基于CNN-LSTM-Attention网络的浮选品位监测模型构建  

Construction of flotation grade monitoring model based on CNN-LSTM-Attention network

作　　者：王春景 刘丹[1,2] 邵平 陈鑫 文书明 余龙舟[3] WANG Chunjing;LIU Dan;SHAO Ping;CHEN Xin;WEN Shuming;YU Longzhou(Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming Yunnan 650093,China;State Key Laboratory of Complex Nonferrous Metal Resources Clean Utilization,Kunming Yunnan 650093,China;Yunnan Amade Electrical Engineering Company,Kunming Yunnan 650033,China)

机构地区：[1]昆明理工大学国土资源工程学院,云南昆明650093 [2]省部共建复杂有色金属资源清洁利用国家重点实验室,云南昆明650093 [3]云南阿姆德电气工程有限公司,云南昆明650033

出　　处：《化工矿物与加工》2024年第12期1-7,共7页Industrial Minerals & Processing

摘　　要：针对工业生产中浮选品位波动大、监测难的问题,根据精矿品位与加药量、充气量、矿浆pH等浮选关键参数在时序上的相关性,运用长短期记忆网络(LSTM)提取时序特征、卷积网络(CNN)提取空间分量以及注意力机制(Attention)提取重点特征信息的方法,构建了一种将卷积网络、LSTM网络与注意力机制相结合的神经网络模型(CNN-LSTM-Attention模型),并将其与LSTM模型和CNN-LSTM模型进行对比,结果表明:本文模型预测结果的平均绝对误差为0.0635、均方根误差为0.0838、决定系数达0.9939;该模型的准确性优于其他模型,适用于浮选品位的实时监测。In response to the problems of large fluctuation in flotation grade and difficult monitoring in industrial production,according to the correlation of flotation key parameters such as concentrate grade,dosage,aeration and slurry pH in time series,a neural network model(CNN-LSTM-Attention model)combining convolutional network,LSTM network and attention mechanism was constructed by using long short-term memory network to extract time series features,convolutional network to extract spatial components and attention mechanism to extract key feature information.Compared with the LSTM model and CNN-LSTM model,the results show that the average absolute error of the prediction results of the proposed model is 0.0635,the root mean square error is 0.0838,and the determination coefficient is 0.9939.The accuracy of the model is better than other models,which is suitable for the real-time monitoring of flotation grade.

关 键 词：泡沫浮选 神经网络 卷积网络 时间序列 注意力机制 品位监测 

分 类 号：TD923[矿业工程—选矿] TP391.4[自动化与计算机技术—计算机应用技术]