Hybrid Short-Term Load Forecasting using the Hadoop MapReduce Framework

With the wide deployment of communication networks and advanced metering infrastructures (AMI) in smart grids, data from the electric demand side is growing exponentially and becomes more complex. To meet the high precision and high efficiency requirements of short-term load forecasting (STLF) in the big data era, this paper describes how an efficient hybrid STLF model can be realized through the utilization of Hadoop MapReduce framework. First, the variational mode decomposition (VMD) is used to decompose the original historical load data into a series of modal function components with different characteristics. Then, leveraging the Hadoop MapReduce architecture, distributed deep belief networks (DBNs) are built to predict each load component generated by the VMD. To further improve the prediction accuracy, the mutual information theory is introduced to select matching factors from the historical load, temperature and date type as the input variables. Test results demonstrate the efficiency of the proposed approach.