Control of Non-Standard Dynamic Objects With the Method of Adaptation According to the Misalignment Based on Neural Networks

The article discusses the procedure for building a neural network control system for dynamic objects. The methods of continuous adaptation of a neural network controller with neural network adjustment to changes in the dynamic characteristics of an object and adaptation to a mismatch signal are compared. To eliminate the shortcomings of the first method, the control loop is supplemented with a block for detecting changes in the state of the control object, designed to detect the signal of mismatch with the "set point". In order to ensure the reliability of detection of the mismatch signal, paired response is taken into account within the calculated value of the average delay time. The neural network model of the control object is adjusted outside the control loop. To detect mismatch in a timely manner, an algorithm of cumulative sums is configured, in which the defining characteristics are the average delay time and the average time between false positives.

Перевод пока недоступен