Non-intrusive Load Identification System Development

Electricity,as a kind of clean,pollution-free,high-efficiency energy source,has become more and more important in China’s energy structure in recent years.Monitoring the energy consumption and status of electrical appliances in buildings,while formulating corresponding energy-saving strategies can not only improve energy efficiency,bring huge economic benefits,but also detect electrical anomalies timely and reduce accidents caused by electricity use.Therefore,strengthening the monitoring of power load is of great significance to society and individuals.At present,mainstream load monitoring methods fall into two categories,what are cost-effective but relatively easy intrusive method and low-cost but difficult non-intrusive method.Scholars’ research is most focused on the latter.Although non-intrusive load monitoring has experienced more than 30 years of development since it was proposed,the original methods have not been effective in practical applications due to the increasing variety of electrical appliances.Besides,about the whole set of non-intrusive load identification system,little research has been made.Firstly,through comprehensively analyzing the positive and negative aspects of existing methods,harmonics is taken as the research object.Based on harmonics,a load decomposition and event detection method is proposed.Secondly,apply decision tree and logistic regression for load identification,and test accuracy of the two algorithms on test set.Finally,some solutions to the key performance of system are designed.The specific research contents are shown as follows:1.The existing non-intrusive load monitoring methods are in-depth researched.The pros and cons of load characteristics are analysed in practical applications.Considering the difficulty of load decomposition in the time domain,current harmonics is selected as the research Object.2.System requirements are analyzed,while functional requirements and performance requirements are proposed as the targets for later research.An overall architecture of the system is designed and clarify technical difficulties.3.A soft threshold wavelet transform is used to filter and denoise the current waveform,and Discrete Fourier Transform is used to extract the current harmonics.The current harmonics is stackable is demonstrated,what is based on to realize the load decomposition.According to the fact that the current statistic changes when the load is switched,an event detection method is proposed.4.Training sets is built to train the decision tree and the logistic regression model.The test set is built to test the accuracy of the two models.Compare model in terms of training speed and accuracy.5.Based on the demand analysis the corresponding solutions for the critical system performance are proposed.After testing and comparing each scheme,the optimal scheme is selected to build the system,and finally the system performance is greatly improved.