Goal

Determine the operating modes of the well, based on the data of temperature and pressure taken throughout the mine.

Our solution

We developed an autoencoder based on the neural network on the PyTorch engine. It is capable of restoring the input signals on the output, comparing the restored result with the original one. Thus, it allows to receive information about deviations in the equipment operation.

In this case, the data on the mine temperature change during a certain time interval were used as the initial data.

The visual representation below shows that the temperature fluctuations have a different character throughout the timeline.

We trained the autoencoder with the use of data from a short section of the graph. Automatic clustering has shown that the temperature values can be combined according to a certain feature.

Based on the obtained clusters, we plotted the deviations of the well from the ‘normal’ state selected in advance (the section chosen for training the neural network).

Automatic clustering identifies the main groups of deviation vectors that correspond to the operating modes of the mine.

The same graph for the entire section chosen for study.

As the result, we determined three main operating modes of the mine. They correspond to the well operation in the water pumping mode, in the absence of water mode and in the mode of filling with water.