基于收得率预测的合金配料优化研究

Research on Optimization of Alloy Ingredients Based on Yield Prediction

本文研究了钢水“脱氧合金化”配料安排的优化方案。首先,对低合金钢种前期冶炼采集的历史数据进行清洗,计算C、Mn元素的收得率;其次,应用主成分分析法对数据进行降维,将各项指标简化为四个主成分;进而,基于变异系数法建立C、Mn元素收得率影响因素的综合评价模型,并基于LSM-SVM算法对元素收得率进行预测,随机抽取200个基础数据作为训练样本,以50组基础数据作为预测样本进行预测并分析方差与误差;为提高预测的准确性,借助BP神经网络,对相同数据集进行训练。最后,在保证钢水质量的约束条件下,以合金使用的总成本最低为目标函数,建立基于收得率预测的合金配料优化模型,借助遗传算法,以10组锅炉为例给出具体的合金配料方案,有效降低配料成本。

This paper studies the optimization scheme of the “deoxidation alloying” batching arrangement of molten steel. Firstly, we wash the historical data collected during the early smelting of low-alloy steel grades to calculate the yield of element C and Mn. Secondly, we use principal component analysis method to reduce the dimensionality, and the indicators are simplified into four principal components. Furthermore, based on the variation coefficient method, a comprehensive evaluation model of the factors affecting element yield is established, then the element yield of C and Mn is predicted based on the LSMSVM algorithm: 200 sets of basic data are randomly selected as training samples while other 50 sets of basic data are used as prediction samples. Then we analyze the variances and errors. To improve the accuracy of predictions, we use BP neural networks to train the same data set. Finally, under the constraints of ensuring the quality of molten steel and commanding the minimum cost of the alloy as the objective function, an optimization model of alloy ingredients based on the yield prediction is established. With the help of genetic algorithm, allocation plans of alloy ingredients are specifically given to 10 boilers, which effectively reduce the cost of ingredients.