Multi-criterion mining algorithm for efficient Home Energy Management System

This paper presents new multi-criterion mining algorithm, which as a module of the Home Energy Management System (HEMS) can identify the operating states of a combination of controllable appliances. There is a close interaction between HEMS and this algorithm. The new incoming samples, collected by HEMS, are associated with appropriate cluster using the mining algorithm. This information is submitted back to HEMS which appropriately controls the appliances by switching them on/off. The strength of our mining algorithm is the combined usage of geometrical and statistical approach to the data clustering. The criteria that embody the geometrical approach include: identification of corresponding min-max parallelepipeds, embodying the new instance by the neighboring parallelepipeds and minimum distance criterion. Box-dimension with its definition as a measure of self-similarity is used as a statistical approach criterion. As a result from the testing of these criteria, we have achieved time optimization and improved accuracy, due to reducing of the number of potential clusters (reducing the number of testing instances but not excluding them for further analysis) without reducing the dimensionality of the clustering problem. The successful identification of the operating states of the appliances, done with our continuously learning algorithm, enables HEMS to improve energy efficiency and cost savings.