QQ群聊

• CN 62-1112/TF
• ISSN 1005-2518
• 创刊于1988年

Based on Particle Swarm Algorithm to Optimize the BP Neural Network of Leaching Rate Prediction in Leaching Mining

BU Jinge,, CHEN Jianhong,

School of Resources and Safety Engineering，Central South University，Changsha 410083，Hunan，China

 基金资助: 国家自然科学基金青年基金项目“基于人工智能的矿山技术经济指标动态优化 ”.  51404305国家自然科学基金项目“基于属性驱动的矿体动态建模及更新方法研究”.  51504286中国博士后科学基金面上项目“辰州矿业采掘计划可视化编制与优化研究”.  2015M 572269湖南省科技计划项目“辰州矿业采掘计划可视化编制与优化研究”.  2015RS4060

Received: 2019-06-11   Revised: 2019-09-30   Online: 2020-03-06

Abstract

With the development of mining technology，the development of mineral resources in China is progressing steadily.Nowadays，the mining trend is green mining mode with environmental protection and high safety.However，many mining methods are faced with serious pollution and low recovery rate.Leaching mining is a kind of mining method which combines mining，sorting and hydrometallurgy.In order to explore how to improve the leaching rate during the leaching process,in this paper,the leaching process of antimonial sulfide ore was taken as an example to analysis the main factors that influence the leaching rate.The BP neural network prediction model was established and optimized by the particle swarm algorithm,so it can conduct the variable training with existing sample data.Finally, the BP neural network model, the PSO - BP neural network model were used to predict leaching rate, respectively，and compared to two kinds of model error precision of the predicted values and actual values.The research results show that the impact containing antimony sulfide ore leaching rate of interaction between these factors and nonlinear relationship and leaching rate is more by 40 groups will affect the relationship between parameters and the leaching rate of leaching rate of data through the neural network training model to predict the 8 groups of leaching rate data，compare the leaching rate of output value and the actual values can be found using the particle swarm algorithm to optimize BP neural network model training accuracy is higher，the more accurate predictions for leaching rate，the prediction data set of normalized linear curve slope is more close to 1.Through further error analysis，it can be seen that compared with BP neural network，the relative errors and variances of the predicted results of the model optimized by particle swarm optimization algorithm and the actual values are significantly reduced.Therefore，this prediction model has certain reference value for the optimization of leaching rate in the current leaching mining area.

Keywords： leaching mining ; leaching rate ; variable training ; BP neural network ; particle swarm optimization ; error analysis

BU Jinge, CHEN Jianhong. Based on Particle Swarm Algorithm to Optimize the BP Neural Network of Leaching Rate Prediction in Leaching Mining[J]. Gold Science and Technology, 2020, 28(1): 82-89 doi:10.11872/j.issn.1005-2518.2020.01.076

2 BP神经网络

图1

Fig.1   Basic structure diagram of BP neural network topology

$M=fD*P,V$

$Ii（1）=xi,$$i=1,2$
$Oij(1)=Iij,$$i=1,2;j=1,2,⋅⋅⋅,n$

$Iij（2）=Oij(2)-αij2βij2,i=1,2;j=1,2,⋅⋅⋅,n$
$Oij（2）=expIij(2)$$i=1,2;j=1,2,⋅⋅⋅,n$

$I3=∑p=1mOp3Wp$
$O（3）=I（3）∑p=1mOp（3）$

$αikt+1=αikt-∂J∂αik+λαikt-αikt-1$
$βikt+1=βikt-∂J∂βik+λβikt-βikt-1$

BP神经网络虽然可以处理非线性参数，但也存在一定的不足和局限性。比如：在样本的选择中，要求样本必须具有代表性，且如果样本数据参数较少，会导致BP神经网络在训练参数时无法优化到预期值。同时BP神经网络的初始权重随机性较大，因此常会出现训练过程无法高速收敛和结果精度低的现象。

