Applications of hierarchical support vector machines for identifying load operation in nonintrusive load monitoring systems

Abstract

Due to the global warming and climate changes, using and conserving of the electrical energy effectively becomes a very important issue recently. If each appliance operates efficiently, the power energy can be used and conserved effectively. Further, the greenhouse gases can also gradually be reduced. This paper* proposes a Non-Intrusive Load Monitoring (NILM) system that employs a decision mechanism based on hierarchical support vector machines (HSVM) to identify the energizing and de-energizing statuses of each load. In the proposed system, the load energizing and de-energizing transient features are extracted from load current waveforms that are acquired at the power entering point. The HSVM performs load identification tasks based on the features. Although a support vector machine (SVM) identifier can solve nonlinear separable cases and optimize the decision hyper-plane, it is a two-class identifier. On the other hand, the NILM problem is a multi-class identification problem. Since any multi-class identification problem can be decomposed into a series of two-class identification sub-problems, the proposed system applies a group of binary SVMs that are arranged into a form of hierarchical tree to identify the operation status of each load. Each node within the tree represents a binary SVM identifier, and each identifier is assigned to solve a corresponding two-class sub-problem. Through this hierarchical decision process, the class of an undetermined datum can be identified. Finally, the results in different experiment environments show that the proposed system is able to identify the operation status of each load, and it also has appropriate robustness.