NILM Load Identification Based on CNN and V-I Trajectory

In the process of grid intelligence and digitization, non-intrusive load monitoring (NILM) is a new technology that enables energy consumption monitoring and rational power supply. It identifies the energy consumption of various electrical appliances by monitoring the overall power consumption in a household or building. Traditional NILM methods typically use signal processing and statistical techniques to analyze power consumption signals. In recent years, convolutional neural networks (CNN) have achieved great success in the fields of image and signal processing. This paper aims to explore the use of CNN for event-based NILM load recognition. Firstly, we will discuss event detection methods and data preprocessing techniques for NILM. Then, we will introduce the basic principles and structures of CNN, as well as its application in recognizing V-I trajectory of power data. Afterwards, we will select appropriate loss functions and optimization algorithms to train the CNN and evaluate its performance in load recognition tasks. The experimental results demonstrate that the proposed method achieves excellent performance in the identification of six typical residential loads, with an overall load identification F1-score of 96.2%. Specifically, the F1-score for identifying purely resistive loads reaches 99.75%. This significantly improves the level of NILM load recognition, enabling more precise and intelligent device identification and energy management.