Non-intrusive Load Identification Technology Based On Voltagereactive Current Trajectory

In recent years,with the frequent occurrence of energy shortages and envi-ronmental pollution,realizing effective energy monitoring and management has become a new research hotspot.Among them,monitoring the electricity load usage and accurate identification of households is an important part of realizing effective energy monitoring.For reasons of economy and information security,non-intrusive household load identification schemes are more practical than intrusive schemes.Aiming at the existing research framework of non-intrusive household load identification based on traditional image features,this paper conducts research and improvement in three aspects:event detection,feature image construction,feature extraction and load identification,and proposes a method based on voltage-reactive current(V-I_f)Non-intrusive load identification technical solution for trajectory.In this technical solution,this paper proposes the Cumulative Sum(CUSUM)event detection algorithm,the V-I_f trajectory gray image construction method,and the Triplet Conv-olutional Neural Network(Triplet CNN)load feature extraction and recognition algo-rithms.The improvement of the three parts has resulted in better recognition results.First of all,in view of the error detection and missed detection problems of the two existing event detection algorithms in different situations,this paper proposes a composite CUSUM algorithm for secondary event detection in windows which detects the event,which is a combination of the fast event detection algorithm and the bilateral CUSUM algorithm based on sliding window.This algorithm improves the missed detection problem of the bilateral CUSUM algorithm based on sliding window,and verifies its performance on the real data set.Secondly,based on the existing steady-state current waveforms or voltage-current trajectory images,there is insufficient discrimination between resistive load classes,and a new voltage-reactive current(V-I_f)trajectory image and image generation algo-rithm are proposed.The steady-state current removes the active component,and the reactive component in the resistive load current is normalized and amplified,and together with the voltage data,it becomes a V-I_ftrace image with a greater degree of discrimination.At the same time,the active power is quantified as the gray value of the image to further enrich the electrical information of the load image,and to verify the advantages of the V-I_f trajectory gray image in the subsequent load identification algorithm.Aiming at the generated V-I_f trajectory grayscale image with good distinction between classes and the characteristics of less sample data,a Triplet CNN network is designed for non-intrusive household load feature extraction and recognition.This paper combines the V-I_f trajectory grayscale image with the Triplet CNN network and performs performance verification on a real data set.It verifies the advantages of the V-I_f trajectory grayscale image as a feature image and the recognition performance of the Triplet CNN network is better than Siamese CNN Network and CNN network of the same type of literature.