Research On Non-intrusive Decomposition Of Residential Electricity Load Based On Machine Learning

With increasing environmental protection efforts and the gradual exhaustion of energy,energy conservation and emission reduction have become a powerful measure to create a clean and environmentally friendly society.In order to meet the goal of energy saving and emission reduction,the power grid is gradually developing towards a smart and economized direction.As one of the key technical links on the user side of the smart grid,the home load monitoring technology is the foundation for building a strong smart grid.In order to break through the bottleneck of traditional electricity meters that only measure power and cannot analyze user load composition and behavior habits,non-intrusive load monitoring(NILM)technology has been favored by many scholars at home and abroad due to its low cost and protection of user privacy attention.Load detection technology is divided into two categories.One is intrusive load monitoring(Intrusive Load Monitoring,ILM),which requires senso stalled on each household load to measure the power information of each electrical device.The second is the non-intrusive load monitoring technology.It only needs to install a sensor at the home bus to obtain the total household electricity consumption information,and use the load decomposition algorithm to realize the decomposition of the household load,and obtain the energy consumption information and operating status of each device.The research on the non-intrusi-ve decomposition of residential electricity load based on machine learning proposed in this paper mainly includes the following contents:(1)Noise reduction processing on the collected electrical signals.When collecting household load data,the collected power data is doped with various noises due to various reasons during the collection process,and the noise will affect load identification and decomposition.This paper uses wavelet packet to denoise the steadystate signal,and mathematical morphology filter to denoise the household load switching transient signal.After a large number of experiments,it is shown that the selected filtering method can achieve a better filtering effect.(2)Non-intrusive power load identification in the steady-state process.The steadystate process defined in this paper is that the household load does not undergo switching between on and off states,and there is no switching event,and it operates stably for a period of time.An unsupervised machine learning algorithm is used for the power characteristics of the steady-state process,and the Fuzzy C-Means(FCM)improved by the gray wolf algorithm is mainly used to cluster the power equipment to obtain the clustering centers and the clustering centers of various loads.Label,and finally use the maximum and minimum closeness to identify the electrical load.(3)Non-intrusive household load identification based on event detection for transient processes.The transient process defined in this paper is the transition from one steady-state process to another steady-state process,and the event detection method is used to locate the load switching process and determine the moment of load switching.The improved morphological filter is used to locate the load switching events,extract the power equipment characteristics of the transient process,and use the whale algorithm to improve the supervised machine learning algorithm of the support vector machine to identify the extracted transient load.(4)Use factor hidden Markov model to decompose the total load.This paper uses the mean shift algorithm without setting the number of clustering categories to establish a Hidden Markov Model(HMM)for each single household appliance.For the case of decomposing multiple loads,in order to reduce the computational complexity,the factorial Hidden Markov Model(FHMM)is used to decompose the total load current,and the total load current decomposition problem is transformed into a problem of finding the maximum probability optimization.The improved Viterbi algorithm optimizes FHMM,reduces the scope of the solution space,and speeds up the efficiency of load decomposition calculation.The calculation example proves that under the premise of ensuring the decomposition effect,the calculation efficiency of decomposition is effectively accelerated.