首页 | 本学科首页   官方微博 | 高级检索  
相似文献
 共查询到20条相似文献,搜索用时 968 毫秒
1.
应用前馈人工神经网络对广域单调的两组样本进行了模拟反演,引入单调前馈网络的概念对其权值和阈值定解问题和泛化能力进行了较诉研究。表明前馈人工神经网络是一个表达形式简单的复杂系统,其单调特征是隐性的,而且训练网络的成熟性对样本数量和样本内在规律性有一定依赖。强调了前馈人工神经网络的应用效果,指出单调与复合问题还需进一步深入研究。  相似文献   

2.
基于遥感分类的深度信念网络模型研究   总被引:1,自引:0,他引:1  
使用Python和TensorFlow作为工具实现基本的深度信念网络,提出了一种改进的Dropout策略,该策略每次只选择部分局部区域数据进行权重的清零,既保持了图像本身的局部信息,又增强了该模型的泛化能力,并使用差分进化进行神经网络的权值和偏移值的优化。通过实验对比表明,改进后的Dropout策略对于防止过拟合有明显的效果,使用差分进化算法对深度信念网络进行参数调整比传统的浅层分类器效果更好。  相似文献   

3.
基于深度信念网络的遥感影像识别与分类   总被引:1,自引:0,他引:1  
高分辨率遥感影像具有高维、多尺度、非平稳的内部特性和海量、多源、异构的外部特征,具有丰富的空间信息。探讨了利用新兴的深度信念网络研究高分辨率遥感影像的智能提取与分类,通过大量实验对比分类精度、Kappa系数以及参数敏感度分析,提出了网络层数、隐含层神经元个数、迭代次数等参数的最优设置方案,相比传统的浅层网络分类器,改进后的深度信念网络更好地拟合了样本的内在结构,遥感图像分类精度达到92%左右,取得了很好的分类效果。  相似文献   

4.
陈军  陈高 《地球科学》2000,25(6):651-655
利用具有全局优化功能的遗传算法直接反演多层密度界面.首先根据重力反演的特点对遗传算法进行改进, 使遗传算法基因交换过程中交换位置的确定同重力异常的拟合情况相结合, 给出适合于重力反演特点的遗传算法.然后利用改进后的遗传算法直接反演多层密度界面.理论模型和实际剖面的计算表明改进是有效的.   相似文献   

5.
基于微粒群算法的神经网络在岩性识别上的应用   总被引:1,自引:0,他引:1  
针对前馈多层神经网络BP算法存在的易陷入局部最优的不足,采用了微粒群算法(PSO)加以改进。通过胜利油田及其它油田的岩性识别所做的实例验证,并与BP神经网络进行比较,结果表明,在同等条件下,此方法效果较为满意。  相似文献   

6.
牙形石是一类在地层学研究中具有重要意义的古生物化石。但由于牙形石鉴定中存在诸多重要、却难以定量分析的“模糊特征”,使得牙形石的鉴定和分类工作长期依赖于古生物工作者的经验。利用数学中的非线性方程将这些“模糊特征”进行数学描述,并利用人工神经网络中的反向传播算法设计出一套多隐含层的前馈网络分类器,对这些“模糊特征”进行分类尝试,取得了良好的效果,为牙形石的进一步数值化研究奠定了基础。  相似文献   

7.
土壤侵蚀产沙量的人工神经网络模拟   总被引:12,自引:0,他引:12  
在分析土壤侵蚀产沙过程和神经网络模型特点具有某些相信的基础上,采用三层前馈网络模型(BP算法),模型的第一层有5个结点,分别代表降雨强度、降雨历时、降雨量、前期降雨量(用前10天降雨总量表示)、径流深;第三层只有一个结点,表示土壤侵蚀产沙量;隐层的结点数采用“试错法”确定为3个。利用四川某地水土保持试验观测资料,对土壤侵蚀产沙量进行模拟及预测,通过分析比较,显示了具有较子的模拟预测效果。  相似文献   

8.
杨琼波  崔东文 《水文》2023,43(1):17-23
针对月降水量时间序列多尺度非平稳性特点,提出小波包分解(WPD)-白骨顶鸟优化算法(COA)-极限学习机(ELM)相融合的降水量预测模型。首先,利用WPD将非平稳月降水量时间序列分解为若干子序列分量;然后在不同维度条件下利用6个典型函数对COA进行仿真测试;利用COA优化ELM输入层权值和隐含层偏值,对每一个子序列分量分别建立COA-ELM模型进行预测,将预测结果叠加重构后即为最终预测结果;最后,以云南省龙潭站汛期和主汛期月降水量数据为例进行实验,并与WPDCOA-BP、WPD-ELM、WPD-BP预测模型进行比较。结果表明:COA在不同维度条件下均具有较好的寻优精度和全局搜索能力。WPD-COA-ELM模型对实例汛期、主汛期月降水量时间序列预测的平均绝对百分比误差分别为3.91%、3.59%,预测精度优于WPD-COA-BP模型,远优于WPD-ELM、WPD-BP模型。WPD能科学降低月降水时间序列数据的复杂性,提高预测效果;COA能有效优化ELM输入层权值和隐含层偏值,提高ELM网络性能。  相似文献   

