概率积分预计参数的神经网络优化算法  

Neural network optimization algorithm for the prediction parameters of probability integral method

作　　者：吕伟才 黄晖[2] 池深深 韩必武[2] LYU Weicai;HUANG Hui;CHI Shenshen;HAN Biwu(School of Geomatics,Anhui University of Science and Technology,Huainan, Anhui 232001,China;Huainan Mining Group Co.,Ltd.,Huainan,Anhui 232001,China;School of Earth and Environment,Anhui University of Science and Technology,Huainan,Anhui 232001,China)

机构地区：[1]安徽理工大学测绘学院,安徽淮南232001 [2]淮南矿业(集团)有限责任公司,安徽淮南232001 [3]安徽理工大学地球与环境学院,安徽淮南232001

出　　处：《测绘科学》2019年第9期35-41,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41474026,41602357,41404002,41704008);淮南矿业(集团)有限责任公司项目(HNKY-JTJS(2018)-178,HNKY-JTJS(2017)-122,HNKY-JTJS(2013)-28)

摘　　要：在分析BP神经网络不足的基础上,为提高概率积分法进行开采沉陷预计时所采用的预计参数的正确性,该文建立了地质采矿条件与预计参数之间的非线性关系,以我国43个地表移动观测站的实测数据为训练和测试样本,采用多种群遗传算法(MPGA)优化BP神经网络的权值和阈值,构建新的概率积分法参数解算方法。计算结果表明,较单纯的BP神经网络算法和标准的遗传算法而言,MPGA算法优化的BP神经网络算法解算的预计参数具有更高的相对精度,这对于获取待研究区域的高精度概率积分法预计参数具有良好的指导意义。Based on the analysis of the shortcomings of BP neural network,in order to improve the correctness of the predicted parameters used in mining subsidence prediction by probability integral method,the non-linear relationship between geological mining conditions and predicted parameters was established,the weights and thresholds of BP neural network were optimized by using multi-population genetic algorithm(MPGA)with the measured data of 43 surface movement observation stations in China as training and testing samples,a new method for calculating the parameters of probability integral method was constructed.The results showed that,compared with the BP neural network algorithm and the standard genetic algorithm(GA),the BP neural network algorithm optimized by MPGA algorithm had higher relative accuracy,which had a good guiding significance for obtaining the prediction parameters of the high-precision probability integration method in the area to be studied.

关 键 词：开采沉陷 概率积分法 预计参数 BP神经网络 多种群遗传算法 

分 类 号：TU196[建筑科学—建筑理论]