Research On Event-Based Non-Intrusive Load Monitoring Algoithms

Nowadays,energy is an important factor restricting the development of human society,and improving energy efficiency is the main method to improve energy sustainability.Smart power technology can monitor the load of power users.Through direct load control,peak shaving and valley filling,user load becomes a relatively controllable resource to support demand-side management(DSM)of power and improve energy utilization.As an important part of smart power technology,the development of load monitoring(LM)technology is accelerating with the increasing demand on the user side and the energy supply side.Among them,non-intrusive load monitoring(NILM)technology has become the mainstream due to its low cost.Existing research of event-based non-intrusive load monitoring technology mainly revolves around how to identify and decompose the user’s electrical load composition based on the extracted electrical characteristics of the load.At the same time,the development of event detection algorithms provides support for event-based load decomposition.Based on the above,this thesis proposes an event-driven load decomposition algorithm based on targeted load multi-dimensional features mining and a hybrid event detection algorithm:(1)We propose an event-driven non-intrusive load monitoring algorithm based on targeted mining multidimensional load characteristics.The proposed NILM solution aims to separately model different electrical appliance types to mine the unique electrical characteristics of the electrical appliances from multi-dimensional features,so that all appliances can achieve the best classification per-formance.First,we convert the multi-classification problem into a serial multiple binary classification problem through a pre-sortmodel.Then,Contrastive Loss k-Nearest Neighbour Model(CTLKNN)with trainable weights is proposed to targeted mine appli-ance load characteristics.The simulation results show the effectiveness and stability of the proposed method.Compared with thereferences,the proposed method has improved the identification performance of all electrical appliance types to different degrees.(2)We propose a hybrid event detection algorithm based on moving average change.Irregular fluctuations of the signal caused by the switching event will lead to false detection of the event detection algorithm.This article summarizes the transformation rules of switching signals,and proposes a hybrid event detection algorithm(HEDA)to filter out different types of false detections.The proposed algorithm is improved based on the event detection based on moving average change(MAC)algorithm.It uses low-pass filtering to reduce steady-state fluctuations.At the same time,a false detection removal algorithm is introduced to filter out non-linear loads.The fluctuation and the false detection caused by the shock.Experiments show that the MAC is less prone to false detections than other basic algorithms,and the low-pass filter processing can smooth the steady-state fluctuations of nonlinear equipment very well.The proposed HEDA can also well avoid false detections caused by steady-state fluctuations and transient oscillations.