A Methodology for Electricity Demand Forecasting Using a Hybrid Approach

Abstract

Load forecasting (LF) plays a crucial role in energy production planning and scheduling, simplifying budgeting processes, and improving power supply reliability. The available integrated solutions are superior to conventional approaches while considering the uncertainties of weather conditions. The primary objective of LF is to establish an optimal load model for the power grid, conducted offline, to achieve accurate predictions, thereby minimizing operational costs and enhancing grid stability. In this work, an integrated LF model is proposed that uses modified combined ensemble empirical mode decomposition with adaptive noise (MCEEMDAN), Shannon entropy (SE), and long short-term memory (LSTM) techniques. To demonstrate the efficacy of the proposed method, this manuscript utilizes a real-time dataset containing actual load data, social & temporal variables and meteorological parameters including temperature, humidity, and rainfall, gathered from Raipur region in Chhattisgarh state, India. A comparative analysis of the proposed method is conducted against other available approaches, including various time-series decomposition methods, different machine learning techniques, and alternative test system.