基于模拟退火的Gauss-Newton算法的神经网络在渗流反分析中的应用

提出了模拟退火的Gauss-Newton算法的神经网络,克服了经典BP网络存在的一些缺陷。并以正弦函数的迭代收敛为例,证明了该方法的正确性,有效性和优越性。同时将该方法用于同乐坪大坝的渗流反分析,利用反演出的渗透系数进行渗流场计算。得到的水头预报值与观测值相吻合,可知反演结果是正确的,说明该方法用于实践工程的渗流参数识别是可行的。     

CONVEX MODEL FOR SEISMIC DESIGN OF STRUCTURES—II: DESIGN OF CONVENTIONAL AND ACTIVE STRUCTURES

The optimal design of the members of conventional structures or structures equipped with active bracing systems, known as active structures, is presented for uncertain excitations. Three approaches are used for obtaining the optimal structural design: (1) the time-history analysis of an actual earthquake record (AR), (2) the global energy-bound convex model adjusted with an excitation-specific reduction factor (RGEB), and (3) the global energy-bound convex model adjusted with an average reduction factor (ARGEB) for a set of excitations with common characteristics. The optimal structures obtained using the RGEB and ARGEB convex models have different sizes for their conventional members from the designs based on a time-history analysis of the actual earthquake (AR). The optimal design of the structure is carried out using a modified annealing algorithm. The advantage of using convex models to perform the optimization is that they represent a more general excitation than a single earthquake record. In addition, the RGEB and ARGEB convex models require considerably less computational effort since the constraints of the optimization become time-independent. A comparison between optimal designs of structures with conventional members only, and active structures indicates that the latter are more efficient by combining the conventional and active members.     

用转换函数研究那木错站地壳结构

青藏高原是研究地球动力学问题天然的试验场,也是研究全球变化的关键地区之一,中国科学院青藏高原研究所于2004年在西藏那木错建立了圈层相互作用综合观测研究站.为了解那木错站下方的地质构造,于2005年8月在本站布设了宽频带地震仪(记录器为Reftek130,摆为STS2),并于2006年5月取得首批数据.本文利用宽频带地震仪提供的三分量地震波形记录,应用转换函数及快速模拟退火算法对那木错站下的地壳横波速度结构进行了反演.反演结果表明,那木错站Moho面深度在85公里左右,地壳结构复杂,尤其在中上地壳,明显呈高低速互层结构,反映了该地区构造活动、物质交换活跃,表明这些地区还未达到均衡;同时,为了便于研究该地区地壳结构,还对IRIS提供的临近地区的地震宽频带进行了处理,为研究那木错地区圈层结构提供地球物理依据.     

裂变径迹法在研究造山带隆升过程中的应用介绍

介绍了４种类型的造山带隆升－冷却模式,不同模式对应不同的磷灰石裂变径年龄－－高程图特征;了不同构造造背景下的热历史特征一裂变径迹年龄序列。裂变径变宅遮挡同统计分布型式对应于不同的热历史,具有不同的年代学意义。     

Grain Coarsening of Hematite in the Gushan Iron Deposit,Anhui Province,China

The iron ores of the Gushan mine occur in the contact zone of a Mesozoic diorite intrusion and are composed primarily of hematite microcrystallites and chalcedony,The hematite microcrystallites have undergone post-mineralization recrystallization and coarsening with resultant formation of lath-shaped hematite porphyroblasts.Microscopic investigation reveals that recrystallization and coarsening of the hematite ores of the Gushan mine took place without the formation of new nuclei,due to the coalescence of the microcrystallites.The whole process could have begun with the mutual approach of the microcrystallites,followed by grain rotation to realize paralleism and ending by the welding of these grains to form optically homogeneous porphyroblastic hematite.     

