一种基于机器学习算法的区域/单站ZTD组合预测模型

针对天顶对流层总延迟(ZTD)具有一定的时空变化特性,提出了一种基于BP神经网络、长短期记忆网络(LSTM)算法的区域/单站ZTD组合预测模型. 以连续14天香港连续运行参考站(CORS)网络18个监测站观测数据为例,利用BP神经网络、LSTM及本文算法进行了区域、单站及二者组合ZTD预测模型研究. HKWS测站的预测结果表明:利用前13天数据预报第14天数据,区域、单站、组合模型ZTD预测的均方根误差（RMSE）分别为10.2 mm、10.4 mm、8.5 mm,组合模型相对于区域、单站模型预测精度分别提升了17.2%、18.4%.      

Detection of clusters in traffic networks based on spatio‐temporal flow modeling

Spatio‐temporal clustering is a highly active research topic and a challenging issue in spatio‐temporal data mining. Many spatio‐temporal clustering methods have been designed for geo‐referenced time series. Under some special circumstances, such as monitoring traffic flow on roads, existing methods cannot handle the temporally dynamic and spatially heterogeneous correlations among road segments when detecting clusters. Therefore, this article develops a spatio‐temporal flow‐based approach to detect clusters in traffic networks. First, a spatio‐temporal flow process is modeled by combining network topology relations with real‐time traffic status. On this basis, spatio‐temporal neighborhoods are captured by considering traffic time‐series similarity in spatio‐temporal flows. Spatio‐temporal clusters are further formed by successive connection of spatio‐temporal neighbors. Experiments on traffic time series of central London's road network on both weekdays and weekends are performed to demonstrate the effectiveness and practicality of the proposed method.     

基于深度学习的人群活动流量时空预测模型

传统的时空预测方法缺乏对复杂时空非线性关系的描述,且难以顾及空间多尺度特征对于预测结果的影响.针对这一问题,本文提出了一种融合空间多尺度特征的时空网络模型(MST-Net),将流量预测的回归问题转换为具有时空特性的判别模型.首先,通过并联卷积提取空间多尺度特征;然后,通过引入注意力机制的门控循环单元提取时间特征;最后,...     

Spatio‐temporal prediction of land surface temperature using semantic kriging

Spatio‐temporal prediction and forecasting of land surface temperature (LST) are relevant. However, several factors limit their usage, such as missing pixels, line drops, and cloud cover in satellite images. Being measured close to the Earth's surface, LST is mainly influenced by the land use/land cover (LULC) distribution of the terrain. This article presents a spatio‐temporal interpolation method which semantically models LULC information for the analysis of LST. The proposed spatio‐temporal semantic kriging (ST‐SemK) approach is presented in two variants: non‐separable ST‐SemK (ST‐SemK_NSep) and separable ST‐SemK (ST‐SemK_Sep). Empirical studies have been carried out with derived Landsat 7 ETM+ satellite images of LST for two spatial regions: Kolkata, India and Dallas, Texas, U.S. It has been observed that semantically enhanced spatio‐temporal modeling by ST‐SemK yields more accurate prediction results than spatio‐temporal ordinary kriging and other existing methods.     

Extracting commuter-specific destination hotspots from trip destination data – comparing the boro taxi service with Citi Bike in NYC

ABSTRACT

Taxi trajectories from urban environments allow inferring various information about the transport service qualities and commuter dynamics. It is possible to associate starting and end points of taxi trips with requirements of individual groups of people and even social inequalities. Previous research shows that due to service restrictions, boro taxis have typical customer destination locations on selected Saturdays: many drop-off clusters appear near the restricted zone, where it is not allowed to pick up customers and only few drop-off clusters appear at complicated crossing. Detected crossings imply recent infrastructural modifications. We want to follow up on these results and add one additional group of commuters: Citi Bike users. For selected Saturdays in June 2015, we want to compare the destinations of boro taxi and Citi Bike users. This is challenging due to manifold differences between active mobility and motorized road users, and, due to the fact that station-based bike sharing services are restricted to stations. Start and end points of trips, as well as the volumes in between rely on specific numbers of bike sharing stations. Therefore, we introduce a novel spatiotemporal assigning procedure for areas of influence around static bike sharing stations for extending available computational methods.     

