Research On Hot-rolling Plate Quality Model Of Wavelet Neural Network Based On PSO

With economic development, the market for hot-rolling steelplate quality have become increasingly demanding. In recent years,in order to meet the market demand, research of scholars is veryactive to the quality of hot-rolling plate model based on artificialneural network technology. They made some hot-rolling platequality models, such as BP neural network, RBF network and fuzzyneural network model so on, but the test hits of these hot-rollingplate quality model can not meet the requirement of the manufacturer,therefore, The following work has been done:Firstly, relevant theoretical knowledge and practical applicationof the BP neural network, wavelet neural network and particle swarmoptimization algorithm (Partical Swarm Optimization, PSO) wereanalyzed. At the same time, hot-rolling process of plate was studied,the production sample data and the sample input and output variableswere determined, the quality of hot-rolling plate requirements wereexpounded. Secondly, a new wavelet neural network model based on particleswarm optimization was constructed. By analyzing wavelet neuralnetwork model based on particle swarm optimization, we can seethat the network uses an iterative search formula of PSO algorithmto adjust the network parameters, can be a good escape from localminima with local optimization features, but the error can not be thefastest to reach minimum.Based on the new network firstcalculated error with wavelet neural network, and then adjust thevarious parameters of wavelet neural network with gradientdescent method and particle swarm algorithm.Network not onlyreach the minimum fast, but also escape from local minima.Examples of simulation models were constructed.The resultsshowed that the new wavelet neural network model based on PSOimproves test hits of9.8%, the training time is shortened.Finally, the paper designed and implemented a hot-rolling platequality prediction system. The system has functions of datasamples stored in the database, set the network parameters, thealgorithm parameters settings, display the results of the trainingand test. The system can be given hot rolling plate quality datamodeling and forecasting.