摘要
为了提高BP神经网络对开采沉陷数据拟合和预测的精度,使用Matlab编写用于高精度拟合和预测开采沉陷数据的可执行程序。以多项式拟合的结果为基准,比较分析12种BP神经网络算法对二维数据沉降值的拟合精度,为提高程序对多维度数据的预测精度,讨论不同拓扑结构对下沉率预测的影响。研究结果表明:多项式拟合的结果存在异常区域,其拟合优度R_(1)=0.98194、残差平方和e_(1)=3.8971,均弱于BP神经网络拟合的精度;在12种优化算法中,列文伯格-马夸尔特算法以最少的迭代次数获得最高的拟合优度;随着拓扑结构的改变,预测精度有所不同,经分析3∶20∶10∶1拓扑结构的代码对多维度数据的拟合优度最高,预测值的标准差最小,最终确定适合下沉率预测的BP神经网络算法及拓扑结构。研究结果可为其他地表移动参数的预测提供可交互的程序。
In order to improve the accuracy in fitting and prediction of mining subsidence data by BP neural networks,an executable program for fitting and predicting the mining subsidence data with high accuracy was developed using Matlab.The results of polynomial fitting were used as a benchmark to compare and analyze the fitting accuracy of 12 BP neural network algorithms for two-dimensional data subsidence values.To improve the prediction accuracy of the program for multidimensional data,the influence of different topological structures on the prediction of subsidence rate was discussed.The results showed that the results of polynomial fitting had anomalous regions,the fitting superiority R_(1)=0.98194,the residual sum of squares e_(1)=3.8971,and all of them were weaker than the accuracy of BP neural network fitting.Among 12 optimization algorithms,the Levenberg-Marquardt algorithm obtained the highest fitting superiority with the least number of iterations.With the change of topological structure,the prediction accuracy varied,and the code with 3∶20∶10∶1 topological structure had the highest fitting superiority for multidimensional data and the smallest standard deviation of predicted values,then the BP neural network algorithm and topological structure suitable for subsidence rate prediction were finally determined.The results can provide an interactive procedure for the prediction of other surface movement parameters.
作者
李星
高建良
张学博
王春霞
LI Xing;GAO Jianliang;ZHANG Xuebo;WANG Chunxia(School of Safety Science and Engineering,Henan Polytechnic University,Jiaozuo Henan 454003,China;School of Mining and Civil Engineering,Liupanshui Normal University,Liupanshui Guizhou 553004,China)
出处
《中国安全生产科学技术》
CAS
CSCD
北大核心
2023年第6期90-97,共8页
Journal of Safety Science and Technology