基于LRTC-TNN的瞬时水流量数据连续插值方法

Continuous Imputation Method of Instantaneous Water Flow Data Based on LRTC-TNN

瞬时水流量数据在采集、整理、存储过程中均存在不同程度的数据缺失问题,不但会造成数据分析上的偏差,还会影响后期决策,尤其是连续水流量缺失问题。国内外关于水流量数据缺失值插补的研究方法很多,然而针对相邻时间存在连续缺失值的插补问题还没有完备的解决方案。因此,基于瞬时水流量数据集的低秩假设,提出一种基于非凸低秩张量补全模型(A Nonconvex Low-Rank Tensor Completion Model-Truncated Nuclear Norm,LRTC-TNN)的瞬时水流量缺失值插补方法。通过乘子交替方向法(Alternating Direction Method of Multipliers,ADMM)求解最优的LRTC-TNN模型。利用通用速率参数自动确定张量模态的截断,运用张量补全的策略对连续缺失值进行预测。将该方法用于某地水厂管道瞬时水流量数据插值实验中并与其它最新的和传统的方法进行对比,取得了非常好的效果。

In the process of collection,sorting and storing instantaneous water flow data,there are different degrees of data missing,which will not only cause deviation in data analysis,but also affect later decision-making,especially the problem of continuous water flow missing.In the research on missing value imputation of water flow data at home and abroad,there is no complete solution to the imputation problem of continuous missing values in adjacent time.Therefore,based on the low rank assumption of the instantaneous water flow data set,we propose a nonconvex low rank tensor completion model-truncated nuclear norm(LRTC-TNN)to impute the missing values in the instantaneous water flow time series data.The optimal LRTC-TNN model is solved by the alternating direction method of multipliers(ADMM),and the truncation of tensor modes is automatically determined by using the general rate parameters.The missing values are predicted by using the tensor completion strategy.The proposed method is applied to the imputation experiment of instantaneous water flow data in a water plant and compared with other latest and traditional methods,which is efficient,especially for the imputation of continuous missing values.