基于GA-PSO-BP神经网络的LF终点温度预测  

GA-PSO-BP neural network based prediction model for LF end point temperature

作　　者：李军[1] 贺东风[1] 徐安军[1] 田乃媛[1] 

机构地区：[1]北京科技大学钢铁冶金新技术国家重点实验室,北京100083

出　　处：《炼钢》2012年第3期50-52,共3页Steelmaking

摘　　要：针对LF冶炼特点和现有钢水温度预报方法存在的不足,提出了一种新的预测LF终点温度的BP神经网络模型。用遗传算法(GA)和粒子群算法(PSO)混合优化BP神经网络的权值和阈值,提高BP神经网络的预测精度。混合模型既克服了传统机理模型难以准确实现的困难,也弥补了传统BP算法的不足,结合了2种算法的优点,改善了预测模型的收敛性能。开发了基于Java语言的现场应用软件。仿真结果表明,该算法可以提高预测精度和速度,预测误差在5℃以内的炉次达到了88%。In view of the characteristics of the LF （ladle furnace） metallurgical process and the disadvantages of traditional hot metal temperature prediction methods a new BP neural network model for prediction of the LF end point temperature is developed. The genetic algorithm（GA）and particle swarm optimization（PSO）are properly combined to optimize the weight and bias value and improve the prediction model precision of BP neural network. The hybrid model can not only overcome the difficulties of the traditional prediction model of its prediction inaccuracy, but also compensate the insufficiency of the traditional BP algorithm. The hybrid algorithm combines the advantages of both and improves the convergence performance of the prediction model. Simulation results show that by use of this model not only the prediction precision and speed can be improved but also heats of the refined hot metal with prediction deviation of less than 5℃ reaches 88%.

关 键 词：LF精炼炉 钢水温度预测BP神经网络 遗传算法 粒子群算法 

分 类 号：TF769.2[冶金工程—钢铁冶金]