Application of genetic algorithm to configure artificial neural network for processing a vector multisensor array signal

The possibility of applying genetic algorithms to configure a topology of artificial neural network which process a multisensor array vector signal for gas-analytical tasks has been considered. Such a configuration is implemented according to the criteria of increasing the percentage of correct recognition and reducing the computing cost by using a set of predefined possible values of the parameters of the neural network architecture. The optimized characteristics are the number of hidden layers of neurons in each layer and the amount of training sampling. The tests have been performed in the ©Matlab neural network operated under Levenberg-Marquardt back-propagation learning function. The obtained results confirm the efficiency of the genetic algorithm to optimize the topology of designed artificial neural network with advanced characteristics.