DBSRad‐LSTM: DBSCAN Clustering for Load Forecasting in Microgrids Using Radial LSTM

ABSTRACT In this study, an innovative approach with enhanced recursive feature clustering for load forecasting in smart solar microgrids by integrating density‐based spatial clustering (DBSCAN) and radial basis function neural networks (RBFNN) encoder is proposed. Our methodology is based upon novel density‐based clustering with feature recursive forwarding RBFNN‐LSTM for high‐accuracy micro‐ and macro‐feature learning in temporal data. DBSRad‐LSTM performance is evaluated using three distinct datasets: Panama electricity consumption, Italy solar electric load, and a custom dataset tailored for smart grid applications. Through rigorous comparative analysis, our DBSRad‐LSTM model outperformed traditional machine learning models such as gated recurrent unit (GRU), long short‐term memory (LSTM), and convolutional neural networks (CNN) across several metrics. Specifically, DBSRad‐LSTM demonstrated superior performance in terms of accuracy, thereby contributing enhanced load forecasting capabilities. The proposed model integrating RBFN linear functionality with the attention of LSTM and DBSCAN clustering to enhance the learning of temporal data outperformed CNN, SVMCNN, and GRU on Panama power consumption, Italy electric load and bespoke datasets, obtaining a higher R 2 value of 0.89 and much lower MSE 0.015, RMSE 0.123, and MAE of 0.009. Achieving a 9%–25% improvement in error metrics and an average 13% better fit. By offering a distinct clustering‐based approach that improves on existing methods, this research makes a substantial contribution to the field of smart grid management and opens the door for more precise and effective energy distribution systems.