Non-Intrusive Load Identification Based On Group Intelligent Decision Using Structured Graph

A transition from traditional fossil-based generation to clean and energy-efficient generation has already begun.Under this new environment of electric power industry,the balance between energy consumption and power supply needs to be promoted actively.To ensure this dynamic balance,demand-side resource should be allocated scientifically and scheduled flexibly by advanced information technology.Nonintrusive load monitoring(NILM)plays an important role in fully perceiving the load operation of smart grid,which is the premise of flexible scheduling on the demand side.At present,the proposed load identification models of NILM highly dependent on prior data.Meantime,it is necessary to improve the precision of identification and decrease the computing time.To realize high accuracy and fast identification for independent users,this paper proposes a group decision classifier based on the structured load feature graph.Focusing on residential appliances,the completed processes of NILM are accomplished.The main contributions of this article are:(1)Due to the strong random electricity consumption behaviors,diversified modes,and unbalanced distribution of samples,the problem of obtaining special prior data for different users needs to be solved.Inspired by the knowledge graph holding strong power for knowledge representing,a load feature graph that preserves sufficient information is constructed.The variable waveform data is transformed into structured features with the consistency of the same type and the divisibility of different types,which are suitable for load identification and applicable to different users.(2)Focusing on the problem of a single load identification model and alleviating the identification results are affected by feature selection,the convolutional neural networks and combined support vector machines are used to classify images and features respectively.As one of the powerful structures of deep learning,CNN is used for accurate identification of current curves.Besides the unique current waveforms in some types of loads,significant operational characteristics and features in data formats are comparatively common.Combined SVM classifiers are employed to adapt these nonlinear data and classify them.An AdaBoost algorithm is used to adjust the weight of training sets.After that,a logical operating rule is proposed to obtain the final identification result.The identification accuracy is improved.(3)A load identification algorithm which is flexible for various households is required to realize automatic NILM.The wavelet transform model is used to rapidly locate load events based on the total power curve and obtain the waveform of unknown load.Furthermore,to minimize running time,a customized load identity library focusing on the individual user is proposed.Based on the library,the load identification could match templates directly by the genetic algorithm to reduce computational cost.(4)In view of above processes and considering the realizability of the NILM system,an architecture of end-cloud collaboration is proposed.The identification process with high requirements on data sources and computing power is deployed on the cloud-side while the others are deployed on the end-side to make up for the data upload delay.The architecture could achieve continuous and real-time perception of residential power consumption.Finally,a NILM system is developed based on Huawei cloud ModelArts platform.