A Research On Non-invasive Load Monitoring And Decomposition Algorithm For Practical Application

As an important secondary energy source for the survival of society,how to efficiently manage and use electric energy has always been a problem for people.With development of technology,traditional power grid has completed its transformation to smart grid and is gradually moving towards energy internet.Fully understanding the consumer's power consumption behavior is the basis for grid companies to build consumer portraits and formulate demand response policies.Traditional load monitoring is to read the electricity quantity information on the home bus through the electricity meter,but it can?t analyze negative composition inside home.By installing an intelligent terminal on the socket to collect the usage information of each electrical appliance,the equipment-level electricity consumption data can be obtained.However,such a method has high cost,poor operability and low user acceptance.Thus,non-invasive load monitoring through the number of loads on the bus has become a hotspot.This paper researches supervised and unsupervised algorithms.In supervised algorithms,the characteristics of load data are first analyzed to extract low-frequency and high-frequency features.The training load decomposition model of Long Short-Term Memory(LSTM)network is used to analyze the influence of different features on the recognition results.The accuracy of algorithm is verified on public data set and self-collected data set.Experiments show that the ability of LSTM to find time series features fits the scenario very well,and accuracy is obviously improved after adding high-frequency features.In unsupervised algorithms,a load decomposition algorithm is proposed to solve the problem that data labels are difficult to obtain in real scenarios.This algorithm transforms the non-invasive load monitoring problem into a combinatorial optimization problem,and searches for the optimal combination of electrical appliances through the idea of swarm intelligence.The proposed algorithm combines the advantages of existing two swarm intelligence algorithms and has higher optimization accuracy.The unsupervised learming method of the algorithm is also more suitable for popularization and application.Finally,the intelligent energy management system is designed and implemented based on the above algorithm.The system includes many functions such as data acquisition,electrical equipment working status identification and historical energy analysis,which is an exploratory attempt to apply non-invasive load monitoring in practice.In the future,the system will expand functions of demand response capability evaluation and equipment status evaluation,and then the system will become an important link between users and power grid companies.It5s helpful to promote the deep integration of energy network and information network.