基于深度信念网络的磷石膏充填材料强度预测  

Strength Prediction of Phosphogypsum Filling Material Based on Improved Deep Belief Network

作　　者：张安安[1] 邓芳明 ZHANG An-an;DENG Fang-ming(Jiangxi Academy of Science,Nanchang 330029,China;School of Electrical Engineering and Automation Engineering East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]江西省科学院能源研究所,南昌330029 [2]华东交通大学电气与自动化工程学院,南昌330013

出　　处：《科学技术与工程》2020年第18期7220-7225,共6页Science Technology and Engineering

基　　金：国家自然科学基金(51662008);江西省重点研究开发计划(20171BBG70078)。

摘　　要：针对目前现有的强度预测方法精度低,提出提取输入参数的深层连接的深度信念网络(DBN)强度预测模型,并采用量子粒子群优化算法(quantum particle swarm optimization,QPSO)来确定DBN的隐含层节点个数和学习率。为获得最优的预测性能,以充填材料的成分及其尺寸作为基于DBN的预测模型的输入,输出充填材料的抗压强度。实验结果显示,该预测方法仅用了1.89 s的预测时间且精度达到99.84%,相比于广泛应用的BP神经网络、RVM(relevance vector machine)、SVM(support vector machine)三种算法在精度和时间上都有显著提升。Aiming at the low accuracy of the existing strength prediction methods, a novel, fast, and accurate method was proposed on the paste filling material strength of different components. The deep belief network(DBN) that can extract the deep layer connection of input parameters was employed to establish the prediction model. To optimize the prediction performance, the quantum particle swarm optimization algorithm(QPSO) was applied to determine the number of hidden layer nodes and learning rate of DBN. The components of filling material and their dimension were employed as the input of prediction model based on DBN, the output result was the compressive strength of filling material. The experimental results are superior to those by BP neural network, relevance vector machine(RVM), and support vector machine(SVM) methods, and the prediction time(1.89 s) and precision(99.84%) achieves significant improvements.

关 键 词：磷石膏 抗压强度 强度预测 量子粒子群优化算法 深度信念网络 

分 类 号：TD989[矿业工程—选矿]