A Hybrid Forecasting Methodology: Statistical-Neural Fusion for Electricity Demand Prediction

Abstract

Accurate electricity demand forecasting is crucial for efficient planning and operation of electrical systems, necessitating the integration of diverse demographic, economic, statistical, and engineering data through advanced methodologies to mitigate risks of under or over-investment , this imperative is recognized as a multifaceted problem requiring diverse techniques , this study addresses the complex, non-linear nature of long-term electrical energy loads by designing and developing an advanced hybrid forecasting methodology termed "Statistical-Neural Fusion," which aligns with the need for robust deep learning frameworks, this hierarchical approach utilizes a Level 0 "Statistical Expert" (Multiple Linear Regression model) to capture initial linear relationships and generate a "Smart Feature" forecast based on historical data including Year, Population, and Gross Domestic Product (GDP), considering demand determinants. Subsequently, a Level 1 "Mastermind" (Integrated Neural Network, specifically an MLP) refines and corrects this initial forecast by learning complex, non-linear patterns from an enhanced dataset comprising both original inputs and the "Smart Feature," aligning with the understanding of demand forecasting over various time scales , this method was applied to forecast energy production and maximum load for the national electricity grid up to 2034, relying on a comprehensive database of historical operational, demographic, and economic data.