High performance load pattern identification in the scenario of high proportion data missing

This paper presents an effective identification method of load pattern in the case of high proportion data missing (HPDM). Data preprocessing generally discards quantities of missing data, which leads to the waste of the information. Traditional missing value completion methods such like nearest neighbor interpolation (NNI) and cubic spline interpolation (CSI) are difficult to sufficiently utilize the information redundancy and correlation of load data set, so they cannot cope with the problem of HPDM well. Hence, this paper presents low rank matrix completion (LRMC) method for HPDM, which can make full use of the approximate low rank characteristic of the load data matrix to rebuild the lost data segments. Meanwhile, Bi-directional long short-term memory (Bi-LSTM) neural network which is an excellent feature extraction model for time series data, is used for load pattern classification. At last, the experimental results based on UCI load data set validate the performance of the presented method on HPDM.