Research On Household Non-invasive Load Monitoring Algorithm

Energy Internet of Things technology is a new global energy strategy at nowadays.How to integrate the power grid with other energy networks will be the core issue of energy Internet.Therefore,the construction of smart power grid should be the foundation and key of energy Internet system.In this paper,the academic researches are carried out on the non-intrusive load monitoring algorithm of household users’s,which belongs to technical field of the non-intrusive load monitoring(NILM),where is located in the power distribution junction of power grid.The real-time electricity consumption information of each electrical appliance inside the power user can be obtained by the NILM system,that the power data can be collected and analyzed,no matter when the user load can be adjusted and monitored to meet the requirements and standards of the power grid demand-side management.The first work in this paper is that performs median filtering on the measured electrical signals of the home entrance ports,which initially eliminates the influence of data redundancy and noise on load identification.Subsequently,whether the load operating state has changed by event detection,when a variety of load difference characteristics are extracted at the same time,which including active and reactive power characteristics,and also steady-state voltage-current trajectory(Ⅵ trajectory)characteristics that the V-I trajectory characteristic is determined by the steady-state voltage-current Power waveform,which characterizes the change law of load current with voltage.Then,Fourier changes are used to extract the current and voltage high-order harmonic characteristics and harmonic distortion rate to characterize the load power characteristics.The paper also quantifies these characteristics and construct a load feature library preparing for home user load identification.In the case that the load type is difficult to predict in the actual electricity scene,this paper studies the traditional fuzzy c-means(FCM)algorithm that can accurately identify the types with large differences in load.However,it cannot accurately distinguish low-power household appliances.Aiming at the shortcomings of the traditional FCM algorithm,this paper proposes a non-invasive method of firefly fuzzy C-means clustering based on Kernel Entropy Component Analysis(KECA)Load identification algorithm.Three indicators are given which namely the total load identification accuracy,the single-type electrical appliance identification accuracy and the algorithm iteration rate to evaluate the load identification.The calculation Experimental analysis verifies the necessity of reasonable selection of load characteristics for NILM.the result shows that the combination of FCM,KECA and Firefly algorithms(FA)can achieve good and solve overlapping load characteristics for low-power loads in different scenarios.In order to satisfy NILM’s accuracy and effectiveness and it also to solve the problem of using single feature to optimize in the load decomposition,the contribution of this paper is that a multi-dimensional load feature identification algorithm named Multi-Dimensional NILM is proposed.The main idea is to establish multi-objective genetic optimization function based on low-frequency power features,which Non-dominated Sorting Genetic Algorithm Ⅱ(NSGA-Ⅱ)with elite strategy is used to solve the minimizing objective function.In addition,this paper uses a penalty function in NSGA-Ⅱ to weaken the mutual influence of load decomposition appliances.The presented algorithms can operate in real time using low sampling rates without training the system.The proposed method shows good performance in standard measures when tested on the popular AMpds and a private datasets.