基于双隐含层BP神经网络的某金矿回收率预测研究  

Prediction of the Recovery Rate of a Gold Mine Based on Double Hidden Layer BP Neural Network

作　　者：张帅[1] 赵鑫[1] 彭祥玉 王宇斌[1] 桂婉婷 田家怡 ZHANG Shuai;ZHAO Xin;PENG Xiangyu;WANG Yubin;GUI Wanting;TIAN Jiayi(School of Resources Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,Shaanxi,China)

机构地区：[1]西安建筑科技大学资源工程学院,陕西西安710055

出　　处：《黄金科学技术》2024年第1期170-178,共9页Gold Science and Technology

基　　金：陕西省自然科学基金项目“双重难选碳质金矿中的石墨吸附机理研究”(编号:2019JQ-545)资助。

摘　　要：为掌握某金矿选矿工艺影响因素对金实际回收率的作用规律并预测金的回收率,采用正交试验方法开展了金矿浮选试验,通过Pearson系数分析金回收率对不同工艺因素的敏感性,并利用双隐含层BP神经网络对金回收率进行预测。结果表明:生产过程中金回收率对不同因素的敏感性由大到小依次为2#油用量、Na2S用量、丁基黄药用量、CuSO_(4)用量和磨矿细度。在此基础上,选用2#油用量、Na_(2)S用量和丁基黄药用量3个主要影响因素,使用不同隐含层激活函数的BP神经网络对金回收率进行预测。预测结果表明:当使用“logsig”作为激活函数时,其拟合度与精度较高,拟合优度R2为0.9792,相对平均误差仅为0.666%,说明该模型能够较好地预测金回收率。研究结果对贵金属矿山生产中金属回收率的预测有一定的参考意义。In order to grasp the action law of process factors affecting the actual recovery rate of a gold ore and predict the gold recovery rate,the flotation test was carried out by the method of orthogonal experiment.The sen‐sitivity of process factors was analyzed by Pearson coefficient,and the gold recovery rate was predicted by us‐ing double hidden layer BP neural network.The results show that the sensitivity of the gold recovery rate to dif‐ferent factors in the production process is in descending order:2#oil dosage,sodium sulfide dosage,butyl xan‐thate dosage,copper sulfate dosage and grinding fineness.The reasons for the significant differences in the ef‐fects of 2#oil dosage,sodium sulfide dosage and butyl xanthate dosage on gold recovery rate were also eluci‐dated.On this basis,used three main influencing factors such as 2#oil dosage,the study selected different input layer to the first implicit layer functions,such as tansig,purelin and logsig,and the rest of the activation func‐tions remained unchanged.The first hidden layer to the second hidden layer function was logsig,and the second hidden layer to the output layer function was purelin.When research used logsig as the activation function,the fitted degree and accuracy are high,the goodness of fit R2 is 0.9792,and the relative average error is only 0.666%.The model can be used to predict the recovery rate of gold.The research has certain reference signifi‐cance for the prediction of metal recovery rate in the production of precious metal mines.

关 键 词：BP神经网络 Pearson系数 激活函数 影响因素 金矿 回收率 

分 类 号：TD953[矿业工程—选矿]