SHORT-TERM LOAD FORECASTING IN MICROGRIDS: A CLUSTERING-ENHANCED DEEP LEARNING APPROACH

Abstract

Accurate short-term load forecasting is vital for efficient microgrid energy management, unit commitment, and renewable energy integration. Traditional and deep learning models often struggle with the complex, time-varying patterns in residential, commercial, and industrial loads. To address this, clustering algorithms are applied to group similar consumption patterns, enhancing forecasting accuracy. This study presents a clustering-enhanced long short-term memory (LSTM) framework that segments hourly load profiles using K-means, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Gaussian Mixture Models (GMM), and Hierarchical Clustering before training. Using 8,760-point synthetic load profiles per sector, baseline models and standalone LSTMs were compared against clustering-enhanced LSTMs. Results show that clustering reduces mean squared error (MSE) by up to 48% and mean absolute error (MAE) by up to 28% in residential forecasts (GMM), improves commercial forecasting by 18.7% (MSE) and 14.4% (MAE) with Hierarchical Clustering, and yields modest gains of up to 2.4% (MSE) and -0.026% (MAE) in stable industrial profiles with K-means. The proposed framework offers a scalable, sector-specific solution to improve microgrid forecasting and support renewable integration.