Research On Residential Electricity Behavior Recognition Technology Based On Multi-feature Fusion

Resident power consumption is an important part of intelligent power consumption.The monitoring of resident power consumption behavior is of great significance for power grid dispatching and safe,stable and economic operation.The traditional idea of intrusive residential power monitoring is to install measuring transmission devices for each power terminal(electrical appliances),which requires a lot of financial and material resources and has poor economy.Hart proposed a non-intrusive load monitoring(NILM)technology to solve the above shortcomings.This paper mainly uses NILM to identify residential electricity consumption behavior.NILM has three main links:data processing,feature extraction and load identification.In the link of feature extraction,this paper proposes a method of load multi-feature fusion to select the optimal combination of features according to the characteristics of typical residential load steady-state power consumption.In the link of load identification,a residential power identification model based on steady-state power characteristics is proposed.The measured data prove that the optimal combination of features can improve the accuracy of residential power identification compared with active power as a single load feature.Secondly,in view of the limitation that the steady-state electricity characteristics can not correctly identify the status switching of power equipment,a residential power load identification model based on transient event identification is proposed,and the above model is tested by the measured data.The experimental results show that the residential power consumption behavior recognition model which combines the optimal combination of steady-state multi-features and transient event recognition can be applied to both steady-state electricity consumption and transient events.The specific contents of this paper are as follows:Firstly,the paper introduces the principle of NILM technology and the realization method of each link,and then focuses on the link of load identification link——pattern recognition principle and algorithm.Supporting vector machine(SVM),decision tree and k-NN classification are the representative of supervised classification algorithm and K-means clustering is the representative of unsupervised classification algorithm.The principles of each algorithm are introduced respectively with examples.Secondly,energy-saving lamps,notebook computers,air conditioners,refrigerators,hair dryers and electric kettles are selected as typical residential electrical loads from resistance type electrical appliances,motor type electrical appliances,electronic type electrical appliances and power electronic type electrical appliances.Current and voltage waveforms of normal working state and start-stop transition phase are measured respectively.The power consumption characteristics of typical residential appliances are analyzed from four aspects:steady-state working current waveform coefficient,V-I curve,harmonic emission level and power characteristics,and the fundamental current,harmonic current,active power and reactive power are extracted as the steady-state power consumption characteristics of typical residential appliances.The current waveforms are analyzed and separated from the start-stop and shift event waveforms of each electrical appliances.The characteristics of power load events,such as total transient time,peak current and peak current,can be extracted to characterize the changes of power load events.Then,the single feature analysis and multi-feature correlation analysis are carried out for the steady-state electrical characteristics of typical residential appliances.According to the principle of multi-feature selection,the PCA principal component analysis method is used to reduce the dimension of 10-100 harmonic currents of loads.The new features obtained from the dimension reduction and other steady-state load characteristics are reconstituted into feature sets,and Fisher linear discriminant method is used to select the optimal combination of Steady-state load characteristics.Finally,a residential power identification model based on steady-state electricity characteristics is proposed,and the optimal load feature combination is verified by the measured data of a residential power supply line to improve the accuracy of residential power identification.Since only using steady-state electricity characteristics to identify residential power consumption is not suitable for the situation of residential power status switching,a residential power load identification model based on transient event recognition is proposed,and the residential power transient events are identified by the measured data.And the results show that the fusion of transient event recognition can make up for the deficiency of steady-state load identification and improve the accuracy of residential power identification.