Model Study On Molten Steel Breakout Prediction In Wide And Thick Slab Continuous Casting Mold

In order to reduce the false alarm rate in molten steel breakout prediction, this paper studied the formation mechanism of sticking breakout and variation pattern of mold copper plate temperature and mold friction under sticking in combination with production practice of4#continuous casting machine in a steel plant. By optimizing the single thermocouple time-sequential and grouping thermocouple space neural network model, introducing FCM-AGA algorithm to the breakout prediction model, establishing the friction monitoring neural network and breakout probability adjustment neural network, the multi-couplings breakout prediction model was set up. And the false alarm rate was effectively decreased. The major conclusions are summarized as follows:(1) The new sticking shell hot spot propagation model was set up, based on this model the proportional coefficient β between vertical propagation velocity and casting speed was0.7, and the intervals of time-delay sequences length of upper and nether thermocouples’ time-sequential neural network were [7,9] and [6,8] respectively.(2) Based on theoretical analysis and mold friction data of production field, mold friction had no fixed pattern and presented random variation.(3) By using FCM-AGA algorithm which is organically combined fuzzy clustering method with adaptive genetic algorithm, the mean square error of RBF neural network decreased to8.5×10-5.(4) By introducing friction monitoring neural network to the single thermocouple time-sequential and grouping thermocouple space neural network model, model II was set up. The false alarm rate of model Ⅱ decreased to a certain extent. Compared with single thermocouple time-sequential and grouping thermocouple space neural network model (model I), the false alarm rate of model Ⅱ decreased by3.21%.(5) By introducing the breakout probability adjustment neural network to model II, the multi-couplings breakout prediction model (modelⅢ) was set up. Compared with model I, model II and model of4#continuous casting machine, the false alarm rate of model III decreased by15.87%,12.38%and15.59%respectively.