3.1 粒子群算法

$Pgk=P1k,P2k,⋅⋅⋅,PnkfPgk$
$=minfP1k,fP2k,⋅⋅⋅,fPnk$
$Pi=ϕ1Pid+ϕ2Pgdϕ1+ϕ2$
$mbest=m∏i=1mPi$
$Widk+1=τPid-Pgd±γmbest-Widk×ln1u$

图2

Fig.2   Process of particle swarm optimization neural network

（1）计算粒子的适应度值，并确定个体极值和全局最优极值。

（2）运用第3.1节中的式（10）~（13），对粒子的速度位移进行更新计算，得到粒子适应度更新值。

（3）根据新的适应度值重复更新粒子的个体极值和全局极值。

（4）反复迭代后，当误差达到期望值或达到设定的最大迭代次数时，结束粒子群算法，此时根据得到的最优结果设置新的神经网络权值与阈值。

图3

Fig.3   Structure diagram of PSO-BP neural network with leach rate as output variable

4.2 神经网络模型的构建与预测结果

Table 1  Sample data of training group

185110300448.17
28558300484.73
385310500385.06
485410300383.75
585310300476.75
685110300346.25
785510300112.03
885110500358.32
985410900390.93
1085510300390.50
1185310300251.19
128556300476.00
1385310300110.75
1485210300464.25
1585412300491.08
1685110300236.75
1785210300110.24
1885112300459.07
198516300444.14
2085510900393.75
2185110900374.21
2285110300447.25
2385210900388.00
2485312300487.78
258528300463.65
2685510300493.06
2785512300493.58
2885410300486.25
2985110700363.53
308536300465.57
318556300476.00
3285410300111.01
3385410500389.96
3485510300493.06
3585210500372.67
3685210300466.20
3785510500393.24
3885210700374.61
3985410700391.48
4085510300266.25

$Np=Ap-Amean,p/Astd,p$

$EMS=12P∑p=1P∑n=1Ntpn-dpn2$

Table 2  Sample data of prediction group

18548300480.83
285510700393.55
385310300478.06
485210300466.02
585410300383.75
685310300376.75
785310300478.06
885310700385.26

图4

Fig.4   Linear regression curve of particle swarm optimization neural network leach rate prediction group

图5

Fig.5   Linear regression curve of BP neural network leach rate prediction group

Table 3  Statistics of prediction results of BP neural network and PSO-BP neural network models

180.8379.45-0.017180.45-0.0047
293.5592.70-0.009393.21-0.0036
378.0678.210.001978.090.0038
466.0266.920.013666.130.0017
583.7583.05-0.008383.40-0.0042
676.7574.98-0.023175.91-0.0109
778.0678.870.010478.760.0089
885.2685.300.000485.290.0004

Table 4  Error analysis of prediction results of BP neural network and PSO-BP neural network models

5 结论

（1）溶浸开采过程中的浸出率受诸多因素的影响，在含锑硫化矿的浸出过程中，浸出率主要受温度、反应时间、固液比、搅拌速度和HCl浓度的影响，其中HCl浓度占主导因素，其次为温度、搅拌速度、固液比和反应时间，其影响占比依次减小，这些因素相互作用，协同影响着浸出率。

（2）采用粒子群算法对BP神经网络进行优化，有效地提高了神经网络对浸出率的预测精度，预测结果的相对误差整体降低，模型对应结果的方差也相对降低，表明经粒子群优化的神经网络模型的预测更精确、更稳定。

（3）在溶浸开采过程中，影响浸出率的因素非常繁杂，部分影响因素与浸出率多呈非线性关系，因此利用粒子群算法优化BP神经网络模型，以此模型来模拟预测受多因素影响的溶浸开采浸出率是一种精准且有效可行的预测方法。

参考文献 原文顺序 文献年度倒序 文中引用次数倒序 被引期刊影响因子

[J].中国有色金属学，20182810）：2103-2111.

Tian QinghuaHong JianbangXin Yuntaoet al.

Prediction of oxidation leaching behavior of antimony-containing sulfide ore based on artificial neural network model

[J].The Chinese Journal of Nonferrous metals20182810）：2103-2111.

[J].矿业研究与开发，1998183）：12-14.

Ma Shengli.

