Ministry of Education Key Laboratory of Coal Science and Technology, State Key Laboratory of Clean and Efficient Coal Utilization, Taiyuan University of Technology
Shaanxi Coal Industry Chemical Technology Research Institute Limited Liability Company
College of Mechanical and Vehicle Engineering, Taiyuan University of Technology

通过关联配合煤中硫、灰分、挥发分的质量分数等主要的煤质指标,利用基于Adma算法为优化器的GRU神经网络模型,不断对模型参数进行调整后,通过sigmoid激活函数判断模型准确率的多标签多分类方法,建立了焦炭质量预测模型。结果表明:当三层GRU网络的隐层神经元数量为(64,64,64);学习率为0.01;样本批次大小为64;样本训练次数为50;丢弃率为0.3时,得到了模型的最优参数,此时模型预测准确率达到97%。采用GRU神经网络多标签多分类焦炭预测模型不仅具有高精度、低损失函数等特点,而且针对小样本配煤数据预测焦炭质量可以达到很好的效果,对实际的配煤炼焦具有一定的参考意义。

By correlated the main coal quality indexes such as sulfur mass fraction, ash mass fraction, volatile mass fraction in coal, a coke quality prediction model was established using a GRU neural network model based on the Adma algorithm as the optimizer, continuously adjusting the model parameters, and the multi-label multi-classification method was developed on the basis of using the Sigmoid activation function to judge the accuracy of GRU neural network. The results show that when the number of hidden layer neurons in a three-layer GRU network is (64, 64, 64), the learning rate is 0.01, the batch size is 64, the training frequency is 50, and the dropout rate is 0.3, the model achieves optimal performance. Under these conditions, the predic-tion accuracy of the model reaches 97%. The coke quality prediction model is proposed, which is based on the GRU neural network and the multi-label multi-classification method, not only demonstrates high precision but can also produce accurate predictions of coke quality using small sample coal blending data. These results hold significant value for actual coal blending coking and can serve as an important reference.

郝晓东,乔星星,王 影,等.基于 GRU 神经网络多标签多分类的焦炭质量预测模型[J].煤炭转化,2023,46(6):90-100

HAO Xiaodong,QIAO Xingxing,WANG Ying,et al.Multi-label multi-classification coke quality prediction model based on GRU neural network[J].Coal Conversion,2023,46(6):90-100