Key Technologies For Autonomous Non-intrusive Load Monitoring

Load consumption details monitoring can acquire the real-time electricity consumption information of each(main)appliance inside the power users' premise.It is a technical component of advanced metering infrastructure of smart grid,and is located in between electricity distribution and power ultilization,thus has great engineering significance.Non-intrusive load monitoring(NILM)can realize load consumption details monitoring by analyzing the aggregate load data,which has the advantages of low cost,easy implementation,high reliability and user acceptability.For any given unseen scenarios,the autonomous NILM can adaptively establish and update the personalized load signature database of appliances without any entry(to meter the consumption data of individual appliances)or appliance survey based on prior knowledge about the operation characteristics of appliances,thus realizes NILM.In this regard,the following studies have been carried out:(1)In order to identify the unseen appliances(working states thereof),a method of fully unsupervised non-intrusively adaptive modeling of appliance load signature(abbreviated as Adaptive Modeling)is proposed for the first time.Among which,the sequential pattern mining technique is firstly employed to establish the complete working cycles of appliances,the “Load Event Association Rule” is proposed to establish the association between different working cycles of the same appliance,and an algorithm of incremental appliance finite state machine(FSM)model topology generation is presented to automatically establish and dynamically update the working state set of the appliance and the switching relationship between its different working states,based on which,the desired appliance load signature can be acquired.Test results show that the proposed method can automatically establish and dynamically update the load signature set of various types of unknown appliances only using the aggregate load data,thus filling an international technical gap in the field of NILM.(2)To establish a complete load signature database of appliances,it is also necessary to use the identification results of the modeled unseen appliance to automatically assign it the physical name(abbreviated as Automatic Naming),based on prior knowledge about the operation characteristics of appliances.For this purpose,a novel automatic appliance model naming method integrating the non-parametric characteristics is proposed.Among which,the concept of non-parametric operating characteristics of appliances is proposed for the first time,and the algorithms for distinguishing the two non-parametric characteristics of “fixed duration” and “periodic operation” are presented.Furthermore,based on the definition of non-parametric appliance set,a twolayer decision-making method combining the non-parametric characteristics for automatic naming is established.Test results show that the proposed method can effectively overcome the problem of the weak generalization ability of parametric characteristics,and thus improve the accuracy and robustness of appliance naming,and reduce the time required for inferencing.(3)In order to optimize the monitoring performance of the modeled appliances with the load signature given,a novel non-intrusive appliance working state identification method based on dynamic time warping(DTW)algorithm is proposed.Among which,the DTW algorithm is adopted for the first time to measure the similarity between the variable-length transient power waveform(TPW)sample and template time-series,and the DTW-based integrated distance combining multiple types of TPW signatures is defined.Test results show that the proposed method can not only improve the accuracy and robustness of appliance working state identification with the load signature given,but also is easy to implement at reasonable cost.(4)An incremental learning based fully autonomous NILM framework is proposed,and the corresponding implementation architecture is designed.Based on that,the hardware and software system developed based on this paper have been successfully deployed in fields.The field results show that the system can run autonomously for any given unseen scenarios,adaptively establish and update the personalized load signature database of appliances,and monitor the working states of the modeled appliances in near-real time.