC using LSTM neural networks and multiple linear regression for real-world electric vehicles
