Non-Intrusive Residential Electricity Load Decomposition And Analysis Of User’s Electricity Consumption Behavior

With the development of the energy internet,users are not only eager to get reliable power supply,but also expect personalized energy-saving suggestions,which brings higher requirements to the grid companies’ refined and comprehensive energy services.Existing electricity collection devices and systems cannot meet the above changes,and need to be intelligently transformed to collect and identify the types of indoor electrical appliances and their power consumption.Considering the feasibility and cost of deployment and maintenance,non-intrusive power load decomposition is the only feasible intelligent power collection and transformation plan,which has been widely approved in the industry.On this basis,modeling and categorizing the user’s electricity consumption behavior is the key to the refinement of on-demand electricity service customization,which is of great significance to user energy saving,and can indirectly reduce grid line losses,improve open capacity and grid reliability To optimize power grid construction and dispatch.However,the traditional load decomposition method is realized by establishing a huge feature database and using offline data.When faced with a huge feature database calculation,these methods cannot meet the requirements of running in a real-time analysis system environment,and there are load decomposition and event detection accuracy is not high,etc.problem.In recent years,the use of deep learning methods to achieve non-intrusive load decomposition has become a hot spot,using features based on convolutional neural network structure extraction and sequence-to-point structure based on sliding windows to reduce the amount of calculation,and using sequence-to-point neural based on multidimensional convolution The network model performs nonintrusive load decomposition,collects long-term sequence total load data,uses a multi-dimensional convolution structure for feature extraction,and uses a sequenceto-point structure for data calculation,which can perform electrical appliance characteristics while ensuring load decomposition accuracy The automatic extraction greatly reduces the calculation time of the load decomposition model applied in the real-time analysis system.Using public data sets to test and verify the performance of the algorithm model,compared with other algorithms used in non-intrusive load decomposition,the recognition rate is higher.In addition,the algorithm model has also achieved good performance in the actual test of residential electricity consumption environment.After realizing the non-intrusive load decomposition research,analyze and mine the electrical load data obtained by the load decomposition to obtain information on the electricity consumption behavior of residential users,and classify the electricity consumption behavior of the users.Application requirements,and finally design and implement the software module of the user’s electricity non-intrusive load monitoring system.