【摘要】 热连轧粗轧生产过程中，为解决换规格后宽度设定精度低的难题，提出了一种蚁群优化算法协同深度极限学习机(ant colony optimization deep extreme learning machine, ACO-DELM)的热连轧粗轧宽度预测方法.该方法将蚁群优化算法应用于DELM网络中，以提高其预测精度和泛化能力.先利用数据预处理方法对原始数据进行异常值的剔除和数据归一化.然后，使用蚁群优化算法对DELM的隐藏层节点数、迭代次数进行优化，在隐藏层节点数达到95个、迭代次数为480次时，DELM模型的预测性能最佳，其在预测不同规格带钢平均宽度时，决定系数R~2达到0.998 9,97.98%的样本点预测误差分布在-7～7 mm.应用结果表明，与传统的深度极限学习机(DELM)、卷积神经网络(CNN)等模型相比，ACO-DELM模型在预测精度和泛化能力上有明显的提升，可有效应用于热轧带钢的平均宽度预测.更多还原

【Abstract】 In order to solve the problem of low precision of width setting after changing specifications in the rough rolling process of the hot strip rolling mill, a rough rolling width prediction method based on the ant colony optimization algorithm and deep extreme learning machine(ACO-DELM) was proposed. The method applied the ant colony optimization algorithm to the DELM network to improve its prediction accuracy and generalization ability. Firstly, the original data were removed from the outliers, and the data were normalized by using data pre-processing methods. Then, the number of hidden layer nodes and the number of iterations of DELM were optimized by using the ACO algorithm. When the number of hidden layer nodes reached 95 and the number of iterations was 480, the prediction performance of the DELM model was the best. When predicting the average width of strip steel of different specifications, the determination coefficient R~2 reached 0.998 9, and the prediction error of 97.98% of the sample points was distributed in-7～7 mm. The application results show that compared with traditional deep extreme learning machine(DELM) and convolutional neural network model(CNN), the ACO-DELM model has obvious improvements in prediction accuracy and generalization ability, and can be effectively applied to the prediction of the average width of hot-rolled strip.更多还原