Non-intrusive Load Monitoring Method Research Based On The Neural Network Pattern Recognition

As a significant part of micro-grid,residential buildings contribute significantly to electricity energy consumption.Residential electricity saving plays an important role in promoting energy saving and emission reduction of the whole society and relieving the energy crisis.A perfect residential buildings energy monitoring system(REMS)is the foundation of carrying out energy demand optimization work,it can tell the residents the household appliances that are turned-on within a given period of time and their individual energy consumption,improving users’ enthusiastic on participating in demand response(DR),developing a reasonable energy saving plan and target to buy energy-saving equipment,thereby reducing energy consumption and electricity costs.Therefore,to achieve household electricity visualization services is of great realistic significance in improving the energy efficiency in China,carrying out the sustainable energy development,building a conservation-minded society,easing energy pressure and so on.However,the smart meters in the market currently can only provide the energy consumption information of a whole family,they cannot tell users a specific electricity information,and guide them to use appliances scientifically and reasonably.For this purpose,non-intrusive load monitoring(NILM)technique is proposed,it just need to employ monitoring algorithms into the existed smart meter or the computer software.On the basis of summarizing the previous studies,this paper puts forward an intelligent NILM system,and presents a detailed analysis of the accuracy and effectiveness of the proposed method.Firstly,this paper explains the research background and significance of NILM technology and introduces the basic concept of NILM,which provides a solid foundation for the proposed NILM technique.On the other hand,the research status and bottlenecks of NILM technique are presented,including the high sampling frequency of the power monitoring equipment,low decomposition rate,and difficult to deal with the conditions which the same kind of appliances used at the same time.Based on these conditions,the proposed method focus on utilizing single or very few measurement parameters obtained at a low sampling rate(under 1HZ),thus to realize load decomposition.Secondly,the theory of neural network pattern recognition(NNPR)model is proposed,including mathematical foundation,structure and development of NNPR.By comparing with the traditional pattern recognition scheme,the advantages of the NNPR scheme are clarified.Thirdly,the proposed method in this paper is based on neural network pattern recognition,it not only takes into account the characteristics of the load itself(i.e.current and active power),but also makes full use of the information in the historical data(i.e.the time information).By changing the number of nodes in the output layer of the NNPR model,the problem of recognizing multi-state appliances effectively is solved.Then,the concept of current density curve is proposed depend on the historical data,which determines the number of load states and the current amplitude of each operation mode.Moreover,all of the historical data can be labeled.At last,we present a novel computational formula to estimate the energy consumption of individual appliance.Experimental results indicate that the proposed method provides a very high identification accuracy(beyond 90%).Besides,the algorithm validation results show that,for the vast majority of household appliances,the proposed method can get a higher identification accuracy,compared with the method that do not take into account the time coefficient.