低温雨雪过程的粒子群-神经网络预报模型

利用逐日气温和降水量数据、NCEP/NCAR再分析资料以及预报场资料,通过分析提取我国南方区域持续性低温雨雪过程及其预报因子,使用粒子群-神经网络方法建立非线性的统计集合预报模型 (PSONN-EPM),对我国南方区域持续性低温雨雪过程进行预报试验。结果表明:以过程的冷湿程度及影响范围为标准,将低温雨雪过程分为一般过程和严重过程,并建立不同的预报模型效果较好。通过10 d独立样本预报试验看,基于粒子群-神经网络方法建立的集合预报模型比基于逐步回归方法建立的预报模型的预报平均相对误差小,对严重过程预报能力高于对一般过程预报,且这种非线性统计集合建模方法在建模过程中不需要调整神经网络参数,在实际预报业务中值得尝试。     

基于LSTM的海洋表面短期风速预测研究

为实现对海面风速精确的短期预测,提出了一种基于长短期记忆(LSTM,longshort-termmemory)神经网络的短期风速预测模型,选取OceanSITES数据库中单个浮标站点采集的风速历史数据作为模型输入,经过训练设置最佳参数等步骤,实现了以LSTM方法,对该站点所在海区海面风速在各季节性代表月份海面风速的24 h短期预测。同时通过不同预测时长的实验以及与BP(back propagation)神经网络神经网络和径向基函数神经网络(radialbasisfunctionneuralnetwork,RBF)的预测效果对比实验,证明了LSTM预测方法相比上述两种神经网络预测方法,在海表面风速预测应用中的优越性。最后通过多个海域对应的站点风速数据预测实验,证明了LSTM神经网络模型的普遍适用性,由相关系数和预测误差的分析可知该方法具备应对急剧变化数据的预测稳定性,可以作为海洋表面风速短期预测的一种可靠方法。     

Operational Evaluation of the Quantitative Precipitation Estimation by a CINRAD-SA Dual Polarization Radar System

In this paper, the quantitative precipitation estimation based on hydrometeor classification (HCA-QPE) algorithm was proposed for the first operational S band dual-polarization radar promoted by the CINRAD/SA radar of China. The HCA-QPE algorithm, the localized Colorado State University-Hydrometeor Identification of Rainfall (CSU-HIDRO) and Joint Polarization Experiment (JPOLE) algorithms, and the dynamic Z-R relationships based on variational correction QPE (DRVC-QPE) algorithm were evaluated with the rainfall events from March 1 to October 30, 2017 in the Guangdong Province. The results indicated that even though the HCA-QPE algorithm did not use the observed rainfall data for correction, its estimation accuracy was better than that of the DRVC-QPE algorithm when the rainfall rate was greater than 5mm h^(-1); and the stronger the rainfall intensity, the greater the QPE improvement. And the HCA-QPE algorithm worked better than the localized CSU-HIDRO and JPOLE algorithms. This study preliminarily verified the improved accuracy of QPE by a dual-polarization radar system modified from CINRAD-SA radar.     

基于随机补片和DeepLabV3+的建筑物遥感图像变化检测

为有效解决传统遥感图像变化检测预处理复杂的问题,提出一种基于随机补片和DeepLabV3+的建筑物遥感图像变化检测方法。以ResNet50特征提取网络为基础,创建DeepLabV3+语义分割网络,并在图像和标签中创建大小为224像素×224像素的随机补片作为网络输入,训练建筑物提取网络;修改建筑物提取网络输入层为6通道,通过矩阵运算将两期遥感图像转换为一幅6通道非RGB图像,利用转换后的非RGB图像进行网络训练并验证变化检测精度。实验1利用ENVI5.3软件,采用马氏距离法进行变化检测;实验2采用改进的U-Net网络和随机补片,完成网络训练和精度验证;实验3使用实验2的训练数据和验证数据,采用随机补片和DeepLabV3+网络进行变化检测网络训练及精度验证。实验结果表明,该方法实验1、实验2、实验3建筑物变化检测平均交并比分别为24.43%、83.14%、89.90%,边界轮廓匹配分数分别为61.47%,80.24%、96.51%。     

Sea Level Prediction in the Yellow Sea From Satellite Altimetry With a Combined Least Squares-Neural Network Approach

