针对国内外转炉炼钢终点控制的现状, 建立了一种用于终点预测的神经网络模型。以炉口辐射信息获取系统为实验平台, 运用光纤谱分复用和颜色空间模型转换技术, 分析发现了光谱与图像信息特征量在吹炼过程中呈现出中前期类似、末期相反的规律。从得到的特征规律曲线中选用一些关键特征量, 在改进的修正系数算法基础上, 进行了模型的训练和预测分析。实验结果表明：响应时间在2 s以内, 满足快速判定的时间要求; 改进算法的模型预测精度高于常规算法, 该系统可以正常工作在转炉炼钢的恶劣环境下, 达到了预期效果。

Considering the present situation of the basic oxygen furnace (BOF) steelmaking endpoint control, a neural network model was established to judge the steelmaking endpoint. Based on the furnace mouth radiation information acquisition platform, the spectrum and image characteristics were analyzed using the fiber spectrum division multiplexing technology and the color space conversion method. The results indicate that they are similar at early-middle stage but inverse at the steelmaking late stage. Some appropriate variables were selected from the law curve as the neural network model parameters and the model was trained and forecasted on the basis of an improved back propagation (BP) neural network correction coefficient algorithm. The experimental results show the response time is less than 2 s which meets the requirements of online endpoint judgment, and the prediction accuracy of the proposed algorithm is better than that of the conventional algorithm. The system works stably and the anticipated effect is achieved.