Energy Forecasting in Smart Grids: Toward an Artificial Intelligence-Driven Approach

Abstract

This thesis focuses on developing an accurate electricity consumption forecasting model within smart grids by leveraging modern artificial intelligence techniques to enhance energy management and improve efficiency in residential environments. The study began with the preparation and preprocessing of detailed electricity consumption data for a single household, including measurements such as active power, reactive power, voltage, current intensity, and consumption from sub-metered appliances. Initially, a Long Short-Term Memory (LSTM) neural network model was applied for load forecasting. However, after comparative evaluation, the XGBoost model demonstrated superior accuracy and performance in reducing prediction errors. Consequently, efforts were concentrated on optimizing the XGBoost model through hyperparameter tuning and measure the features (using normalization or standardization), resulting in highly accurate forecasts with minimal error margins. To extend the scope of the research, the model was applied to six different households in a specific area, with hourly and daily predictions performed to provide a detailed and real-time view of electricity consumption patterns. This expansion allowed for a comprehensive understanding of consumption variability and enhanced the model’s predictive capability. In the final phase, an integrated energy management framework was developed based on the forecasting results. This framework optimizes the utilization of multiple energy sources including solar, wind, battery storage, and the electrical grid thus enabling efficient energy distribution and reducing reliance on the grid when possible. The framework represents a significant step towards building intelligent and sustainable energy systems capable of adapting to demand fluctuations and maximizing operational efficiency. This thesis makes a valuable scientific and practical contribution to the field of smart grids by combining AI techniques with energy management strategies. It opens new avenues for innovative solutions that support energy sustainability and help reduce environmental impacts in the future.