The application of neural networks is the main element in the recognition of images, classification and prediction of space-time sequences in solving a wide range of problems in many industries.
At the same time, most modern works on the classification of time sequences are focused on one-dimensional structures. Within in this paper, transformations from the one-dimensional to the two-dimensional structures of the original time series representations were used to solve the recognition problems. Implementation of this method in the field of diagnosing the operation of pumping equipment allows for better recognition of spatial structures and training of neural networks with a small number of initial data (images).
The aim of the work is to improve the efficiency of determining the technical condition of rod pumps during the operation by dynamometry. A comprehensive approach to the interpretation of rod pumps dynamogram cards was proposed and described. Using the encoding of dynamogram card in various types of images, the optimal methods for their representation are established. The following ways of dynamogram representation were analyzed: initial representation (image), plot in polar coordinates, recurrent diagrams and cross-correlation matrix with sequential delays. Various methods of computer vision were used for classification problems of dynamometers. The complex analysis carried out made it possible to establish the most optimal approach to the presentation of dynamogram cards based on the accuracy of recognition.
As a result of the work carried out, methods for presenting data were shown that showed high classification accuracy and a low level of learning error in small samples of the original data, i.e. these representations provide the best way of «isolating» the topological features of the original dynamogram cards (among the methods compared). Due to the fact that often deep-pumping equipment is operated in downhole conditions with the presence of several types of complications, a new architecture of the classifier for diagnosing dynamogram cards operation was proposed. In it, it is possible to implement complex diagnostics of the equipment condition taking into account all known technological factors (complications) that affect the operation of the equipment.
References
1. Valiakhmetov R.I., Yamaliev V.U., Shubin S.S., Alferov A.V., Application of heuristic algorithms in analyzing data to solve the problem of detection of electric centrifugal pumping units (In Russ.), Izvestiya Tomskogo politekhnicheskogo universiteta. Inzhiniring georesursov, 2018, V. 329, no. 2, pp. 159–167.
2. Takhautdinov Sh.F., Farkhullin R.K., Muslimov R.Kh. et al., Processing of practical dynamometers on a PC (In Russ.), Kazan': Novoe Znanie Publ., 1997, 76 p.
3. RD 39-1-9-98-84, Metodika diagnostirovaniya i optimizatsii rezhimov raboty ustanovok ShGN po dinamograficheskim issledovaniyam (Methods of diagnosing and optimizing operating modes SRP units by dynamographic studies), Shevchenko: KazNIPIneft', 1984, 101p.
4. LeCun Y. et al., Backpropagation applied to handwritten zip code recognition, Neural computation, 1989, V. 1, no. 4, pp. 541 – 551.
5. Krizhevsky A., Sutskever I., Hinton G.E., Imagenet classification with deep convolutional neural networks, In: Advances in neural information processing systems, 2012, pp. 1097–1105.
6. Xia X., Xu C., Nan B., Inception‒v3 for flower classification, Proceedings of 2nd International Conference on Image, Vision and Computing, 2017, pp. 783–787.
7. West J., Ventura D., Warnick S., Spring research presentation: A theoretical foundation for inductive transfer, Brigham Young University, College of Physical and Mathematical Sciences, 2007.
8. Wang Z., Oates T., Encoding time series as images for visual inspection and classification using tiled convolutional neural networks, Proceedings of Workshops at the Twenty-Ninth AAAI Conference on Artificial Intelligence, 2015, Part 1.
9. Wang Z., Yan W., Oates T., Time series classification from scratch with deep neural networks: A strong baseline, Neural Networks (IJCNN), 2017 International Joint Conference on, IEEE, 2017, pp. 1578–1585.
10. Eckmann J. P., Kamphorst S. O., Ruelle D., Recurrence plots of dynamical systems, EPL (Europhysics Letters), 1987, V. 4, no. 9, p. 973.
11. Harikrishnan K.P. et al., Determining the minimum embedding dimension for state space reconstruction through recurrence networks, URL: https://arxiv.org/pdf/1704.08585.pdf
12. Krämer K.H. et al., Dimension-scalable recurrence threshold estimation, URL: https://arxiv.org/pdf/1802.01605.pdf.
13. Yamaliev V.U., Salakhov T.R., Shubin S.S., Primenenie elementov teorii determinirovannogo khaosa k resheniyu zadach tekhnicheskogo diagnostirovaniya UETsN (In Russ.), Elektronnyy nauchnyy zhurnal Neftegazovoe delo = , 2014, no. 4, pp. 174–191.
14. Kang W.Y., Park K.W., Zhang B.T., Extremely sparse deep learning using inception modules with dropfilters, Proceedings of 14th IAPR International Conference on Document Analysis and Recognition (ICDAR), IEEE, 2017, pp. 448–453.
15. Akbar S. et al., The transition module: a method for preventing overfitting in convolutional neural networks, Computer Methods in Biomechanics and Bio-medical Engineering: Imaging & Visualization, 2018, р. 1‒6.
