基于PSO-BP神经网络的转炉炼钢碳含量预测

Prediction of Carbon Content in Converter Steel-making Based on PSO-BP Neural Network

针对转炉炼钢终点控制问题,对炼钢过程中采集的炉口火焰光谱信息进行研究,采用阿特曼Z-score模型将光谱数据做标准化处理,获取量纲一致的样本集信息。采用主成分分析方法,提取了13个有效的光谱特征频率。基于耦合粒子群算法(PSO)与BP神经网络算法,构建基于炉口火焰光谱信息的炼钢后期钢水碳含量连续预报模型。预报效果采用正交试验方法验证与测试,研究结果表明:PSO-BP神经网络模型的整体测试误差在±0.05以内,具有一定的准确性与普适性,可为炼钢终点控制策略的优化提供定量指导。

Aiming at the end-point control of converter steelmaking,the spectrum information of the furnace mouth flame collected in steelmaking process was studied,and the spectrum data were standardized by using the Atman Z-score model to obtain the sample set information with the same dimension.Using Principal Component Analysis(PCA),13 effective spectral characteristic frequencies were extracted.Using coupled particle swarm optimization(PSO)and BP neural network algorithm,a new continuous prediction model of carbon content in molten steel based on the spectral information of furnace mouth flame was proposed.The prediction results were verified and tested by orthogonal test method.The results show that the whole test error of PSO-BP neural network model is within±0.05,which has certain accuracy and universality,and it can provide quantitative guidance for the optimization of control strategy of steelmaking endpoint。