Investigation of hit-and-run crash occurrence and severity using real-time loop detector data and hierarchical Bayesian binary logit model with random effects

Objective: Most of the extensive research dedicated to identifying the influential factors of hit-and-run (HR) crashes has utilized typical maximum likelihood estimation binary logit models, and none have employed real-time traffic data. To fill this gap, this study focused on investigating factors contributing to HR crashes, as well as the severity levels of HR.

Methods: This study analyzed 4-year crash and real-time loop detector data by employing hierarchical Bayesian models with random effects within a sequential logit structure. In addition to evaluation of the impact of random effects on model fitness and complexity, the prediction capability of the models was examined. Stepwise incremental sensitivity and specificity were calculated and receiver operating characteristic (ROC) curves were utilized to graphically illustrate the predictive performance of the model.

Results: Among the real-time flow variables, the average occupancy and speed from the upstream detector were observed to be positively correlated with HR crash possibility. The average upstream speed and speed difference between upstream and downstream speeds were correlated with the occurrence of severe HR crashes. In addition to real-time factors, other variables found influential for HR and severe HR crashes were length of segment, adverse weather conditions, dark lighting conditions with malfunctioning street lights, driving under the influence of alcohol, width of inner shoulder, and nighttime.

Conclusions: This study suggests the potential traffic conditions of HR and severe HR occurrence, which refer to relatively congested upstream traffic conditions with high upstream speed and significant speed deviations on long segments. The above findings suggest that traffic enforcement should be directed toward mitigating risky driving under the aforementioned traffic conditions. Moreover, enforcement agencies may employ alcohol checkpoints to counter driving under the influence (DUI) at night. With regard to engineering improvements, wider inner shoulders may be constructed to potentially reduce HR cases and street lights should be installed and maintained in working condition to make roads less prone to such crashes.     

基于贝叶斯最大熵的黑土区小流域土壤有机质空间分布预测

进行w（SOM）空间预测研究,对掌握区域w（SOM）空间分布现状、实施精准农业以及保护区域生态环境都有着重要意义.以土地利用类型为辅助变量,将不同土地利用类型所对应w（SOM）的概率分布作为“软数据”,采用BME（贝叶斯最大熵）方法对我国东北黑土丘陵区海沟河小流域表层w（SOM）的分布情况进行空间预测,并与以w（TN）和土地利用类型为辅助变量的CK（协同克里格）方法进行比较,探讨两种方法的可行性与精度.结果表明:我国东北黑土丘陵区海沟河小流域表层w（SOM）平均值为24.04 g/kg,空间变异程度为中等.w（SOM）与w（TN）呈极显著正相关,与土地利用类型存在较强的相关性,不同土地利用类型w（SOM）差异明显,w（TN）与土地利用类型能够用来辅助w（SOM）的空间分布插值.相较于CK方法,BME方法能更好地利用“软数据”进行空间插值,使对w（SOM）预测结果的平均误差（ME）、平均绝对误差（MAE）和均方根误差（RMSE）均有所降低,精度大幅提高,空间插值效果明显优于CK方法.研究显示,研究区w（SOM）以阶梯状自东向西依次递减分布,在南北方向上变化不大,空间变化特征较为明显,BME方法利用“软数据”插值后的结果能较好地反映研究区w（SOM）空间分布的实际情况.      

CHANNEL NETWORK EXTENSION BY LOGGING ROADS IN TWO BASINS,WESTERN CASCADES,OREGON1

ABSTRACT: Based on field surveys and analysis of road networks using a geographic information system (GIS), we assessed the hydrologic integration of an extensive logging-road network with the stream network in two adjacent 62 and 119 km² basins in the western Cascades of Oregon. Detailed surveys of road drainage for 20 percent of the 350 km road network revealed two hydrologic flow paths that link roads to stream channels: roadside ditches draining to streams (35 percent of the 436 culverts examined), and roadside ditches draining to culverts with gullies incised below their outlets (23 percent of culverts). Gully incision is significantly more likely below culverts on steep (< 40 percent) slopes with longer than average contributing ditch length. Fifty-seven percent of the surveyed road length is connected to the stream network by these surface flowpaths, increasing drainage density by 21 to 50 percent, depending on which road segments are assumed to be connected to streams. We propose a conceptual model to describe the hydrologic function of roads based on two effects: (1) a volumetric effect, increasing the volume of water available for quickflow and (2) a timing effect, altering flow-routing efficiency through extensions to the drainage network. This study examines the second of these two effects. Future work must quantify discharge along road segments connected to the stream network in order to more fully explain road impacts on basin hydrology.     

水质综合评价的人工神经网络模型

为探讨水质综合评价的客观方法,以成都市金堂县东风水库水质资料为例,建立了地面水水质综合评价的BP网络和Hopfield网络模型。BP网络模型以单输出代替多输出可保证评价结果的唯一性。Hopfield网络更优于BP网络,既适用于定量指标的水质参数又适用于定性指标的水质参数,而且使水质评价形象化     

水质评价模式识别的BP神经网络方法

水质受多个因素的影响，而且这些作用多是非线性的。其水质评价的实质是实现一个非线性映射。由于具有任意个隐层节点的前馈神经网络可以以任意精度逼近一个连续函数，因此，目前得到普遍应用的是采用BP算法的多层前馈神经网络。文中运用人工神经网络的反向传播（BP）算法进行了实例评价，与其它方法相比，评价精度较高，方法简单易行。