Research On Non-intrusive Load Monitoring Technology For Edge Computing Framework

Non-intrusive load monitoring technology can help users understand their own power consumption behavior without infringing on users’ privacy,and help power companies guide users to reasonably respond to demand-side management in order to effectively exploit the potential of interaction.With the construction and promotion of the Ubiquitous Electric Internet of Things,users will interact with the power grid more friendly and closely.Therefore,non-intrusive load monitoring technology will get a wider range of applications in the future.In order to improve the accuracy and scalability of this technology,this paper conducted a research on non-intrusive load monitoring technology for edge computing framework,and the research was carried out from the perspective of the technology itself and the technical application methods.The specific work is as follows:(1)The technology foundation of non-intrusive load monitoring for edge computing framework was researched.Firstly,based on the features of edge computing framework,the technical scheme of non-intrusive load monitoring was elaborated;Then,different kinds of load equipment were analyzed and classified;Next,a test platform for data collection and algorithm verification was constructed;Finally,some evaluation indicators for the implementation effect of non-intrusive load monitoring technology were designed.(2)A non-intrusive load event detection method based on Bayesian iteration was proposed.Firstly,this methodology built an event detection model based on the posterior probability of the observed variables;Then,the detection model was solved based on the Bayesian iterative algorithm;Finally,the solution speed was accelerated in consideration of the application situation of load monitoring.(3)A non-intrusive load disaggregation method based on equipment operating states mining was proposed.Firstly,this methodology extracted the power features from the load event detection results,and obtained clusters which represent different types of load events based on the Mean-shift clustering algorithm in the feature plane.Then,the effective equipment operating states were mined by introducing ZLSC constraints between the clusters and using the GSP algorithm to mine load event pairs from power and time dimensions.Finally,equipment identification was conducted by comparing the operating status mined with the equipment templates stored in the database to complete the load decomposition.(4)A non-intrusive load monitoring technology application method for edge computing framework was proposed.Firstly,the methodology constructed an edge computing model for nonintrusive load monitoring task from regional users,based on the existing low-voltage power information collection system.In this model,non-intrusive load monitoring task for each user could be flexibly executed at local or edge nodes.Then,the cost of task execution corresponding to allocation strategy was analyzed.Finally,based on the deep neural network,a task allocation strategy was generated in order to optimize the cost of task execution for all users in the entire region.