Research on Breakout Prediction System Based on Multilevel Neural Network

Abstract

A new type of breakout prediction system based on multilevel neural network for continuous casting was proposed, which consists of a pattern recognition unit of single-thermocouple temperature pattern based on BP neural network, a logic judgment unit of multi-thermocouple temperature pattern and a decision making unit of fuzzy neural network based on T-S (Takagi-Sugeno) model. In the training of BP (error back propagation) network, the data fitting method based on the maximum entropy function was used to process input data, and the structure of the network was simplified. According to the law of crack growth in the bonding area and adopting the horizontal network prediction model to logically judge multi-thermocouple temperature pattern, the prediction time was shortened. By considering the overall influencing factors of breakout and using the fuzzy neural network to establish the model, the breakout prediction system made the final judgement. The result showed that the breakout prediction system based on multilevel neural network can effectively decrease the false alarm rate and improve the prediction accuracy.