一种改进遗传神经网络的建筑基坑沉降预测模型

目前常见的沉降预测方法有灰色系统模型、时间序列分析法、BP神经网络及其改进算法等。针对BP神经网络容易出现过拟合和局部最优的缺点,部分学者利用遗传算法进行神经网络初始权值和阈值优化。但是遗传算法对于因监测数据质量问题而造成变形预测结果不佳的优化效果有限。因此引入自适应增强算法对遗传神经网络预测模型进行改进。并利用某高层建筑基坑实测50期监测数据进行仿真预测。实验结果表明,利用自适应增强算法改进之后的遗传神经网络预测模型在满足工程监测精度要求的前提下,在MAPE、MAE、MSE三项精度指标上分别提高80.57%、81.04%、70.83%。     

基于相空间重构和高斯过程回归的对流层延迟预测

天顶对流层延迟(zenith tropospheric delay,ZTD)是影响GPS定位精度的关键因素,为了提高ZTD的预测精度,提出一种基于相空间重构的高斯过程回归预测模型.针对ZTD时间序列的混沌特性,利用国际GNSS服务(International GNSS Service,IGS)站提供的ZTD数据,采用C...     

Urban event detection with big data of taxi OD trips: A time series decomposition approach

Big urban mobility data, such as taxi trips, cell phone records, and geo‐social media check‐ins, offer great opportunities for analyzing the dynamics, events, and spatiotemporal trends of the urban social landscape. In this article, we present a new approach to the detection of urban events based on location‐specific time series decomposition and outlier detection. The approach first extracts long‐term temporal trends and seasonal periodicity patterns. Events are defined as anomalies that deviate significantly from the prediction with the discovered temporal patterns, i.e., trend and periodicity. Specifically, we adopt the STL approach, i.e., seasonal and trend decomposition using LOESS (locally weighted scatterplot smoothing), to decompose the time series for each location into three components: long‐term trend, seasonal periodicity, and the remainder. Events are extracted from the remainder component for each location with an outlier detection method. We analyze over a billion taxi trips for over seven years in Manhattan (New York City) to detect and map urban events at different temporal resolutions. Results show that the approach is effective and robust in detecting events and revealing urban dynamics with both holistic understandings and location‐specific interpretations.     

大气可降水量预测模型的自适应Kalmam滤波改进

针对现有可降水量预报模型存在预报精度不高等问题,该文提出采用方差分量估计自适应卡尔曼滤波对可降水量数据进行预处理,用以提高径向基神经网络预测模型的预测精度,从而形成高精度预报模型。通过比较不同基站不同时间的数据,分析使用方法的预报精度。实验结果表明:将预测模型应用于全国7个测站进行实验,预测相对精度的平均值可达95%以上,预报残差在10-5左右,残差值小于0.001的占90%以上。在影响因素方面,使用较短时间作为模型原始数据进行预测会得到较好的预测结果。实验证明本预测方法在预报大气可降水量值方面具有较高的精度。     

A novel framework for detecting non-recurrent road traffic anomalies by combining a temporal graph convolutional network and hierarchical time memory detector

A non-recurrent road traffic anomaly refers to a sudden change in the capacity of a road segment, which deviates from the general traffic patterns, and is usually caused by abnormal traffic events such as traffic accidents and unexpected road maintenance. Timely and accurate detection of non-recurrent road traffic anomalies facilitates immediate handling to reduce the wastage of resources and the risk of secondary accidents. Compared with other types of traffic anomaly detection methods, prediction algorithms are suitable for detecting non-recurrent anomalies for their potential ability to distinguish non-recurrent anomalies from recurrent congestion (e.g., rush hours). A typical prediction algorithm detects an anomaly when the difference between the predicted traffic parameter (i.e., speed) and the actual one is greater than a threshold. However, the subjective setting of thresholds in many prediction algorithms greatly affects the detection performance. This study proposes a novel framework for non-recurrent road traffic anomaly detection (NRRTAD). The temporal graph convolutional network (T-GCN) model acts as the predictor to learn the general traffic patterns of road segments by capturing both the topological effects and temporal patterns of traffic flows, and to predict the “normal” traffic speeds. The hierarchical time memory detector (HTM-detector) algorithm acts as the detector to evaluate the differences between the predicted speeds and the actual speeds to detect non-recurrent anomalies without setting a threshold. In the experiments with traffic datasets of Beijing, NRRTAD outperformed other methods, not only achieving the highest detection rates but also exhibiting higher resilience to noise. The main advantages of NRRTAD are as follows: (1) adopting the T-GCN with a weighted graph to integrate differentiated connection strengths of multiple types of topological relations between road segments as well as temporal traffic patterns improves the prediction performance; and (2) utilizing a flexible mechanism in the HTM-detector to adapt to changing stream data not only avoids subjective setting of a threshold, but also improves the accuracy and robustness of anomaly detection.     

