Optimizing K-means clustering center selection with density-based spatial cluster in radial basis function neural network for load forecasting of smart solar microgrid

Abstract

Many researchers have investigated estimating and forecasting load power by utilizing many approaches and techniques in neural networks. In this case study, a novel method is proposed to achieve higher accuracy in load-predicting performance in the smart solar microgrid. The K-means cluster is optimized with a density-based spatial cluster and is then utilized to determine the center points in the radial basis function neural network. The proposed method is analyzed and evaluated in the dataset, which is accumulated from the advanced meter infrastructure (AMI) in the smart solar microgrid in 6 months. The proposed methodology is deployed in load power forecasting in various horizons ranging from 10, 20, and 30 min. This optimized technique was inspected and compared against persistence methods, which only apply K-means cluster for center selection in RBF neural network, by using MATLAB simulations. The experimental results proved that the developing enhancement could achieve the maximum improvement of 7.432% R-square, 70.519% mean absolute percentage error (MAPE), and 80.769% root mean squared error (RMSE). The optimized algorithm could effectively eliminate the maximum average of 2.418% of the outer points in the dataset, which decreased the learning time during the modeling process and acquired better convergent velocity and stability compared with the persistent method. Moreover, when combined with enhanced methodology, the 10-min interval data had higher effectiveness and accuracy than the 20-min and 30-min data.