Neural Network with Extended Input for Estimating Electricity Consumption Using Background-based Data Generation

Abstract

Non-intrusive load monitoring (NILM) can infer the electricity consumption of individual target appliances by only collecting and analyzing the aggregate power data at the single power entrance point. This detailed information obtained is of great significance to smart grid operation and energy saving. For solving the problems of feature loss and training data acquisition in current deep neural network-based NILM methods, this paper presents an extended input neural network method using background-based data generation for estimating electricity consumption. First, the extended input method is proposed, which ensures a complete working cycle to be fed into the neural network each time. Then, a synthetic method based on the power use background is used to generate abundant and desirable training data. Finally, the proposed methods are validated through comparison tests.