A Constructive Study and AI-Based Framework for Electricity Demand and Peak Load Forecasting in the Tamil Nadu Power System

Electricity demand forecasting is a critical component of modern power system operation, particularly for regions experiencing high variability in consumption patterns and increasing penetration of renewable energy sources. Tamil Nadu, one of the largest electricity-consuming states in India, exhibits extreme seasonal demand variation, dual daily peak behavior, and a pronounced duck-curve effect due to large-scale solar integration. Conventional statistical forecasting techniques fail to capture these nonlinear and dynamic characteristics, resulting in inaccurate demand estimation and inefficient power procurement. This paper presents a constructive study of existing electricity demand forecasting techniques and proposes an Artificial Intelligence (AI)-based hybrid ensemble framework for accurate dayahead and week-ahead electricity demand and peak load forecasting. The proposed framework integrates Long Short-Term Memory (LSTM) networks, XGBoost regression, and Prophet models, incorporating weather compensation and calendar effects. Experimental analysis demonstrates a day-ahead Mean Absolute Percentage Error (MAPE) of 2.5% and a week-ahead MAPE of 3.9%, with peak demand prediction accuracy within ±2%. The proposed system enhances grid reliability, optimizes renewable energy integration, and reduces operational costs.