Quantifying Flexibility Of Smart Home Appliances For Demand Response Via NILM Algorithm

The power consumption characteristics of smart home appliances(SHAs)are highly related to the regional daily load pattern and SHAs account for a high proportion of household loads.SHAs have great potential in the demand response(DR)market and can effectively reduce the peak-to-valley difference in grid loads.Recognizing SHA load patterns and quantifying their flexibility in participating in DR will help grid operators to develop reasonable control strategies.SHA load patterns are difficult to obtain because they are hidden in the household aggregated curves.Non-intrusive load monitoring(NILM)can break the shackles of traditional load monitoring methods,but most NILM algorithms are limited to supervised methods or preset training processes.The load signatures of SHAs extracted by probability statistics method can help to improve the estimation accuracy of NILM.This paper analyzes the power consumption characteristics of SHAs,including air conditioners(ACs),electric vehicles(EVs),electric water heaters(EWHs),clothes dryers and dishwashers.The load patterns(starting and ending times of power consumption events,duration of events,and time intervals of adjacent events)of SHAs under different time types are counted and identified,and their distribution curves are fitted with known probability distribution functions.Based on the statistical characteristics of SHAs,the optimal clustering algorithm is used to mine the similarity of the electricity consumption behavior among SHA users.In order to disaggregate specific appliances from the household aggregated curve,a training-free unsupervised non-intrusive load extraction(NILE)algorithm and a NILM algorithm are proposed.NILE automatically recognizes EV loads charged by different rated power levels through steps such as load data preprocessing,event detection and matching,charging event recognition,and charging event integration.Event detection based on the skipping power difference can effectively identify the changing edges of power events.Charging event recognition based on bounding box fitting and load signatures can effectively mark the start time,end time and power amplitude of the event,and can avoid confusion with other appliances.NILM gradually separates the high-power SHAs through data preprocessing,power block segmentation based on skipping power difference and sets of segmentation points,and power segment extraction based on simulated annealing for the combinatorial optimization problem.In addition,it is verified on the actual data set that the proposed algorithms have higher estimation accuracy.To quantify the DR flexibility of SHAs,a price-based DR model is established.By defining the DR objective function and constraint set and setting the SHA potential available time periods,the timing of the occurrence of power consumption events for SHAs is rescheduled.This paper combines the DR model with the nurse scheduling problem and uses integer programming to solve it.The flexibility indicators are defined according to the changes before and after SHA’s participation in DR.The case results show that EVs have the highest flexibility in participating in DR,followed by EWHs and clothes dryers,and ACs have the least flexibility.