Hierarchical sparse learning for load forecasting in cyber-physical energy systems

Cyber-physical energy systems, which emerges as the approach for integrating physical layers and control networks, have drawn extensively attention in recent years. Electric load forecasting is believed to be an important issue in CPES for its applications in prices determination and automatic generation control. Conventional deterministic load forecast method have drawbacks to providing information about the probability distribution of the prediction results, which are important for stochastic decision in power systems. This paper explores a hierarchical probabilistic approach for short-term load forecast, which combines sparse Bayesian learning with empirical mode decomposition, in order to obtain a componential forecasting results, as well as the forecasting uncertainty. Mahalanobis distance based similar day weighting is introduced to prune the training data. The numerical testing results illustrate that the proposed approach exhibits better performance in comparison with original SBL model and weighted SBL without componential analysis.