Research On The Tensor Completion Method Of Missing Data In Low-voltage Area Network

The user power data of low-voltage area network is the basis of line loss calculation of low-voltage area network,load forecasting,non-intrusive load identification and other applications.However,in the process of collection and transmission,the user data often appears to be missing,which becomes the bottleneck of subsequent applications.In order to break the bottleneck,according to the characteristics of user data in low-voltage area network,we proposed a tensor-based high-precision data completion method.(1)First of all,we analyzed the reason of data missing in the low-voltage area network in detail,and summarized the classic data missing mode.Then,from the perspective of relative value,absolute value and volatility of data,the evaluation index of missing data completion performance is proposed.At the same time,the theoretical basis of tensor is introduced.(2)Aiming at the problem of single measurement missing data completion in the lowvoltage area network,firstly,the user space correlation,the periodic and timely sequence characteristics of single measurement data in low-voltage area network are analyzed by taking the current data as an example.The third-order tensor model of current data in lowvoltage area network is formed by user-days-time interval.In order to extract the highorder correlation contained in multi-user current data,a low-rank tensor completion model is established by combining tensor model with singular value decomposition.Alternating Direction Method of Multipliers(ADMM)is used for optimization solution iteratively.Finally,the simulation example is analyzed by using the actual multi-user current data of a low-voltage area network.The result shows that the proposed method not only can complete the random missing data and the all-day missing data simultaneously,but also has smaller completion error compared with the cubic interpolation,Kalman filter and matrix completion methods.(3)Aiming at the problem of multiple quantity measurement missing data completion in the low-voltage area network,firstly,we analyze the characteristics of multiple quantity measurement data which different from the feature of single measurement data.Considering the basic law of circuits,it not only contains the characteristics of the single measurement data,but also includes the correlation between multiple measurements and the coupling relationship between various characteristics.Therefore,the data structure of the low-voltage area network is the high-order coupling data structure of spatial-timequantity measurement.In order to extract the high-order coupling correlation of multimeasure data,a three-order multi-measure data tensor model was constructed with three characteristic orders of measurement-time-user.Then,combined with tensor theory and singular value decomposition,a tensor completion model of missing data of multi-user and multi-quantity measurements is established,and the model is solved iteratively by using the alternating direction multiplier method.Finally,the method is used to complete the missing data of multiple measurements by analyzing the simulation examples in two cases of random missing data and the whole day missing data.The result shows that this method is more accurate than the single measurement method.