Estimating historical hourly traffic volumes via machine learning and vehicle probe data: A Maryland case study

This paper focuses on the problem of estimating historical traffic volumes between sparsely-located traffic sensors, which transportation agencies need to accurately compute statewide performance measures. To this end, the paper examines applications of vehicle probe data, automatic traffic recorder counts, and neural network models to estimate hourly volumes in the Maryland highway network, and proposes a novel approach that combines neural networks with an existing profiling method. On average, the proposed approach yields 24% more accurate estimates than volume profiles, which are currently used by transportation agencies across the US to compute statewide performance measures. The paper also quantifies the value of using vehicle probe data in estimating hourly traffic volumes, which provides important managerial insights to transportation agencies interested in acquiring this type of data. For example, results show that volumes can be estimated with a mean absolute percent error of about 21% at locations where average number of observed probes is between 30 and 47 vehicles/h, which provides a useful guideline for assessing the value of probe vehicle data from different vendors.     

Modeling the cost of bird strikes to US civil aircraft

The objective of our analysis is to develop a model of damage costs that arise from collisions between aircraft and birds, based on data drawn from the Federal Aviation Administration National Wildlife Strike Database (NWSD). We develop a two-part model, composed of two separate statistical models, that accounts for the effects of aircraft mass category, engine type, component of the aircraft struck, and the size and number of birds struck. Our results indicate the size of bird, number of birds, and engine ingestions are the largest determinants of strike-related costs. More generally, our result is a model that provides a better understanding of the determinants of damage costs and that can be used to interpolate the substantial amount of missing data on damage costs that currently exists within the NWSD. A more complete accounting of damage costs will allow a better understanding of how damage costs vary geographically and temporally and, thus, enable more efficient allocation of management resources across airports and seasons.     

Analysis of common-cause and special-cause variation in the deterioration of transportation infrastructure: A field application of statistical process control for structural health monitoring

We present a statistical process control framework to support structural health monitoring of transportation infrastructure. We contribute an integrated, generally-applicable (to various types of structural response data) statistical approach that links the literatures on statistical performance modeling and on structural health monitoring. The framework consists of two parts: The first, estimation of statistical models to explain, predict, and control for common-cause variation in the data, i.e., changes, including serial dependence, that can be attributed to usual operating conditions. The ensuing standardized innovation series are analyzed in the second part of the framework, which consists of using Shewhart and Memory Control Charts to detect special-cause or unusual events.We apply the framework to analyze strain and displacement data from the monitoring system on the Hurley Bridge (Wisconsin Structure B-26-7). Data were collected from April 1, 2010 to June 29, 2011. Our analysis reveals that, after controlling for seasonal effects, linear trends are significant components of the response measurements. Persistent displacement may be an indication of deterioration of the bridge supports. Trends in the strain data may indicate changes in the material properties, i.e., fatigue, sensor calibration, or traffic loading. The results also show that autocorrelation and conditional heteroscedasticity are significant sources of common-cause variation. Use of the control charts detected 43 possible special-cause events, with approximately 50% displaying persisting effects, and 25% lasting longer than one week. Analysis of traffic data shows that unusually heavy loading is a possible cause of the longest special-cause event, which lasted 11 days.     

The inherent precision of regression estimated route probabilities

From basic assumptions about independent and consistent driver behaviour, and with data from traffic counts, we derive statistical properties of regression or correlation estimates of route selection probabilities, turning probabilities and travelling times. Our modelling is conditional in a way that avoids most traffic generation problems and permits an asymptotic analysis of the precision under mild assumptions allowing non-stationarity. This allows us to put together non-stationary data from the corresponding time intervals during several days when we aim at high precision estimates.     

基于单参数的Lagrangian支持向量回归算法及其应用

支持向量机(Support Vector Machines,SVM)是基于统计学习理论框架下的一种处理非线性分类和非线性回归的有效方法。由于具有完备的理论基础和出色的学习性能,该方法已成为当前国际机器学习界的研究热点,能较好地解决小样本、高维数、非线性和局部极小点等实际问题。这里提出了一种基于单参数的拉格朗日(Lagrangian)支持向量回归算法,并将该算法应用在外贸货物吞吐量预测中。估算结果证明了这种改进的支持向量回归算法在吞吐量预测中的有效性和实用性。     

