A data mining based load forecasting strategy for smart electrical grids

Smart electrical grids, which involve the application of intelligent information and communication technologies, are becoming the core ingredient in the ongoing modernization of the electricity delivery infrastructure. Thanks to data mining and artificial intelligence techniques that allow the accurate forecasting of power, which alleviates many of the cost and operational challenges because, power predictions become more certain. Load forecasting (LF) is a vital process for the electrical system operation and planning as it provides intelligence to energy management. In this paper, a novel LF strategy is proposed by employing data mining techniques. In addition to a novel load estimation, the proposed LF strategy employs new outlier rejection and feature selection methodologies. Outliers are rejected through a Distance Based Outlier Rejection (DBOR) methodology. On the other hand, selecting the effective features is accomplished through a Hybrid technique that combines evidence from two proposed feature selectors. The first is a Genetic Based Feature Selector (GBFS), while the second is a Rough set Base Feature Selector (RBFS). Then, the filtered data is used to give fast and accurate load prediction through a hybrid (KNB)-B-3 predictor, which combines KNN and NB classifiers. Experimental results have proven the effectiveness of the new outlier rejection, feature selection, and load estimation methodologies. Moreover, the proposed LF strategy has been compared against recent LF strategies. It is shown that the proposed LF strategy has a good impact in maximizing system reliability, resilience and stability as it introduces accurate load predictions. (C) 2016 Elsevier Ltd. All rights reserved.