基于改进BP神经网络的铁水预处理终点硫含量预报模型

铁水预处理脱硫是纯净钢生产中的一项重要任务,其中铁水终点硫含量是反映脱硫站能力和生产效果的重要指标.对梅山钢铁股份有限公司铁水包喷吹CaO Mg粉剂复合脱硫过程,通过采用自适应调整学习率和最大误差学习法对标准BP算法进行了改进,建立了基于改进BP神经网络的铁水预处理终点硫含量预报模型.用梅钢的1154炉数据进行模型训练,经100炉数据现场验证表明,改进的BP算法比标准BP算法预报误差≤0.003%的精度提高28%,有19%的炉次预报值与实际值完全一致,有90%的炉次误差≤0.003%,平均误差为0.0017%.改进的BP算法在铁水预脱硫终点硫含量预报模型应用中获得了更好的使用效果.

Final Sulfur Content Prediction Model Based on Improved BP Artificial Neural Network for Hot Metal Pretreatment

Pre-desulfurization of hot metal has become an important task for clean steel production.The final sulfur content is a key index of desulfurization station for capacity and efficiency evaluation.Based on the practice of CaO+Mg powder co-injection at Meishan Steel,and improved BP algorithm with new method of adjusting study rate and learning method of maximal error,a prediction model of final sulfur content for hot metal pretreatment was established.The data of 1154 heats were used to training the model and the other 100 heats were selected as the test samples.It was shown that,the improved BP algorithm is more accurate than the normal one,the accuracy of prediction with error less than 0.003% was increased by 28%;for 19 percent of the total test heats the predicted values were the same as the actual ones,90 percent heats were with error less than 0.003%,the average error was(0.0017%).Thus the improved BP algorithm is suitable to predict the final sulfur content for hot metal pre-desulfurization.