The Research Of Breakout Prediction System Based On Quantum Wavelet Neural Networks

Breakout is the most catastrophic incident in continuous casting. It shatters equipment, reduces task efficiency and breaks the balance of producing. There are two main methods to avoid or reduce it, one is to study the mechanism of breakout, the other is to explore techniques of breakout prediction.In the paper, the mechanism of breakout incident is firstly introduced. From this, it is necessary to predict the breakout in continuous casting. The status of researching in breakout prediction system at home and abroad indicate that breakout prediction system based on neural network replace breakout prediction system based on the logic has become a trend. The paper analyses the method and principle of breakout prediction deeply. The thermocouples detect the temperature in the mold. Actually, the process of breakout prediction is the process that recognizes waveform which have breakout omen.In order to improve the accuracy of the waveform pattern recognition of the breakout prediction system, the author studied the traditional neural network, and the quantum neural network combines with wavelet theory form the quantum wavelet neural network model has been given. Experiments show that this model has great advantages in pattern recognition. The quantum neurons of hidden layer of the model using a linear superposition of wavelet function as incentive function, such hidden layer neurons not only can express more of the status and magnitude, but also can improve network speed and accuracy of convergence. The model will reasonably allocate uncertainty of decision to all patterns, and reduce the uncertainty of pattern recognition, improve the accuracy of pattern recognition. The same time, the paper presented learning algorithm of quantum wavelet neural network.In addition, through the process of studying, the following results are achieved:1. After studying the quantum neural network and wavelet neural network, the quantum wavelet neural network model and algorithm are proposed. And the algorithm provides an effective method in pattern recognition. The validity of the algorithm is proved by application in gear failure pattern recognition and waveform pattern recognition of breakout prediction system.2. The sample temperature data in the mold collected by thermocouples cannot be used as input data of the neural network directly. When breakout happened, the temperature changed with a special process. The paper has given a method of data preprocessing. After processing, the sampling data has been compressed within value of six categories, which is between 0 and 1.3. In order to improve the accuracy of the breakout prediction system, a large number of sample temperature data must be provided for training the neural network. As a result of the complex environment in the casting scene and the hardness of keep temperature data when breakout happened, it is hard to gain enough sample data. The paper has given a new method of achieving sample data which based probability and statistics.4. The author studied the single-thermocouple and multi-thermocouple network prediction model of the breakout prediction system, and simulated the prediction system in Matlab6.5. At last, the breakout prediction system is developed with Visual C++ based on the windows platform, which can simultaneously see all the function and circulation of all parts in continuous casting. The experiment showed the accuracy of the breakout prediction system has improved.