Neural Network Algorithm for Adjusting the PI Controller in the Shearer Control System

Abstract

The control system of a shearer is considered, which is designed for destroying rock and loading it onto a scraper conveyor. When working out a coal seam by a shearer, external disturbances — the coal resistance to cutting, solid inclu- sions of rock, changes in the width of the screw, which vary indefinitely, lead to a deterioration in the quality of transients. The paper focuses on the shearer control system, the key elements of which are: a movement drive, a cutting drive, a coal face and a standard regulator that provides the system with the desired control quality indicators. A typical cutting current controller in the form of a PI controller with parameters configured for a specific mode of operation of the shearer cannot ensure the optimal functioning of the control system in all modes due to the non-linearity of the control object and the random of changes in the coal resistance to cutting. To improve the control quality indicators, it is necessary to choose the parameters of the PI controller so as to minimize the amplitudes of the current steps of the cutting motor, and therefore reduce the amplitudes of the moment in the transmission of the cutting drive and minimize the system quieting time. In this paper, we propose a tuning algorithm based on obtaining the values of the controller parameters for each of the possible modes of operation of the shearer, identifying the type of disturbing effect by the response curves of the system available to observation. At the same time, the use of an artificial feed forward neural network, acting as an operational means of recognizing a multidimensional response curve in the control loop, is proposed. A neural network of two architectures was used: with a scalar and a vector output function. The curve recognition algorithm satisfies the limitations on the speed of solving the problem of controlling an electromechanical system, since the recognition of a disturbance occurs in a time that does not exceed the output of the process to the maximum of the current step. The correctness of the obtained results was confirmed by the results of computer modeling.