基于PCA决策的PSO-BP模型预测高炉铁水产量

Molten iron yield predicting of blast furnace using PSO-BP model based on PCA decision

铁水产量是衡量钢铁厂产能效益的重要经济指标,根据炉次特征对其精准预测有利于钢铁厂的产能结构优化,可促进高炉的稳定与高产。为提高铁水产量预测准确率,结合机器学习理论,以国内某钢铁厂2022年全年生产冶炼数据为基础,提出基于主成分分析(PCA)决策的粒子群优化-反向传播(PSO-BP)混合预测模型。首先利用主成分分析对原始数据集进行降维处理;然后利用粒子群搜索算法优化BP神经网络的权值矩阵,成功解决BP神经网络收敛速度慢、易陷入局部最优的问题;最后结合炼铁理论,根据主成分分析结果确定模型的输入向量与拓扑结构。测试结果表明,该模型相较于其他传统模型预测误差更小,在误差范围为±50 t的情况下准确率达99.8%,实现对高炉铁水产量的精准预测,可有效指导铁水包的中转调度,为高炉参数调控提供数据支撑。

The yield of molten iron is an important economic indicator to measure the capacity efficiency of steel plants,and its accurate prediction according to the characteristics of furnaces is conducive to capacity structure optimization of steel plants and promotes the stability and high yield of blast furnace.In order to improve the prediction accuracy of molten iron yield,combined with machine learning theory,a hybrid prediction model of particle swarm optimization-back propagation(PSO-BP)based on principal component analysis(PCA)decision-making was proposed based on the annual production and smelting data of a domestic steel plant in 2022.To begin with,principal component analysis was used to reduce the dimensionality of the original data set,and then the particle swarm search algorithm was used to optimize the weight matrix of BP neural network,which successfully solved the problem that BP neural network had slow convergence speed and was easy to fall into local optimality.Finally,combined with the ironmaking theory,the input vector and topology of the model were determined according to the results of principal component analysis.The testing results show that the prediction error of the model is smaller than that of other traditional models,and the accuracy rate is 99.8%when the error range is±50 t,which accurately realizes the prediction of molten iron yield for blast furnace,effectively guides the transfer scheduling of molten iron ladles,and provides data support for blast furnace parameter regulation.