Research on LFFA-BP neural network model in breakout prediction

Abstract

Advance forecasting of breakout in the continuous casting process could help to improve the capacity and quality of slabs. Neural network-based prediction methods are the main methods used for forecasting, but they have the disadvantages of being complicated and time-consuming. To compensate for these shortcomings, Levy Flight (LF) and Firefly Algorithm (FA) are introduced into a Back Propagation (BP) neural network to establish the LFFA-BP network model, which applied to the prediction of breakout. First, the model utilizes LF to change the step size of FA to prevent it from falling into local optimal solutions. Subsequently, the network optimal threshold is determined through the powerful search capability of the FA. Lastly, the network computation time is reduced through the superior convergence speed of FA. The models before and after improvement are used to classify and predict the temperature data collected at a production site. The results show that the identification accuracy of the LFFA-BP breakout prediction model is significantly higher than that of the traditional BP breakout prediction model, since it achieved a prediction accuracy of 99.23% and reporting rate of 100%. The improved model not only accelerates the running speed of the network model, but also ensures its global search capability and robustness, indicating that it has good application prospects.