Research On Non-intrusive Household Load Identification Based On Wavelet Transform And Random Forest

In order to establish the two-way interaction and effective communication between the user and the electric power company,and to realize the efficient power management of residential electricity load,the household electric power load needs to be monitored and identified so as to further realize the accurate classification and measurement of residential electricity load.Because the non-intrusive monitoring device has a simple and convenient installation,small upfront input costs,strong operability in actual situation and many other advantages.The non-intrusive load identification technology for residential load is the hot trend in this field.In this paper,some typical residential electrical equipment is studied.And the working condition of residential electricity load was simulated by a variety of electrical mixed switching experiments in the laboratory.After the voltage and current data are collected synchronously through the data acquisition card,the active power is then been normalized.Through the multi-scale wavelet decomposition,combined with wavelet transform modulus maximum principle and increment ratio method,the timely and accurate detection of transient events which caused by load switching operation is realized.Compared with the general event detection algorithm which based on the rule judgment,this paper makes a quantitative analysis of the location accuracy of transient event,moreover,it also has high precision for the detection of small power electrical appliances which runs in the background of high power electrical appliances.Compared with the detection algorithm of general change-point which based on wavelet transform,wavelet decomposition is performed by using normalized active power in this paper,it overcomes the disadvantage that the general change-point detection algorithm directly using the voltage or current data to implement wavelet transform but cannot be applied to the residential electricity load.After the precise location of the transient event,the features of voltage,current and active power signal after wavelet transform are extracted from two aspects of steady state and transient state in this paper.It includes the energy of detail and approximate coefficients after wavelet transform,the absolute mode average value of each layer coefficient,and the ratio of absolute modulus maximum and absolute mode average value of the detail coefficients etc.And the principal component analysis method was used to reduce the steady state characteristics of 28 groups.Finally,a load characteristic database based on 5 groups of the steady-state characteristic and 10 groups of wavelet transient characteristics is established.Finally,this paper focuses on the process of load identification using random forest algorithm,and the accuracy is analyzed by combining with the data of a variety of electrical appliance mixed experiments.At the same time,the performance difference of the algorithm is compared with other load recognition algorithms in the same experimental example.The experimental data show that it has high recognition accuracy in this paper,and it can identify the similar loads and the same load which have different gears,it can also accurately identify the continuous change load which has the finite multi-transient process.