基于NSGA-BP神经网络算法的高密度电法非线性反演

发布时间：2018-03-29 17:13

  本文选题：高密度电阻率法　切入点：非线性反演　出处：《山东科技大学》2017年硕士论文

【摘要】：煤炭企业在国民经济和社会发展中起着及其重要的地位。我国作为世界上主要的产煤国的同时,也是受矿井灾害最严重的国家之一。目前随着我国煤炭开采强度不断加强,深度不断增加,大型煤田越来越多,矿井地质灾害的威胁也愈来愈严重。地质勘探是煤矿安全生产的基础,在煤矿地质灾害的防治中扮演着举足轻重的作用。高密度电阻率法作为是一种高效的电阻率勘探方法,在矿山地质安全的勘探与评价中被广泛应用。在高密度电法资料的解释方面,目前仍以最小二乘法为代表的线性反演为主,但是这种方法的反演精度并不是很高,因此越来越多的学者投入于非线性反演的研究中,基于BP神经网络的高密度电法的反演成为非线性反演中一个较为活跃的分支。本文针对BP神经网络因权值阈值随机初始化导致的收敛缓慢,易陷入局部最小的不足,结合神经网络节点权值数量级越小,网络泛化能力越强的特点,采用多目标优化算法(NSGA-Ⅱ算法)与BP神经网络算法联合演算,以BP神经网络的训练均方误差MSE和隐含层参数的均方根值同时作为目标函数,对BP神经网络进行多目标优化,从而提高BP神经网络对高密度电法数据的反演精度。本文通过计算机模型介绍了基于NSGA-BP神经网络算法的高密度电法反演方法和具体流程,反演结果表明NSGA-Ⅱ算法能够有效优化了 BP神经网络的权值和阈值,提高BP算法的全局寻优性能和神经网络的泛化能力,比传统非线性反演算法和单一 BP神经网络算法的反演结果更准确。最后以梁家煤矿ZK60号地面注浆孔为研究对象,开展了高密度电法的测量,应用NSGA-BP神经网络方法进行反演分析。结果表明,该方法能够有效的应用于高密度电法实测资料的解释中,为钻孔注浆效果评价提高可靠的依据。
[Abstract]:Coal enterprises play an important role in the development of national economy and society. As the main coal-producing countries in the world, China is also one of the countries most seriously affected by mine disasters. With the increasing of depth and the increasing number of large-scale coal fields, the threat of mine geological hazards is becoming more and more serious. Geological exploration is the basis for the safe production of coal mines. It plays an important role in the prevention and control of coal mine geological hazards. High density resistivity method is an efficient resistivity exploration method. It is widely used in the exploration and evaluation of mine geological safety. At present, the linear inversion represented by the least square method is still the main method in the interpretation of high-density electrical data, but the inversion accuracy of this method is not very high. Therefore, more and more scholars are engaged in the research of nonlinear inversion. The inversion of high density electrical method based on BP neural network has become an active branch of nonlinear inversion. In this paper, the convergence of BP neural network due to random initialization of weight threshold is slow, and it is easy to fall into local minimum. Combined with the characteristics that the weight of the neural network is smaller, the generalization ability of the network is stronger, the multi-objective optimization algorithm (NSGA- 鈪，

本文编号：1682072