Research On Non-intrusive Load Decomposition Algorithms For Residential Users

With the development of smart grid,power grid companies are gradually transforming into highly intensive and technological ones.The demand for intelligent power distribution side is increasing,so the intelligent technology of power distribution side is the current research hotspot.As the key technology of intelligent power utilization,load monitoring can help power grid companies to obtain users’electricity information effectively and promote friendly interaction between users and power grid companies.Load monitoring technology mainly uses intrusive method,that is installing data acquisition devices in electrical equipment.This method has high economic costs and low practicability.Non-intrusive load monitoring(NILM)technology only installs data acquisition equipments in the electricity meter at the entrance.It decomposes the total power load into electrical information of separate appliance by analyzing the voltage and total current,which can obtain the energy consumption of users.Load decomposition method is an important part of NILM.In this thesis,NILM methods are studied from steady and transient processes.Firstly,the basic NILM framework is given.Load characteristics of typical residential appliances are analyzed from steady and transient aspects.The load decomposition algorithms are introduced from mathematical optimization and pattern recognition.The application scenario of steady and transient NILM methods are analyzed.Secondly,a steady NILM method based on dynamic adaptive particle swarm optimization(DAPSO)is proposed.Based on traditional power characteristics,total harmonic distortion coefficient is regarded as a new load feature.The inertia weight is adjusted dynamically and adaptively.DAPSO algorithm is used to decompose load data in different noise scenarios.Experimental results show that DAPSO algorithm improves the effectiveness and robustness.Finally,a transient NILM method based on multi-scale wavelet packet optimization is proposed.Transient load feature library based on active power is established and optimal wavelet basis function is determined by wavelet packet energy entropy.Multi-scale wavelet packet transform is used to map the transient load feature library into wavelet domain energy space and form standardized template.Load decomposition is analyzed according to the matching degree between load data and standardized template.Experimental results show compared with the steady NILM methods,transient methods can effectively solve feature overlap problems.