Then the author has constructed the submodel of the time of blowing oxygen and refrigerant , making use of the rbf neural networks . meantime , applying the genetic algorithm , completed the optimizationof parameter of the number of neuron and spread , and get the model with a better generalization ability 然后分别应用rbf神经网络建立吹氧时间子模型、冷却剂子模型，并结合遗传算法完成rbf神经网络本身其神经元个数、 spread等参数的优化，使模型具有较强的泛化能力。
2.
Then the paper develop a refrigerant model by means of traditional bp algorithm 、 statistics theory 、 experience theory and thoroughly analyze the combination of these three methods and their predict precision . and then the paper put forward an advanced algorithm on the basis of combining the statistics theory and the problems on the traditional bp algorithm . the imitate results show that the algorithm has an ability of fast convergence speed and 然后分别应用传统bp算法、全量法和增量法建立冷却剂子模型，对三者的拟合和预报精度作了深入分析，并结合统计学习理论针对传统的bp算法存在的问题提出了改进的算法。仿真结果表明该算法具有较快的收敛速度和较强的学习能力。
