基于BP神经网络的电弧炉炼钢过程的终点预报

The End-point Prediction of EAF Steel making Based on BP Artificial Neural Network

针对电弧炉炼钢过程的高度非线性、时变性、不确定性,基于BP算法,建立了增量神经网络电弧炉冶炼终点的预报模型。以废钢、铁水、铁矿石、通电时间、吨钢氧耗和电耗相对于参考炉均值的增量为输入节点,对冶炼钢水终点温度和碳、磷的含量进行预报。研究表明,当钢水终点温度和碳、磷含量的控制精度分别在±10℃、±0.03%和±0.04%时,预报值命中率分别为93%、95.2%和86.3%。

According to the high nonlinear time variation, the increment artificial neural network model based on BP algorithm has been developed to predicate the end-point of EAF steel making. The end-point temperature and carbon and phosphorus content are predicted by increment of scrap, hot metal, charging program, power on time, oxygen and electric power consumption per ton steel corresponding to reference furnace average value as input nodes. The results shows that increment of artificial neural model can provide good predication of the end-point of EAF steel making. If the controlled precision of temperature of molten steel and carbon and phosphorus content are ±10℃, ±0. 03 % and ±0. 04 % individually, the percentage of hits of predicted value is respectively 93% , 95.2 % and 86.3 %.