Parallel processing of probabilistic models-based power supply unit mid-term load forecasting with apache spark

Abstract

Mid-term load forecasting (MTLF) of power supply unit (PSU) is an essential part of refined distribution network planning. By analyzing a large amount of historical data accumulated in electric automation systems, accurate MTLF result can be obtained with the help of big-data and parallel computing technology. In this paper, a dynamic bayes network (DBN)-based MTLF model is proposed to forecast the peak power load of next year of each PSU. In the first stage, we improve the accuracy of MTLF model by using dynamic radius DBSCAN algorithm to determine the optimal state division. In the second stage, to improve the computation efficiency, the calculations of multiple probability matrixes and the modified forward algorithm are implemented on an apache spark-based parallel computing platform. The experiment results indicate that the parallel processing of DBN-based MTLF model has superior performance in accuracy, efficiency, and versatility.