Research On Building Power Management System Based On No-intrusive Load Monitoring

With the continuous construction and development of smart grids,the concept of intelligent power consumption has gradually penetrated into millions of households.The intelligent power management system can not only guide users to reduce unnecessary energy consumption,but the management and power supply departments can also use it to predict and control the load in the area to achieve the purpose of " peak reduction and valley filling".To this end,this thesis takes residential buildings as the research object,and proposes an implementation method of building power management system based on Advanced Metering Infrastructure architecture.The collection and analysis of user power consumption data are completed according to the non-intrusive load monitoring technology,and the load is classified and identified by deep learning network.The non-dominated sorting differential evolution algorithm is used to optimize the user’s power consumption behavior and provide the user with a choice of optimization scheme.Using the concept of hierarchical management,the functional requirements of the power management system are deeply analyzed.Unified Modeling Language(UML)is also used to analyze and construct the system architecture.The main functions of the system include providing reasonable power consumption optimization suggestions for users and real-time monitoring whether the three-level power supply and distribution system of the building is in safe,reliable,high quality and economic operation state.The specific work is as following:First,it is proposed to apply the deep learning network to the implementation of nonintrusive load monitoring technology,build a framework with bidirectional GRU and attention mechanism as the core,and conduct experimental verification on the public real dataset REDD.It is mentioned that it solves the problem of low recognition accuracy of current traditional algorithms for multi-state electrical appliances.The system uses non-intrusive load monitoring technology to complete the collection and analysis of user electricity data.Compared with traditional data collection methods,this method is easier to implement and has a lower cost.Secondly,the non-dominated sorting differential evolution algorithm is proposed to optimize the user’s power consumption behavior,and to provide users with reasonable user behavior optimization suggestions.The three objects are optimized on the public dataset REDD,such as typical electrical appliance,typical user’s power consumption data one day and user group’s power consumption data one day.The results show that the non-dominated sorting differential evolution algorithm used in this thesis can reduce the user’s electricity cost while affecting the user’s electricity comfort as little as possible.Then,it analyzes the functions of the system floors and substation floors,and designs the identification of possible abnormalities and bad electricity consumption data,voltage deviation monitoring,and three-phase imbalance monitoring methods that may occur in the system,and proposes corresponding management measures.Finally,the Unified Modeling Language is used to construct the building power management system,which can manage the total power data of the building from the user-level to the substation-level,and realize the intelligence of the building power management.