Studying the adequacy of the neural network level controller in the automated control system of an evaporator

Abstract

The aim of the study is to substantiate the use of neural network control of beet juice level in an evaporator by evaluating the accuracy and adequacy of the model. This allows us to assess how well the model properties describe the course of the real process. The use of mathematical statistics methods is the most common way to test models for adequacy. In the automation scheme of level control, capacitive level gauges are used as a sensor. The actuators are pneumatic seat valves with a built-in throttle and an electropneumatic converter. The use of neural network controllers is found only in some specific cases of intelligent control of the evaporation process, and there are no data comparing the use of intelligent controllers with classical ones. In this paper, the Durbin-Watson d-criterion is used to assess the adequacy of the model. Statistical data on the behavior of the level control system circuits in different operating modes using intelligent and classical controllers were collected and a model of the evaporator unit operation was built. The advantage of the Durbin-Watson criterion is its simple and fast implementation, which does not require large economic and energy costs. The accuracy of the model was also evaluated. The static error of the control quality for the levels in the five enclosures of 25% - 65% (in 10% increments) is within the range of no more than 0.2%. The proposed model of the evaporator station operation is generally characterized by high accuracy.