INCREMENTAL SYNTHESIS OF GROWING MODULAR NEURAL NETWORK FOR CHP-PLANT SUPPLY WATER TEMPERATURE CONTROLLER

Abstract

The paper considers the use of growing modular neural networks for incremental synthesis of the neurocontroller of supply water temperature at CHP-plant. There has been proposed the architecture of the growing modular neural networks on the basis of a three-layer perceptron, allowing the network modules training using genetic algorithm. For test problem it is shown that the training time of growing neural network reduced and its accuracy increased compared to a fixed architecture neural network. The problem of CHP–plant supply water temperature neurocontroller synthesis that provides a reference daily heat output and stable hourly temperature of the return water is successfully solved on the basis of the proposed type of growing network.