Research Of Steel Temperature Model For Continuous Roller Annealing Furnace

Continuous roller annealing furnace is an important finishing equipment of the finishing line. At present, due to the restrictions of test technology and the annealing furnace’s factors itself, making direct and accurate measurement of the temperature distribution of bar annealing process is very difficult. We can know the temperature of bar only after the bar come out from the furnace. Therefore, the establishment of a continuous roller annealing furnace temperature prediction model which can predict the steel temperature is important, it can be use to optimize setting every section temperature of the furnace and increase production quality and quantity.Bar annealing process has characteristics such as big lag, large inertia, multivariable, strong coupling, time-variant, nonlinear and so on, then traditional mechanism model has many assumptions and is not flexible enough. In this thesis, continuous roller annealing furnace was chosen as the research object,and according to its technological features, a steel temperature prediction model which was based on neural network that improved by L-M algorithm was established. Inputs are the furnace’s13sections temperature and the output is bar’s surface temperature. Training the BP neural network by the data collected from the furnace, the simulation results by MATLAB show that the bar temperature prediction model is reasonable and can accurately predict steel temperature. Then according to the typical faults that BP network convergences slowly, and easily falls into the local minimum, it is reasonable to make use of genetic algorithm(GA)and simulated annealing(SA) to optimize the bar temperature prediction model, they all improve the prediction effect. Finally consider that GA can search for the best result in the global scope quickly and SA has good local searching ability, so making the advantages of GA and SA combined, put forward a new kind of mixed optimization algorithm that SA-GA, which can be used for optimizing the BP neural network that improved by LM algorithm. Simulation results show that it is easy to use the mixed algorithm to optimize bar temperature model and can make more accurate prediction effect.By comparing with the simulation results of bar temperature model being improved with various optimization algorithms, it is possible to draw the conclusion: the model optimized by SA-GA is the best, this model’s accuracy and efficiency have been improved significantly and achieves the best prediction effect, moreover the mixed algorithm only requires a large number of actual data, does not need to consider the mechanism model’s parameters and boundary conditions, so the mixed algorithm is more suitable for establishing the bar temperature prediction model of continuous roller annealing furnace.