Effect of the Data Tensorization on the Recovery of Bursts of Missing Values. An Application in Water Networks

摘要

The SCADA systems capture a huge quantity of data from different devices. In order to analyze the historical data collected by a specific sensor, in some cases, it is necessary to restore the lost or discarded data. When we deal with a large amount of unknown consecutive samples, this task is more complicated. The study is focused on reconstructing bursts of lost samples of a water reservoir level meter. We prove that the tensors can be a useful mathematical tool to do this function. We know that it is possible to improve the data reconstructions realized with linear methods by applying tensorization techniques. To do this, it is necessary to organize the data in the tensor, searching to take the maximum advantage of the signal periodicity on different levels. In this work, it is verified that reordering the tensor according to the position of the lost samples, that must be recovered, affects the result of the reconstruction. So that, we purpose an optimal tensor ordering for the bursts restoration.