Scaleable Load Signature Generation with Low-Cost Meters for Machine-Learning-Based Load-Detection

Abstract:
Rising energy prices make it necessary to identify potential savings for households and companies, as energy bills do not offer transparency in the breakdown of consumption. Non-intrusive load monitoring (NILM) helps to extract the consumption of individual appliances from the total consumption of the grid. This makes it possible to identify potential energy savings and the causes of expensive peak loads. This paper presents a method using cost-effective and easy-to-install measurement hardware to capture electrical load signatures for the development and application of NILM algorithms. It describes the data collection, processing, and integration required for training and validating load detection algorithms. Various training and testing scenarios demonstrated that the accuracy of detecting individual consumer loads can be improved using a self-created dataset of load signatures compared to a public dataset.