The thesis research is aimed at the solution of a number of topical issues in Ukraine’s oil industry such as raising the efficiency of oil production units and implementation of intellectual technologies for oil production.
The thesis provides a review of literary sources, which showed that there are currently no automated control systems for oil production on the basis of intellectual wells in Ukraine. This creates the need for research in the area of developing new ways to control electric drives of sucker-rod pumping units. One of the main tasks in designing a control system is to select informative parameters to identify the status of the pumping equipment. The thesis shows that the most appropriate method of recognizing the state of the well and its equipment is the use of neural networks. To determine the optimal structure of the neural network, a series of mathematical experiments were conducted with the real dependencies of the force in the polished rod and the current of the drive motor of the operating oil production facilities. Based on the comparison of the results of the recognition of experimental curves, it is concluded that recurrent neural networks, in particular the Hemming network, are most suitable for the recognition of load curves or time-current curves. The thesis proposes a modification of the Hemming network, in particular, replacing the first layer of the network with the algorithm of bitwise comparison of data arrays. This modification has greatly improved the performance of the neural network and increased the recognition accuracy of the input signal. Based on the proposed modification of the Hemming neural network, an algorithm for its operation was created. It is shown that bringing an input data set to one standard (binary representation) enhances the versatility of the recognition system, enabling it to be used in different control systems where load curves, time-current curves or active power curves are inputs.
The mathematical model of the electromechanical system of the oil production unit developed in the thesis is the basis of an intelligent well control system used to reproduce real processes and to identify the state of the oil production equipment. For conducting mathematical experiments in the MATLAB environment, a model of an electromechanical control system for a sucker-rod pumping unit was created. It is based on high-adequacy mathematical models of frequency-controlled asynchronous electric drive and sucker-rod pump, mathematical model of the neural network; a method of calculating the periodic dependencies of the coordinates of the operation mode of an oil production unit based on solving the boundary value problem. This approach makes it possible to obtain the result of mathematical modeling with high reliability and minimal-volume calculations in the timeless domain, which is important for the development of control systems in real time.
Based on the created model, a number of mathematical experiments were carried out, which confirmed the effectiveness of the proposed modification of the Hamming neural network with respect to the accuracy and speed of recognition of the input arrays of load curves and time-current curves. The example demonstrates that the developed system of sucker-rod pumping unit control can quickly in real time adapt to changing wells operation modes. The thesis presents the modelling of the electromechanical system both in working and emergency modes.
To carry out the experimental studies, a laboratory workbench was created for the electromechanical control system of the sucker-rod pumping unit. The thesis describes the element base and the software used, outlines the schematics of the developed laboratory workbench, the algorithm of operation and the sequence of operations performed during the acquisition of experimental data.
On the basis of the experimental results obtained at the laboratory workbench, it is concluded that the theoretical solutions proposed in the thesis and the application of the developed algorithms are correct.
Key words: sucker-rod pumping unit, asynchronous electric drive, marginal well, uninterrupted operation mode, dynamogram, pattern recognition, neural network, mathematical modeling, control system.
