Search

The paper proposes a method to improve the quality of control of technical objects, the problem of automation of which was solved with the use of the classical control theory and such controllers as PI, PD, PID. The basic idea of the method consists in the gradual replacement of these controllers by their hybrid analogs realized with the use of the fuzzy logic apparatus (fuzzy – PI, fuzzy - PID, etc.). At the first stage, statistical information is collected about the inputs and outputs of the controller used (system error value , error integral θdt, magnitude of the generated control action U). The stage is considered complete when a set of data describing the operation of the controller for all modes of its operation will be created for the automated object. The next step involves the use of the obtained statistical data for synthesizing the base of the control rules of the hybrid controller(s). The solution of this problem is possible in a manual mode, but the process is very labor-consuming, it is associated with a significant risk of errors, manifested in the creation of duplicate or, more critically contradictory, control rules. The identification of these errors may require the developer of great effort and does not guarantee a successful outcome. In this connection, the problem of developing and using specialized software becomes urgent, which was done in this work. As a result, the duration of the developer's time costs decreased by 98%, and the risk of errors was reduced to zero. For ease of use, the result of applying the proposed software is presented in the form of a *.fis file, which can be opened and analyzed in Fuzzy Logic Toolbox. In the future, the software application under consideration can be transferred from the MATLAB modeling environment to any other programming language.

The key purpose of the work is development of a method of control that allows simplifying, automating and unifying the process of design of the hybrid systems which are a basis of modern automation. To achieve a definite purpose, a method of control of a technical object based on the construction of an adaptive system of neuro-fuzzy inference is developed. The objects of the system of neuro-fuzzy inference are the classical and fuzzy models of control. Information exchange between models is provided by means of the developed hybrid control system. The result of the interaction of the two models is the automatic formation of the base of fuzzy controller rules based on knowledge about the control object obtained with its control using the classical controller. In the developed adaptive system of neuro-fuzzy inference signals of error and control in the classical model are used as data for creation a hybrid network. Signals of error and control in a fuzzy mod-el with automatically generated fuzzy inference rules are used as data to verify the created hybrid network in order to detect the fact of its retraining. Thus, during the control of a technical object by means of a hybrid system, the knowledge of an expert in the subject area for adjusting the parameters of the fuzzy controller are completely eliminated, that allows to control difficultly formalizable objects in conditions of uncertainty. To obtain reliable research results, a hybrid control system, consisting of classical and fuzzy models is developed. Numerical values of the error and control signals are obtained at discrete timepoints as a result of the interaction of the two models. Special files for creating and testing a hybrid network in the form of numerical matrixes are generated. The hybrid network is developed in the ANFIS editor of the MATLAB package. The generated structure of the FIS fuzzy inference system of the Sugeno type is graphically shown. The visualization of the dependence of training and verification errors from the number of training cycles is given. The surface of the fuzzy inference system is constructed, that allows estimating the dependence of the output variable on the input.