基于LBS和深度学习的旅游景区客流量的高时频预测

为实现精准的旅游景区客流量的高时频预测,本研究构建了一套基于LBS和深度学习模型的预测方法。此方法可通过对LBS数据的转换实现预测的空间范围与时频控制,并通过方法的核心模型——基于双向循环神经网络和GRU算法构建的深度双向GRU(DBi-GRU)模型完成预测。为检验方法的有效性,研究以深圳大梅沙海滨公园为例对方法进行实验测试。实验使用拟合曲线、误差指标及DM检验3种方法评估DBi-GRU模型的预测效果。此外,实验还设置了其他五种深度学习模型作为DBi-GRU的对照模型,测试基于不同深度学习算法的模型之间的预测水平差异。实验结果表明:(1)本研究提出的DBi-GRU模型在景区客流量高时频预测中具有理想的预测效果,在高峰时段的客流量预测方面也具有较高准确性,预测效果明显优于其他深度学习模型;(2)基于双向循环网络的模型的效果普遍优于基于常规循环网络的模型。尤其是基于双向LSTM算法的模型,虽然预测的准确度略逊色于DBi-GRU模型,但在模型性能上与其的差异并不显著;(3)在相同网络参数下,GRU算法较前人采用的LSTM和RNN算法有着更高的预测准确性。本研究为客流量预测领域的研究提供了一种...     

江西地区地方性震级的量规函数与台基校正值研究

本文采用《江西测震台网地震观测报告》,选取2007年10月~2015年12月所记录的499次M_L≥1.5地震事件,对各子台测定震级与台网平均震级偏差进行定量的统计分析,从震级偏差频次分布、量规函数、台基、方位角等方面分析产生震级偏差的原因。在对量规函数和台基进行校正后,震级偏差绝对值在0.2以内的样本数达到了68.6%,并给出了适合江西地区的量规函数和台基校正值。     

新疆数字化地震台网业务运行管理系统

新疆数字化地震台网业务运行管理系统具有对地震台站业务考评、网络及设备信息、台站参数、系统运维、数据共享、大震速报、速报编目数据统计和台网资料等信息管理功能。该系统在Windows 7系统环境下,使用MyEclipse平台和MySQL数据库通过JSP技术开发动态Web Service系统,使用Delphi XE10环境开发移动APP。在新疆地震台网投入运行后,解决了日常业务中台站与台站、台网与台站数据和信息交换困难的问题,有效提高了新疆地震台网的运行质量监管力度,方便用户使用各类数据,有较好的推广应用价值。     

径向基神经网络GPS高程转换方法

针对GPS高程转换问题,给出了基于径向基神经网络转换GPS高程的模型。用实际观测数据对该模型进行了试验,结果表明,用径向基神经网络转换GPS高程精度高于二次拟合法和BP神经网络法。径向基神经网络能够有效克服BP神经网络局部极小值的缺点,并且具有较高的收敛速度,在GPS高程转换方面具有广阔应用前景。     

������׼վ���������IJⶨ����

???FG5????????????й?????????????????12?????????????????????????????????????????????????5??10 ^-8 ms ^-2??????????????峱????????????????????FG5??????????????Ms6.0???????????     

Earthquake detection probability within a seismically quiet area: application to the Bruchsal geothermal field

