Integrated Genetic Neural Networks and its Application in the Worsted Fore-Spinning Process

The characteristic of worsted fore-spinning process and BP neural network modeling technology all have been summarily analyzed. In order to overcome the problems with slow rate of convergence and falling easily into part minimums in BP algorithm, a new improved genetic BP algorithm was put forward to model the fore-spinning process. Genetic algorithm was used to optimize the weight and threshold matrix of the established neural network. Using the optimization matrix, the BP model was trained secondly. Taking the 100 groups of data gathered from worsted mill to verify the performance, optionally selecting 80 groups to use for the model establishment, the other 20 groups that are not participated in modeling to forecast the roving quality, the relative mean error percent (MEP) between the forecast results and measured value of GNN are all lower than pure NN´s. Meanwhile the correlation coefficients between them of GNN are larger than pure NN´s. Therefore, the accuracy and performance of GNN model are all enhanced greatly. These indicate that the union of Genetic algorithm and neural network is effective and the performances of forecast are remarkable improved.