磷灰石裂变径迹技术与地学应用综述

天然裂变径迹的形成和演化是一个连续不可逆的过程，经历了漫长地质演化的矿物的所有裂变径迹综合起来就可以看作一个热时钟体系。它不仅反映了矿物冷却到封闭温度(退火温度)的时间，还记录了矿物所经历的低于封闭温度的整个热历史的信息。由此可以建立起用于恢复矿物热史信息的各种模型(包括时间——温度模型)。这些模型的建立对沉积盆地沉降／埋藏历史、造山带浅部壳层冷却／剥露历史的研究以及低温热历史重建等地学方面的应用意义重大。     

遗传模拟退火的BP算法在冲击地压中的应用

冲击地压的预测、预报的研究,大多数仍停留在简单的统计研究和单因素的预测方面,因而,结果也不十分理想。笔者采用多层前向网络对该问题进行数学建模,网络的训练算法采用基于遗传模拟退火的BP优化算法。该算法是在遗传算法中引入模拟退火机制,将其同BP算法结合,形成一个混合的优化算法。新算法既有神经网络的学习能力和鲁棒性,又有遗传算法的强全局随机搜索能力。同时,利用华丰矿冲击地压的实际监测数据,通过遗传算法的主要性能指标对新算法的参数进行了比较研究,得到优化后的一组参数。利用该参数,对冲击地压的神经网络模型的结构、权值和阈值进行了优化,得到了非全连接的优化神经网络模型。最后,利用该模型对华丰矿冲击地压进行了短期最大震级的预报。预测结果的相对误差率平均为 7.84 %,预测效果比较理想。     

一种全局优化的隐式交错网格有限差分算法及其在弹性波数值模拟中的应用

在数值模拟中,隐式有限差分具有较高的精度和稳定性.然而,传统隐式有限差分算法大多由于需要求解大型矩阵方程而存在计算效率偏低的局限性.本文针对一阶速度-应力弹性波方程,构建了一种优化隐式交错网格有限差分格式,然后将改进格式由时间-空间域转换为时间-波数域,利用二范数原理建立目标函数,再利用模拟退火法求取优化系数.通过对均匀模型以及复杂介质模型进行一阶速度-应力弹性波方程数值模拟所得单炮记录、波场快照分析表明:这种优化隐式交错网格差分算法与传统的几种显式和隐式交错网格有限差分算法相比不但降低了计算量,而且能有效的压制网格频散,使弹性波数值模拟的精度得到有效的提高.     

Ensemble Forecast for Tropical Cyclone Based on CNOP-P Method: A Case Study of WRF Model and Two Typhoons

In this paper, we set out to study the ensemble forecast for tropical cyclones. The case study is based on the Conditional Nonlinear Optimal Perturbation related to Parameter (CNOP-P) method and the WRF model to improve the prediction accuracy for track and intensity, and two different typhoons are selected as cases for analysis. We first select perturbed parameters in the YSU and WSM6 schemes, and then solve CNOP-Ps with simulated annealing algorithm for single parameters as well as the combination of multiple parameters. Finally, perturbations are imposed on default parameter values to generate the ensemble members. The whole proposed procedures are referred to as the Perturbed Parameter Ensemble (PPE). We also conduct two experiments, which are control forecast and ensemble forecast, termed Ctrl and perturbed-physics ensemble (PPhyE) respectively, to demonstrate the performance for contrast. In the article, we compare the effects of three experiments on tropical cyclones in aspects of track and intensity, respectively. For track, the prediction errors of PPE are smaller. The ensemble mean of PPE filters the unpredictable situation and retains the reasonably predictable components of the ensemble members. As for intensity, ensemble mean values of the central minimum sea-level pressure and the central maximum wind speed are closer to CMA data during most of the simulation time. The predicted values of the PPE ensemble members included the intensity of CMA data when the typhoon made landfall. The PPE also shows uncertainty in the forecast. Moreover, we also analyze the track and intensity from physical variable fields of PPE. Experiment results show PPE outperforms the other two benchmarks in track and intensity prediction.     

多导体柱微波成像的连续编码遗传算法

提出一种能同时重建多个二维良导体目标外形轮廓的新方法——连续编码遗传算法．将目标的横截面轮廓近似表达为三角级数形式,由边界条件得到一积分方程组,在此基础上将成像问题转化为约束优化问题,级数的各项系数为待优化量．积分方程组为约束条件,目标函数定义为实际测量的散射场与反演过程中得到的散射场之间的相对误差函数．采用连续编码遗传算法求解,优化过程通过选择、交叉、变异等遗传操作的选代而实现,待优化目标函数进行线性变换并采用模拟退火原理确定目标后表达为适应度函数,采用联赛选择与比例选择相结合的选择机制和单点交叉方式,变异操作通过对基因施加微小随机扰动实现．上一代中适应度最高的个体直接保留．数值模拟反演实验验证了方法的有效性．与其他反演方法相比,本法具有简单、通用、鲁棒性强等特点．