In applications such as oil and gas production, deep geothermal energy production, underground storage, and mining, it is common practice to implement local seismic networks to monitor and to mitigate induced seismicity. For this purpose, it is crucial to determine the capability of the network to detect a seismic event of predefined magnitude in the target area. The determination of the magnitude of completeness of a network is particularly required to properly interpret seismic monitoring results. We propose a method to compute the detection probability for existing local seismic networks, which (i) strictly follows the applied detection sequence; (ii) estimates the detection capability where seismicity has not yet occurred; and (iii) delivers the results in terms of probabilities. The procedure includes a calibration of a local magnitude scale using regional earthquakes recorded by the network and located outside the monitored area. It involves pre‐processing of the seismograms recorded at each station as performed during the triggering sequence, which is assumed based on amplitude thresholds. Then, the calibrated magnitude–distance–amplitude relations are extrapolated at short distances and combined to reproduce the network detection sequence. This generates a probability to detect a seismic event of a given magnitude at a specified location. This observation‐based approach is an alternative to a fully theoretical detection capability modelling and includes field conditions. Seismic wave attenuation by geometrical spreading and intrinsic attenuation, site effect, and instrumental responses are partly accounted for by the calibration. We apply this procedure on the seismic network deployed in the Bruchsal geothermal field (Germany). Although the system was in good working order, no induced seismicity was identified in the area between June 2010, when monitoring started, and November 2012. The recording of distant seismicity during this time period, however, allowed the application of the proposed procedure. According to the applied network detection parameters, the results indicate that the absence of seismicity can be interpreted as a 95% probability that no seismic event with M_L ≥ 0.7 occurred below the network at 2.4‐km depth, i.e., in the geothermal reservoir.     

中国地震台网初至P波区域三维走时表的建立

基于LLNL-G3Dv3全球P波三维速度模型,应用FMM软件包计算并建立了中国地震台网990个台站的初至P波区域三维走时表．该走时表覆盖了以台站为中心的水平向20°×20°、 垂直向-5.1—80 km （向下为正）的三维空间. 其水平向间隔为0.2°,垂直向间隔为5 km．这样对于任一深度小于80 km的震源,均可以应用此三维走时表计算其到周围10°范围内台站的走时．中国地震台网初至P波区域三维走时表的建立,对于改善区域初至P波走时预测,提高地震定位精度有一定现实意义．      

Structure of residual oil as a function of wettability using pore-network modelling

In the water flooding of mixed-wet porous media, oil may drain down to relatively low residual oil saturations (S_or). Various studies have indicated that such low saturations can only be reached when oil layers in pore corners are included in the pore-scale modelling. These processes within a macroscopic porous medium can be modelled at the pore-scale by incorporating the fundamental physics of capillary dominated displacement within idealised pore network models. Recently, the authors have developed thermodynamic criteria for oil layer existence in pores with non-uniform wettability which takes as input geometrically and topologically representative networks, to calculate realistic S_or values for mixed-wet and oil-wet sandstones [16, 21]. This previous work is developed in this paper to include (i) the visualisation of the 3D structure of this residual oil, and (ii) a statistical analysis of this “residual/remaining” oil. Both the visualisation and the statistical analysis are done under a wide range of wettability conditions, which is reported for the first time in this paper.The structure of residual oil for strongly water wet systems is well known (where residual = remaining oil) and our model agrees with this but this structure changes radically for mixed wet systems (where residual ≠ remaining) and this has not yet been visualised experimentally. We find that for more water-wet systems high final residual oil saturations are reached at relatively small amounts of water injected and this oil is present in the pores as bulk oil. On the other hand, for more oil-wet systems we find a slow decrease of the amount of remaining oil with increasing amounts of injected water. During the process, the remaining connectivity of the oil phase is increasingly provided by oil layers only, hence the slow drainage. The final residual oil saturation, only reached in the theoretical limit of an infinite amount of injected water, is almost entirely contained in large number of (relatively low volume) oil layers, which are present in pores of most radius sizes.     

北京地区主要活断层活动参数的重力反演

采用1991~1999年首都圈重力测网的重力测量资料,根据地质调查结果确定断层初始运动模型,利用稳健—贝叶斯最小二乘方法反演北京地区4条活断层的活动参数(走滑、倾滑、引张)。反演结果表明:在1991~1999年间,北京地区活断层的运动性质均以走滑和垂直升降为主,兼有伸展特征,但运动速率都不大,约为2 mm/a。北京地区北西走向断层反演的运动特征为左旋走滑正断层,北东走向断层反演的运动特征为右旋走滑正断层。     

临汾市地震台网网内地震震级与山西省地震台网的对比研究

选取2012年3月1日至12月31日临汾市地震台网记录到的ML1.0以上地震,对每个地震进行重新定位,并与山西地震台网的定位结果进行对比分析,找到两台网所定震级的偏差系数,用以修正临汾台网的震级,提高测定结果的精度。