Research On Breakout Prediction Model In Slab Continuous Casting Based On BP Neural Network

It is an important approach to improve the productivity and intact rate of continuous caster. Meanwhile, it can reduce the economic losses caused by breakout accidents, improve the quality of continuous casting billet by accurate prediction breakout accidents in continuous casting. In this paper, the influencing factors of sticking breakout and the mechanism of breakout prediction were analyzed. And then, in order to seek an effective breakout prediction model, neural network technology was introduced to research breakout prediction methods. The research has important significance on the theories and practical applications.In this thesis, a breakout prediction model of slab continuous casting is designed to solve the breakout of slab caster No.2 in Panzhihua Iron and Steel Corporation (PISCO). A breakout prediction model based on BP neural network is set up by analyzing the characteristics of BP neural network, its optimization algorithm and the temperature changes of thermocouple during breakout. The model's breakout prediction principle is to judge whether the breakout is take place by identifying temperature changes characteristics of thermocouple. The model combines a time series network with a spatial network. The time series network is designed to identify the thermocouple temperature characteristics in time direction, and the space network is to identify the different thermocouple temperature characteristics in space direction (horizontal and vertical). Then the breakout alarm information is generated by logical judgment.The breakout prediction model for casting slab based on BP neural network is designed and developed by Visual Basic 6.0, and it has strong availability, flexibility and reliability. The model was tested by test samples and off-line simulated forecast with actual history data acquired in Panzhihua Iron and Steel Corporation. In the samples measurement, the test samples does not include training samples, and been input the breakout prediction model to test the generalization performance, missing report and false alarm in the model. On this basis, the model was tested by off-line simulation with history database that does not include training samples and test samples.Testing results of the neural network breakout prediction model show that the recognition effect of model is satisfeid. In the tests, the model does not false alarm by samples measurement using 18 test samples included 10 group breakout alarm samples, 6 groups of normal casting samples and 2 group temperature fluctuations samples. In the off-line simulation, four times slab sticking accidents have been detected quickly and accurately, the missing report rate of the model is 0%, and the false alarm rate is only 0.26%(times / furnace).