An optimized variant of machine learning algorithm for datadriven electrical energy efficiency management (D2EEM)

The electricity at the most demanded energy source around the globe. At the same time, it is critically limited to meet the demand. There are only two logical solutions to meet this demand. First, to increase the power generation capacity, enhance transmission technology, and improve power generation efficiency. The second, is to manage the energy utilization in the premises. Since the electrical energy consumption is different in each application and management of energy utilization in large scale is complex, therefore this study proposed (Data-driven electrical energy efficiency management) D2EEM using optimized ML. This research is mainly focused on data-driven electrical energy efficiency management using artificial intelligence. Particularly, a university campus is selected as a case study in this research. It is a well-established fact that machine learning is outperforming in terms of prediction and classification. Therefore, in this study a new optimized variant of machine learning algorithms is presented. In this study, a benchmark dataset of energy consumption in a university campus of IIT, India (provided by the Smart Energy Informatics Lab, SEIL) was selected for training and testing the proposed variants of machine learning algorithms. The scope of this study is tri folded, First, an exhaustive and parametric comparative study on a wide variety of machine learning algorithms is presented to evaluate the performance of machine learning algorithms in energy load prediction. The deliverable of this phase is the selection of the best candidate of machine learning algorithm for university campus energy load prediction. The second is the optimization of the best selected machine learning algorithms to further improve the efficiency and efficacy of the prediction. Finally, the proposed algorithms were also validated on another dataset of a university campus in a different region. This study recommends a selection trade-off as the function of prediction efficiency and efficacy of the algorithm. Particularly, the proposed optimized Bagged Trees are the most effective algorithm for energy demand prediction applications, and the proposed optimized Medium Trees are the most efficient algorithm for real-time systems. Likewise, optimized Fine Trees have the optimum trade-off between efficacy and efficiency.