A Method Of Non-Intrusive Household Appliance Composition Analysis

Mastering household appliance composition is the basis for designing and implementing appropriate demand side management measures for residential users.It’s almost unworkable to conduct door-to-door surveys or monitor the electrical appliances directly,and the existing non-intrusive load monitoring methods are based on high frequency data measured at the users’ incoming lines,which is difficult to implement in large scale too.In view of the above problems,this paper studies the methods for non-intrusive analysis of household appliance composition which accords with the current metering condition for household electricity consumption.Firstly,the paper studies on the preferences of household appliance combinations(HACs)of families with different generation structures.A density based method is first used to cluster the power for each high-power electrical appliance.After that,based on the statistical information of the ratios of the families with various generation structures and the market shares of the household appliances of different power levels,a fitting method is proposed to analyze the various families’ preferences for the household appliances’ power levels(FPALs).Then a branch and pruning process is conducted to produce the common HACs for each type of the families,whose preference degrees are calculated according to FPALs.Conclusion of the various families’ preferences for the appliance combinations(FPACs)constructs the basis for calculating the priori probabilities of the various HACs.Secondly,a bottom-up simulation method is presented for household load profiling.For this purpose,typical load curve of single operation is concluded by field metering or simulation for each common electrical appliance.Meanwhile,based on the results of a questionnaire investigation,the probability distribution models of the appliances’ starting times and lengths of running are established for families with various generation structures.Next,Monte Carlo simulation is used to produce the daily load curves of the appliances,the addition of which further produces the household daily load curves.This method is useful for calculating the training probability of a certain residential electricity consumption mode under a certain appliance combination.On the basis of the above research,a Bayes classification based non-intrusive method is presented for household appliance combination analysis.Typical residential electricity consumption patterns are first clustered by the K-means method.After calculating the prior probabilities of the HACs on the basis of FPACs and calculating the training probabilities of the various electricity consumption patterns under various HACs by the simulation method,the Bayes formula is used to evaluate the posterior probabilities of the various HACs for the various electricity consumption patterns.The several HACs with maximum posteriori probabilities are most possible for the customers with the corresponding consumption patterns.Case studies are carried out for Shanghai,the results of which verify the feasibility and reasonability of the methods for of FPAC and household load profiling analysis.Taking 1306 families in Shanghai as test samples,typical residential electricity consumption patterns are clustered.Afterwards,non-intrusive analysis of household appliance composition is conducted for each of the patterns.The analysis results are inspected and the results show that the accuracy of the method reaches 82.88% when at most 10 HACs can be related to each consumption pattern.It’s also concluded that the presented method is especially applicable for families with small or medium scales of consumption,since the HACs concluded for such families are quite concentrated.