The Study On The Neural Network Technology Applied In The Sizing Quality Forecast

On the basis of analyzing the complexity of the factors influencing the sizing quality and the relation between the various factors and the sizing quality, it can be concluded that it is very difficult to express the relation between the various factors and the sizing quality precisely with traditional mathematical model or empirical formula, but the application of neural network technology to the sizing quality forecast can receive twice results with half efforts in optimizing the sizing process and raising the sizing management level.In the foundation of learning the neural network technology principle, the history and present situation of the neural network technology application especially in textile industry are analyzed, and the neural network structure, sample data and the forecast model of the sizing quality forecast determined.In the aspect of network structure, the BP neural network trained by teacher is selected. The network calculates the output according to the teacher's sizing input, compares the teacher's sizing output with the network anticipated output result, then computes the error and adjust the internal parameter unceasingly according to the network algorithm, thus establish the network model conformed to the request.In the aspect of network sample data, as a result of time and experimental limitation, the original yarn quality is only considered under the substance of optimization of sizing pulp and process, although various factors influence the sizing quality. In the research process, the actual sizing production data is chosen and analyzed, which comes from a factory in Yancheng.In the aspect of building model of the neural network sizing quality forecast, the different from structures are established, which are single target network forecast model and multi-target network forecast model, through actual network training, simulation process, and comparison of speed and accuracy of two kinds of models, single-target model is determined in this article.The experimental result indicated that the neural network model can forecast the sizing quality accurately and the relative error may control in the contention of±2%.