SYNTHES ALGORITHM SETTING NEURAL NETWORK REGULATOR AVIATION GAS TURBINE ENGINE

Abstract

One of the promising areas in the field of automatic control is the use of intelligent control systems based on artificial neural networks. The main advantage of this control system is the use of such properties of the neural network as the possibility of approximation of arbitrary non-linear dependencies. The main objective of the research is the synthesis of tuning algorithms for neural network regulators by aviation gas turbine engines. The algorithm of initialization (preliminary training) of the neural network is proposed, based on the analysis of the linearized model of the first approximation of the neural network regulator of an aircraft gas turbine engine. For this, two campaigns were used. The first approach consists in the following assignment of a training set so that the gains of a trained neural network take the specified values, for example, equal to 1. This will avoid strong “bursts” of signals in the system during its further study and achieve the required accuracy of learning the characteristics of the control system. The second approach is to determine the approximate values of the constant values of the inputs and output of the neural network as part of the automatic control system for each of the specified R modes of operation of the system and use these values as a training set.

The application of the proposed algorithm allows you to guarantee the convergence of the learning process of the parameters of the neural network and reduce the total quadratic learning error compared with the procedure of randomly selecting the initial values of the neural network by more than 2 times. The effectiveness of the application of learning algorithms for neural network regulators of aircraft engines was investigated, the advantages of using the simplex search method were proved. The results of the simulation of a control system with a neurosemotic controller along the control loop of the rotational speed of an aviation gas turbine engine have shown the effectiveness of using neural networks in problems of multimode control of aircraft engines.