IntelliForecast: A Machine Learning Framework with Feature Engineering for Efficient Forecasting of Electricity Load

Abstract

Modern smart grids of electricity align with the sustainable development goals of the United Nations (UN). Since electricity production and distribution are crucial in sustainable development, research in this area is highly significant. Artificial Intelligence (AI) has emerged as a powerful tool for addressing various challenges across real-life applications, including smart grids. In this regard, electricity load forecasting is indispensable for efficiently managing the demand-supply balance in electricity. This paper aims to develop and propose an intelligent machine learning framework, IntelliForecast, that integrates feature engineering with advanced machine learning models for short-term electricity load forecasting. Specifically, we propose two algorithms: Hybrid Feature Engineering (HFE) for selecting significant features and Learning-based Electricity Load Forecasting (LbELF) for efficient forecasting. Empirical results reveal that the IntelliForecast framework achieved the highest forecasting accuracy of 95.60% for hourly predictions using a Neural Network model optimized with Random Search Optimization (RSO), outperforming Multilinear Regression (MLR) and standalone Neural Network (NN) models. Additionally, the framework reduced Mean Absolute Percentage Error (MAPE) to 0.0169, showcasing its robustness in accurate and efficient forecasting. Our framework can be embedded into modern smart meters, enabling real-time forecasting and facilitating energy trading.