Estimating salinity to complement observed temperature: 1. Gulf of Mexico

This paper and its companion [Thacker, W.C., Sindlinger, L., 2007-this issue. Estimating salinity to complement observed temperature: 2. Northwestern Atlantic. Journal of Marine Systems. doi:10.1016/j.jmarsys.2005.06.007.] document initial efforts in a project with the goal of developing capability for estimating salinity on a region-by-region basis for the world oceans. The primary motivation for this project is to provide information for correcting salinity, and thus density, when assimilating expendable-bathythermograph (XBT) data into numerical simulations of oceanic circulation, while a secondary motivation is to provide information for calibrating salinity from autonomous profiling floats. Empirical relationships between salinity and temperature, which can be identified from archived conductivity–temperature–depth (CTD) data, provide the basis for the salinity estimates.The Gulf of Mexico was chosen as the first region to explore for several reasons: (1) It's geographical separation from the Caribbean Sea and the North Atlantic Ocean makes it a “small ocean” characterised by a deep central basin surrounded by a substantial continental shelf. (2) The archives contain a relatively large number of CTD data that can be used to establish empirical relationships. (3) The sharp fronts associated with the Loop Current and its rings, which separate water with different thermal and haline characteristics, pose a challenge for estimating salinity. In spite of the shelf and the fronts, the relationship between salinity and temperature was found to be sufficiently regular that a single empirical model could be used to estimate salinity on each pressure surface for the entire Gulf for all seasons. In and below the thermocline, root-mean-square estimation errors are small — less than 0.02 psu for pressures greater than 400 dbar, corresponding to potential density errors of less than 0.015 kg/m³. Errors for estimates nearer to the surface can be an order of magnitude larger.     

基于支持向量回归机的船舶操纵运动在线辨识建模

采用滚动时间窗的方法实现支持向量机的在线辨识。以船舶操纵运动响应模型为研究对象,并由10°/10°和15°/15°仿真Z形试验数据构造支持向量机参数辨识所需的训练样本对,应用支持向量机进行船舶操纵运动在线辨识建模,回归操纵运动响应模型中的操纵性指数,并利用建立的响应模型进行Z形试验的数值模拟。将Z形试验数值模拟结果同仿真Z形试验数据进行比较,结果表明,在线式支持向量回归机是一种进行船舶操纵运动在线辨识建模的有效方法。     

SA取消前后RBN—DGPS系统精度评估与分析

在取消SA前后,对GPS、RBN－DGPS的陆地静态定点定位进行实测、统计与分析,重点对RBN－DGPS基准站的伪距改正量残差、经纬度残差等数据进行了大量的实测与传统,并在此基础上,统计了SV、HDOP、VDOP、PDOP、2DRMS的概率。采用回归分析的数学方法,建立了RBN－DGPS用户定位精度随基准站距离变化的数学模型。对我国GPS、RBN－DGPS系统的应用与开发,具有一定的参考价值。     

Improvement in the estimation and back-extrapolation of CO2 emissions from the Irish road transport sector using a bottom-up data modelling approach

Road transport is a major source of CO₂ emissions in Ireland and accounts for almost 96% of the total CO₂ emissions from the transport sector. Following the recent adopted UNFCCC reporting guidelines on annual inventories [24/CP.19], this study applied the 2006 IPCC Guidelines for National Greenhouse Gas Inventories (2006 IPCC GLs) tier 3 approach to estimate CO₂ emissions from road transport at the vehicle category level, for the first time in Ireland. For this, disaggregated datasets were prepared based on year of vehicle registration and mileage since registration of the vehicle. Such an approach provided a more realistic national scenario in comparison to the use of average mileage degradation in emission calculations. This investigation comprised a recalculation of previous emissions estimates (1990–2012) and an estimation of CO₂ emissions in 2013 using a previously unavailable level of data disaggregation for vehicle mileage as well as using vehicle class specific data and an improved bottom-up estimation methodology in COPERT. Historic vehicle fleet data were restructured, annual mileage data were estimated in relation to the fleet data and back extrapolated using a regression approach.The results showed that the mileage degradation was not only subject to fuel technology, engine size, and age but also the emissions class and vehicle category. It was also observed that the disaggregated level of data provided a different CO₂ emissions split among the vehicle categories than that of previous estimations which were based on an aggregated level of data. Previous emissions inventories (1990–2012) were shown to have underestimated the share from diesel fuelled passenger cars by more than 56% in 2012. Diesel fuelled passenger cars were also found to account for the majority of CO₂ emissions from road transport activities in Ireland in 2013. The level and trend assessment showed that emissions from Euro-II and Euro-III classed vehicles especially for passenger cars, which have a significant contribution to the total emission in 2013 have caused an increase in fleet level emissions in Ireland. In addition, the results also showed that the emissions share from Light Duty Vehicles and Heavy Duty Vehicles were overestimated by previous investigations. This paper highlights the importance of the resolution of data used in emissions inventory preparation which may impact upon future projections and policy formulation. The findings of this investigation are also discussed in relation their implications for road transport policy, including carbon taxation and future policy options aimed at achieving EU emissions target in 2020.