Accessible high-quality observation datasets and proper modeling process are critically required to accurately predict sea level rise in coastal areas. This study focuses on developing and validating a combined least squares-neural network approach applicable to the short-term prediction of sea level variations in the Yellow Sea, where the periodic terms and linear trend of sea level change are fitted and extrapolated using the least squares model, while the prediction of the residual terms is performed by several different types of artificial neural networks. The input and output data used are the sea level anomalies (SLA) time series in the Yellow Sea from 1993 to 2016 derived from ERS-1/2, Topex/Poseidon, Jason-1/2, and Envisat satellite altimetry missions. Tests of different neural network architectures and learning algorithms are performed to assess their applicability for predicting the residuals of SLA time series. Different neural networks satisfactorily provide reliable results and the root mean square errors of the predictions from the proposed combined approach are less than 2?cm and correlation coefficients between the observed and predicted SLA are up to 0.87. Results prove the reliability of the combined least squares-neural network approach on the short-term prediction of sea level variability close to the coast.     

边坡稳定分析的神经网络改进模糊点估计法

边坡稳定性分析中,模糊点估计法能同时考虑模糊不确定性和随机不确定性因素。针对传统模糊点估计法计算工作量大的缺点,提出一种神经网络改进模糊点估计法。利用拉丁超立方抽样法和径向基函数神经网络（RBF）建立边坡安全系数的预测模型;对黏聚力和内摩擦角等模糊随机变量取λ截集,并在各截集水平对参数进行组合;利用建立的预测模型对各参数组合的安全系数进行预测;最后由统计矩点估计法计算边坡的可靠度指标。实例分析表明：改进模糊点估计法使用方便、结果可靠,且能通过增加λ截集水平的数目来提高计算精度。对于含有2～4个模糊随机变量的边坡,采用改进模糊点估计法计算可靠度时λ截集水平的数目可近似取25。     

福建强震台网数据记录系统简介

本文概要介绍了强震仪主要技术参数、强震台网计算机系统及强震记录情况。     

Numerical estimation of REV and permeability tensor for fractured rock masses by composite element method

The Monte Carlo method is used to generate parent stochastic discrete fracture network, from which a series of fractured rock samples of different sizes and orientations are extracted. The fracture network combined with a regular grid forms composite element mesh of the fractured rock sample, in which each composite element is composed of sub‐elements incised by fracture segments. The composite element method (CEM) for the seepage is implemented to obtain the nodal hydraulic potential as well as the seepage flow rates through the fractured rock samples. The application of CEM enables a large quantity of stochastic tests for the fractured rock samples because the pre‐process is facilitated greatly. By changing the sizes and orientations of the samples, the analysis of the seepage characteristics is realized to evaluate the variation of the permeability components, the existence of the permeability tensor and the representative element volume. The feasibility and effectiveness are illustrated in a numerical example. Copyright © 2008 John Wiley & Sons, Ltd.     

Numerical and Experimental Analysis of A Ducted Propeller Designed by A Fully Automated Optimization Process Under Open Water Condition

A fully automated optimization process is provided for the design of ducted propellers under open water conditions, including 3D geometry modeling, meshing, optimization algorithm and CFD analysis techniques. The developed process allows the direct integration of a RANSE solver in the design stage. A practical ducted propeller design case study is carried out for validation. Numerical simulations and open water tests are fulfilled and proved that the optimum ducted propeller improves hydrodynamic performance as predicted.     

高轨卫星轨道预报中神经网络模型优化设计

高轨卫星是我国卫星导航系统的重要组成部分。提升该类卫星的轨道预报精度有利于用户定位精度的提高。提出了一种改进高轨卫星轨道预报精度的新方法。该方法避开了精化动力学模型的困难,尝试从轨道预报误差的规律中寻找突破。利用神经网络作为建立预报模型的工具,将某历史时刻的轨道预报误差作为训练样本,利用训练好的神经网络模型补偿当前时刻的预报轨道以提高轨道预报精度。对影响神经网络模型补偿效果的各因素进行了详细分析,制定了适应于高轨卫星短期、中期和长期预报的神经网络最优模型。利用实测数据进行了试验分析,结果表明:预报8,15及30 d应选择的训练步长分别为10,20及25 min;轨道预报8~30 d时,训练噪声均选取0.01。神经网络模型有效地改进了高轨卫星的轨道预报精度,预报4~30 d,轨道精度提高幅度为34.67%~82.37%不等。