9.
顾元  朱培民  荣辉  曾凡平  海洋 《地球科学》2013,38(5):1143-1152
为了解决传统多地震属性的地震相分类方法中"难以引入先验信息用以指导分类,难以给出地震相分类结果可靠程度的定量估计,且各分类参数的权值较难确定"这3个问题,提出了一种新的基于贝叶斯网络的地震相分类方法.该分类方法有效地融合了先验信息和训练样本的分布特征,对提取的多种地震属性进行智能分析,以概率推理的方式得到各地震相类别的概率值,并根据概率分布估计分类结果的可靠程度.详述了贝叶斯网络用于地震相分类的原理与方法,并结合理论地震数据,验证了该方法的可行性和正确性.   相似文献   

10.
毕巍强 《地学前缘》2002,9(2):387-391
文中给出了一种研究全球变化中数据融合的新方法———粗集神经网络结构。用改进的方法训练网络权值 ,分析和仿真结果表明 ,此种模型具有较好的泛化、学习、映射能力。说明新模型能够解决传感器输出为二值或一个范围的多传感器数据融合问题。  相似文献   

11.
12.
Accumulation of organic matter in fens of fluvial valleys is often related to a low terrigenous matter delivery and to palaeoenvironmental conditions inducing low mechanical erosion. These assumptions come from the interpretation of contents in organic (MO) and mineral (MM) matters in sediments, expressed in percents, and then exactly anticorrelated. Calculation of mass accumulation rates of MO (TaMO) and MM (TaMM), expressed in g?m?2?yr?1, shows that TaMO and TaMM generally are not anticorrelated and that high values of TaMO and TaMM could appear simultaneously. That expression of MO and MM accumulation makes it possible to precise the climatic and human impact on sedimentation. To cite this article: J.-J. Macaire et al., C. R. Geoscience 337 (2005).  相似文献   

13.
In situ X-ray diffraction was used to measure the isothermal bulk modulus at room conditions (KT0) of synthetic olivines with different iron contents. The chemical formulae of the olivine samples were (Fex,Mg1?x)2SiO4 with x = 0.45; 0.64; 0.82; 1, with 1% standard deviation (referenced as Fa45, Fa64, Fa82 and Fa100, respectively). All experiments were performed in the multi-anvil apparatus installed at NSLS beamline X17B2, to pressures up to about 7 GPa. Unit-cell volumes under hydrostatic conditions and differential stresses present in the samples were calculated using the method developed by Singh et al. (1998), and pressures measured using NaCl as a standard were then corrected for these stresses. Using a second-order Birch–Murnaghan equation of state, we obtained the isothermal bulk modulus of each composition: KT0Fa45=131.4±2.6 GPa, KT0Fa64=132.1±3.1 GPa, KT0Fa82=136.3±1.7 GPa and KT0Fa100=134.8±1.4 GPa. These values combined with data available in the literature show that the KT0 of Fe-rich olivines increases very slowly with the Fe content, but possibly not in a simple linear trend.  相似文献   

14.
We studied the particle growth in a protoplanetary disk in a high-ionization environment and found that icy planet formation is inactive for a disk with an ionization rate 100 times higher than that of the present Solar System. In particular, in the case of M 10~(-7.4)M_☉yr~(-1), only rocky planet formation occurs. In such a case, all the solid materials in the disk drift inward, eventually reach the inner MRI front,and accumulate there. They form a dense, thin sub-disk of solid particles, which undergoes gravitational instability to form rocky planetesimals. The planetesimals rapidly grow into a planet through pebble accretion. Consequently, rocky planets tend to be much larger than planets formed through other regimes(tandem planet formation regime and dispersed planet formation regime), in which icy planet formation actively takes place. These rocky planets may evolve into hot Jupiters if they grow fast enough to the critical core mass of the runaway gas accretion before the dispersal of the disk gas, or they may evolve into super-Earths if the gas dispersed sufficiently early.  相似文献   

15.
16.
17.
18.
Statistical parameters of waves on the south Aquitaine coast have been determined by comparison of two datasets: in situ local measurements on a short duration and numerical simulation outputs from the WAVEWATCH III model over a period of 6 years and 9 months. Data and simulation results are well correlated in the case of the significant wave height, whereas in the case of the peak periods (Tp) and directions (Dp), this correlation is lower. This is due respectively to the intrinsic unstable nature of Tp and the narrow distribution of Dp. An adjustment of simulation data on measurement is then used to give an evaluation of the wave statistics for the southern fraction of the French Atlantic coast. Eventually, the relationships between the three statistical variables (Hs,Tp,Dp) are analysed in order to determine representative sea states. The classification obtained shows the predominance of swell regimes, distributed on a very narrow angular sector over wind sea regimes. To cite this article: S. Abadie et al., C. R. Geoscience 337 (2005).  相似文献   

19.
20.
"数字国土"在"数字地球"中的地位及其模型探讨   总被引:6,自引:0,他引:6  
建立“数字国土”是“数字地球”计划在国土资源领域的个体体现,借鉴点源信息系统的研究成果和“多”技术集成化研究基础,阐明“多S”技术集成的点源信息系统技术是“数字国土”的支撑技术,并构建了“数字国土”模型。  相似文献   

设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号