Analysis and discussion on optimization of leaching mining

[J].Mining Research and Development1998183）：12-14.

[J].采矿技术，2013133）：42-45.

Ma ChunhuaHuang Zhineng.

Review of research and application of leaching mining technology

[J].Mining Technology2013133）：42-45.

[J].河南师范大学学报（自然科学版），2007353）：167-171.

Qin YinanLiao XiaohuiZhao Qingzhi.

A neural network training method based on particle swarm optimization

[J].Journal of Henan Normal University（Nature Science Edition）2007353）：167-171.

[M].北京原子能出版社199815-28

Wang Changhan.

Leaching Uranium（Ore）

[M].BeijingAtomic Energy Press199815-28.

[J].采矿技术，200663）：39-48.

Wu AixiangWang HongjiangYang Baohuaet al.

Development and prospect of leaching mining technology

[J]. Mining Technology200663）：39-48.

[J].电子设计工程，20111924）：62-65.

Mao JianZhao HongdongYao Jingjing.

Application and prospect of artificial neural network

[J].Electronic Design Engineering20111924）：62-65.

[M].北京北京邮电大学出版社2006.

Han Liqun.

Artificial Neural Network

[M].BeijingBeijing University of Posts and Telecommunications Press2006.

Ren CAn NWang J Zet al.

Optimal parameters selection for BP neural network based on particle swarm optimization：A case study of wind speed forecasting

[J].Knowledge-Based Systems，2014563）：226-239.

[D].长沙湖南大学2006.

Peng Shilin.

Combination of Thermal Environment Parameters of Air-Conditioned Rooms Based on Artificial Neural Network

[D].ChangshaHunan University2006.

[M].西安西安电子科技大学出版社1990.

Jiao Licheng.

Neural Network System Theory

[M].Xi’anXi’an University of Electronic Science and Technology Press1990.

[J].中南大学学报（自然科学版），2011428）：2400-2405.

Liu LangChen JianhongYang Shanet al.

Prediction of PSO-BP algorithm in risk prediction of mine earthquake based on grey correlation analysis

[J].Journal of Central South University（Science and Technology）2011428）：2400-2405.

[J].计算机工程与应用，20134921）：235-239.

Chen JianhongZhou HanlingYu Fengling.

Uranium price forecasting model based on BP improved by QPSO and its application

[J].Computer Engineering and Application20134921）：235-239.

[J].现代电子技术，2019427）：87-94.

Zhu QingWang Jingjing.

Research on university scientific research management evaluation based on particle swarm optimization algorithm and BP neural network

[J].Modern Electronics Technique2019427）：87-94.

[J].内蒙古煤炭经济，20148）：182-184.

Zhang Lei.

Prediction of soft rock roadway deformation based on particle swarm optimization neural network

[J].Inner Mongolia Coal Economy20148）：182-184.

[J].温州职业技术学院学报，200992）：45-51.

Xu Min.

Application of BP neural network in slope stability evaluation on the basis of PSO algorithm

[J].Journal of Wenzhou Vocational and Technical College200992）：45-51.

[J].计算机应用与软件，2009261）：218-221.

Zhao Zhenjiang.

Prediction and research on network traffic based on PSO-BP neural network

[J].Computer Applications and Software2009261）：218-221.

[D].长沙中南大学2007.

Yuan Ziqing.

Research on Seismic Aactivity Response Law and Hazard Prediction of Mine Earthquakes

[D].ChangshaCentral South University2007.

Kim H JJang B SPark C Ket al.

Fatigue analysis of floating wind turbine support structure applying modified stress transfer function by artificial neural network

[J].Ocean Engineering，2018149113-126.

BP神经网络在状态监测数据趋势预测中的应用

[J].微计算机信息，200521）：141-143.

Wang WeiLi Hongru.

Application of BP ANN in predicting condition monitoring data

[J].Microcomputer Information200521）：141-143.

MATLAB 神经网络应用设计

[M].北京科学出版社2001278-283.

Wen Xin.

Neural Network Simulation and Application Based on MATLAB

[M].BeijingScience Press2003278-283.

/

 〈 〉