AN OVERVIEW OF GEOLOGICAL EXPERIMENTS CARRIED OUT ON BOARD MANNED SPACECRAFT AND SPACE STATIONS

An introductory paper to a series of articles on geological applications of remote sensing imagery produced from manned space flights chronicles a progression in research from the simple identification of geomorphological complexes during space flight to specific programs of observation and hand-held photography on board orbiting space stations. The coverage is primarily devoted to Soviet achievements over the period 1961–1982, especially to work conducted on board the space stations Salyut-5 through Salyut-7. Translated by Jay K. Mitchell, PlanEcon, Inc., Washington, DC 20005 from: B. M. Zubarev, V. V. Kozlov, and V. V. Lebedev, Kosmonavty issleduyut Zemlyu [Cosmonauts Study the Earth]. Moscow: Nauka, 1991, pp. 37–41.     

Measuring the structural similarity of network time prisms using temporal signatures with graph indices

The network‐time prism (NTP) is an extension of the space‐time prism that provides a realistic model of the potential pattern of moving objects in transportation networks. Measuring the similarity among NTPs can be useful for clustering, aggregating, and querying potential mobility patterns. Despite its practical importance, however, there has been little attention given to similarity measures for NTPs. In this research, we develop and evaluate a methodology for measuring the structural similarity between NTPs using the temporal signature approach. The approach extracts the one‐dimensional temporal signature of a selected property of NTPs and applies existing path similarity measures to the signatures. Graph‐theoretic indices play an essential role in summarizing the structural properties of NTPs at each moment. Two extensive simulation experiments demonstrate the feasibility of the approach and compare the performance of graph indices for measuring NTP similarity. An empirical application using bike‐share system data shows that the method is useful for detecting different usage patterns of two heterogenous user groups.     

城市交通热点区域识别及空间交互分析

出租车载客数据可以用于研究居民的出行特征,提取城市的交通热点区域,但对城市交通热点区域的交互关系研究相对较少。本文以纽约市的出租车行程记录数据为数据源,利用交通小区划分结合出租车载客数据提取城市交通热点区域,基于复杂网络的方法对不同日期类型和天气情况的城市交通热点区域空间交互网络进行研究并进行社区发现。结果表明,热点区域受城市核心区的影响而聚集在核心区域周围,城市内社区的形成可以克服地形和行政区域等因素的影响。研究结论有望为城市规划、城市交通管理、出租车调度、以及人们的出行等提供信息参考。     

The network K‐function in context: examining the effects of network structure on the network K‐function

The flaws of using traditional planar point‐pattern analysis techniques with network constrained points have been thoroughly explored in the literature. Because of this, new network‐based measures have been introduced for their planar analogues, including the network based K‐function. These new measures involve the calculation of network distances between point events rather than traditional Euclidean distances. Some have suggested that the underlying structure of a network, such as whether it includes directional constraints or speed limits, may be considered when applying these methods. How different network structures might affect the results of the network spatial statistics is not well understood. This article examines the results of network K‐functions when taking into consideration network distances for three different types of networks: the original road network, topologically correct networks, and directionally constrained networks. For this aim, four scenarios using road networks from Tampa, Florida and New York City, New York were used to test how network constraints affected the network K‐function. Depending on which network is under consideration, the underlying network structure could impact the interpretation. In particular, directional constraints showed reduced clustering across the different scenarios. Caution should be used when selecting the road network, and constraints, for a network K‐function analysis.     

Traffic transformer: Capturing the continuity and periodicity of time series for traffic forecasting

Traffic forecasting is a challenging problem due to the complexity of jointly modeling spatio‐temporal dependencies at different scales. Recently, several hybrid deep learning models have been developed to capture such dependencies. These approaches typically utilize convolutional neural networks or graph neural networks (GNNs) to model spatial dependency and leverage recurrent neural networks (RNNs) to learn temporal dependency. However, RNNs are only able to capture sequential information in the time series, while being incapable of modeling their periodicity (e.g., weekly patterns). Moreover, RNNs are difficult to parallelize, making training and prediction less efficient. In this work we propose a novel deep learning architecture called Traffic Transformer to capture the continuity and periodicity of time series and to model spatial dependency. Our work takes inspiration from Google’s Transformer framework for machine translation. We conduct extensive experiments on two real‐world traffic data sets, and the results demonstrate that our model outperforms baseline models by a substantial margin.     

Understanding satellite-derived bathymetry using Sentinel 2 imagery and spatial prediction models

