基于随机森林特征选择的BiLSTM电解槽出铝量预测  

Prediction of aluminum yield from BiLSTM aluminum reduction pots based on random forest feature selection

作　　者：孙少聪 徐杨 曹斌 Sun Shaocong;Xu Yang;Cao Bin(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;Chalco Intelligent Technology Development Co.,Ltd.,Hangzhou 311100,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025 [2]中铝智能科技发展有限公司,浙江杭州311100

出　　处：《轻金属》2023年第10期30-36,共7页Light Metals

基　　金：贵州省科学技术基金资助项目(黔科合基础-ZK[2021]重点001)。

摘　　要：铝电解槽出铝量需要凭借专家经验对槽况的判断,其经验水平决定了出铝量决策的准确度。针对电解槽出铝量预测问题,本文提出了一种基于随机森林(RF)特征选择的RF-BiLSTM电解槽出铝量预测模型。所用模型在BiLSTM模型的基础上,利用随机森林算法对输入BiLSTM模型的特征进行降维处理,并将优化后的特征进行不同模型的对比实验。实验结果表明,与LSTM方法相比,RF-BiLSTM平均绝对误差(MAE)减少21.01,该方法优于现有方法。为铝电解槽出铝量预测问题提供了一定的参考价值。The aluminum yield from aluminum reduction pots requires the judgment of pot conditions based on expert experience,and the experience level determines the accuracy of the aluminum yield decision.With respect to the prediction of aluminum yield from aluminum reduction pots,this paper proposes a prediction model for aluminum yield from RF-BiLSTM aluminum reduction pots based on random forest(RF)feature selection.Based on the BiLSTM model,the random forest algorithm is used to reduce the dimensionality of features which are entered the BiLSTM model,and the optimized features are compared with different models.The experimental results show that compared with the LSTM method,the mean absolute error(MAE)of RF-BiLSTM is reduced by 21.01,which is better than the existing method,thus providing a certain reference value for the prediction of aluminum yield from aluminum reduction pots.

关 键 词：铝电解 出铝量预测 随机森林 BiLSTM 

分 类 号：TF821[冶金工程—有色金属冶金] TP274[自动化与计算机技术—检测技术与自动化装置]