Topology and parameter optimization of ANN using genetic algorithm for application of textiles

The BP feed-forward neural network is popular in solving many non-linear multivariate and complex problems. The most important problem with neural network is to decide optimal structure and parameter settings. Literature presents a multitude of methods but there is no rigorous and accurate analytical method. This paper presents the hybrid approach of genetic algorithm and neural network computing for establishment of the optimum number of neurons on layers, transfer functions, learning rate, momentum and number of epochs for a given problem. The method can be used without restrictions to model a network with many inputs and outputs. The process involves GA evolving several structures, different parameters and fitness level of each structure. GA decides fitness using neural network as the fitness function. This technique can help to eliminate trial and error work for deciding the optimal network. The proposed Neuron-Genetic classifier has been successfully applied for prediction of yarn properties in spinning process of Textile industry.