ABSTRACT

Optical satellite data is an efficient and complementary method to hydrographic surveys for deriving bathymetry in shallow coastal waters. Empirical approaches (in particular, the models of Stumpf and Lyzenga) provide a practical methodology to derive bathymetric information from remote sensing. Recent studies, however, have focused on enhancing the performance of such empirical approaches by extending them via spatial information. In this study, the relationship between multibeam depth and Sentinel-2 image bands was analyzed in an optically complex environment using the spatial predictor of kriging with an external drift (KED), where its external drift component was estimated: a) by a ratio of log-transformed bands based on Stumpf’s model (KED_S) and b) by a log-linear transform based on Lyzenga’s model (KED_L). Through the calibration of KED models, the study objectives were: 1) to better understand the empirical relationship between Sentinel-2 multispectral satellite reflectance and depth, 2) to test the robustness of KED to derive bathymetry in a multitemporal series of Sentinel-2 images and multibeam data, and 3) to compare the performance of KED against the existing non-spatial models described by Stumpf et al. and Lyzenga. Results showed that KED could improve prediction accuracy with a decrease in RMSE of 89% and 88%, and an increase in R² of 27% and 14%, over the Stumpf and Lyzenga models, respectively. The decrease in RMSE provides a worthwhile improvement in accuracy, where results showed effective prediction of depth up to 6 m. However, the presence of higher concentrations of suspended materials, especially river plumes, can reduce this threshold to 4 m. As would be expected, prediction accuracy could be improved through the removal of outliers, which were mainly located in the channel of the river, areas influenced by the river plume, abrupt topography, but also very shallow areas close to the shoreline. These areas have been identified as conflictive zones where satellite-derived bathymetry can be compromised.     

优化短期余水位组合预测模型

针对现有非稳定非线性余水位预测模型较少和精度较低的问题,本文研究基于MEEMD算法与遗传优化BP神经网络的余水位组合预测模型。利用夏威夷岛4个长期验潮站获取的余水位时序数据,首先采用遗传算法MEEMD对余水位时序数据进行处理分析,得到较为稳定的余水位IMF分量;然后将经过遗传算法优化后分解的较为稳定的各个IMF分量作为BP神经网络预测模型的输入变量,分别建立12、24、48 h短期余水位的MEEMD遗传算法优化BP神经网络预测模型。通过与非优化BP神经网络预测模型结果进行对比分析,结果表明,优化前后均方根误差的偏差最高达2.03 cm,验证了预测24 h内的短期余水位仍保持其相关特性。该组合预测模型对于分析余水位变化规律和潮汐预报的精度、水位改正等均有重要意义。     

基于粒子群优化神经网络的概率积分法预计参数的确定

为有效确定概率积分法预计参数,提高预计值的精度。将粒子群优化(PSO)算法和BP神经网络进行融合,采用改进的混合粒子群优化算法优化神经网络的权值和阈值。在分析概率积分法参数与地质采矿条件之间关系的基础上,建立了基于PSO优化BP神经网络的概率积分法预计参数的优化选择模型。以我国典型的地表移动观测站资料为例,将计算结果与实际值进行了对比分析,并与文献[1]中改进BP算法进行了比较。结果表明,PSO-BP神经网络方法用于概率积分法预计参数的选取是可行的,收敛速度更快,计算精度更高。     

DeepIndoorCrowd: Predicting crowd flow in indoor shopping malls with an interpretable transformer network

Accurate and interpretable prediction of crowd flow would benefit business management and public security. The existing studies are challenged to adapt to the indoor environment due to its complex and dynamic spatial interaction patterns. In this study, we propose a crowd flow predicting method for indoor shopping malls, which simultaneously features temporal variables and semantic factors to suit the shopping mall environment. A deep learning model named DeepIndoorCrowd is presented. The model aims at capturing temporal dependencies and the semantic pattern in crowd flow to generate an accurate multi-horizon prediction. With a multi-term temporal dependency capturing structure, the model is effective in learning both daily and weekly patterns of the indoor crowd flow in a shopping mall and is able to provide the temporal interpretation of the prediction result. Moreover, a semantic-temporal fusion module is introduced to utilize the semantic information of stores in prediction, which has proved to be effective in enhancing the model's ability to learn temporal patterns. Experiments were conducted on a real-world dataset to verify the proposed approach. The ablation study demonstrates that the DeepIndoorCrowd can effectively improve the efficiency and accuracy of the prediction up to 18.7%. In addition, some interesting indoor crowd flow patterns were discovered by analyzing the model's interpretation of the prediction result. The proposed prediction method provides an intuitive way of modeling indoor crowd flow, and the experiment's outcome can help indoor managers better understand stores' flow traffic.     

改进后灰色神经网络及短期基坑沉降预测研究

在短期基坑沉降监测中,由于数据量少且呈非线性变化,沉降模型很难准确建立。灰色GM(1,1)对数据少、趋势性强、波动小的数据有较高的预测精度,但不能模拟复杂的非线性函数;BP神经网络可以对非线性数据进行学习训练,具有自学习、自适应能力;通过将GM(1,1)与BP神经网络组合,并优化网络部分的学习率、权值和阈值等,建立一种改进的灰色神经网络模型,该模型具有对非线性数据自学习、自适应能力和预测精度更高等优点。通过某基坑沉降监测分析,验证改进的灰色神经网络模型预测精度更高,适合短期建模,具有很好的实用性。