Research Of Caster Breakout Prediction Model Based On GA-BP Neural Network Pattern Recognition

Casting is a modern steel-making enterprises the main slab casting method, which has a simplified production process, molten steel utilization, low energy consumption, good billet quality, improve working conditions, mechanization and automation of the production process advantages. Application of continuous casting technology revolutionized the foundry production process, personnel distribution and logistics control, application of steel production automation and information technology, thus greatly improving the environment and improve product quality to provide a strong guarantee material conditions.Breakout is a continuous casting process extremely hazardous major industrial accidents, in order to avoid casting breakout, the international community usually begin to study and solve from two directions, one is to improve the continuous casting equipment and process conditions, from to prevent the formation of the mechanism of breakout accident; on the other hand is detected in time to identify the characteristics of breakout, take effective measures to slow down the casting and so avoid breakout. In this paper, the formation mechanism of steel sticking break, breakout pattern recognition were analyzed to understand the dynamic transmission temperature is the basic physical characteristics of the breakout mapping process. The dynamic data thermocouple into static data, in order to achieve a combination of network structure in the area of discrimination law. On this basis, seek a valid GA-BP neural network based pattern recognition caster breakout prediction model.In this paper, mold thermocouple temperature change when the adhesive breakout analyzed, the thermocouple temperature variation with time as pattern recognition feature, which judges whether breakout. This model not only to establish a single galvanic coupling of single timing network, and the establishment of a group of more than galvanic coupling space network, the establishment of the dual space of the model group, the difference between the previous "1" shaped two galvanic network structure and "-1-" four galvanic-shaped network structure, the establishment of a "T" four galvanic-shaped network structure, and two four galvanic space network structure than the right, thus caster breakout recognition mode will be more accurate and more timely.In this paper, BP neural networks and genetic algorithms are discussed, genetic algorithm and BP neural network using genetic algorithm initial weights of BP neural network globally optimal search, in which the population using derivative amend the number of individual parameters to improve the genetic algorithm local search capability is not strong disadvantage, accelerate genetic algorithm convergence rate, after searching the optimal solution to conditions, will give the optimal solution for accurate BP network to solve. This not only avoids the local minima problem, but also to achieve a rapid and efficient global optimization purposes, so that the training results have been greatly improved. BP neural network with the steepest descent method for the study criteria to the error back propagation algorithm to adjust the connection weights, a large number of training data input to the neural network to obtain the optimal solution weights, then get relatively better connected caster breakout prediction model.Use Visual Studio 2010 integrated development environment, complete program development and interface design in C++, MFC interface part of the design is complete. Use the OpenGL cross-language, cross-platform professional graphics program interface functions in software development is completed thermocouple waveform display. Training and offline test even the timing of a single network and group even space network prediction model, the test results show that, with good precision and classification of nonlinear mapping ability caster breakout prediction model GA-BP neural network pattern recognition.