A.V. Terentyeva

Neurocontroller-building technique of air conditioning system management in hardware containers constructed by means of fuzzy logic methods is considered. Development of control model has been performed using MATLAB Fuzzy Logic Toolbox. Fuzzy algorithm processing is the following: fuzzification of input variables; application of fuzzy operator «AND» in the precondition; implication of the precondition and consequence; aggregation of consequences through rules and defuzzification. The results of these steps are shown in the figures. Figures 1 and 2 demonstrate the membership functions graphs for terms of input linguistic variables «Difference of temperatures» and «Rate of temperature change». Figure 3 shows the implementation of the operator «AND» for two input values: the temperature difference - 5 K and the rate of temperature change - 0.05 K/min. In figure 4 the result of fuzzy output sets association for each rule is represented. To improve the accuracy of output result, careful tuning of the membership functions is required, which is a quite laborious process. Therefore, for simplicity the linguistic information on control object is represented in the form of a special fuzzy-neural network. As a result, the fuzzy-neural network structure for membership functions tuning of input and output parameters was defined. In Fig. 8 the structure of the generated fuzzy-neural network is shown. It should be noted that as a result of fuzzy-neural network training there is a membership functions modification and thus improving the accuracy of the output results. It is necessary to stress that described principles of hybrid intelligent technologies allow us to configure effectively control algorithms of air conditioning system, which leads to: - temperature change according to medico-technical rates; - reduce of fluctuations of served room temperature; - minimization of the time-to-set mode; - reduce of energy consumption. The following stage in using intelligent techniques elements in management of air conditioning systems used in complex patterns of weapons and defense technology, is a software development based on the proposed method, and hardware implementation of neurocontroller. Besides, the extension of fuzzy knowledge matrix to increase of fuzzy-neural network learning efficiency is required.