Expressways play a vital role in serving mega-cities, and the safety of expressways

is extremely important. In order to explore the crash mechanisms of expressways,

previous studies have mainly utilized average daily traffic (ADT) as a major contributing

factor. In recent years, several researchers also adopted average hourly traffic (AHT)

and microscopic traffic at five-minute intervals in expressway safety analyses.

Nevertheless, there have been no studies, which have compared the performance of all

three factors: ADT, AHT, and microscopic traffic.

This study collected data from three expressways in Central Florida, including traffic

data at one-minute intervals, detailed crash information, and geometric characteristics. A

Bayesian Poisson-lognormal model was estimated for total crash frequency using ADT,

a Bayesian multilevel Poisson-lognormal model was built for hourly crash frequency

prediction using AHT, and a Bayesian multilevel logistic regression model was

developed for real-time safety analysis using microscopic traffic indicators at five-minute

intervals. The modeling results showed that the crash-contributing factors found by

different models were comparable but not the same. Four variables, i.e., the logarithm of

volume, segment length, number of lanes, and existence of weaving segments, were

found to be positively significant in the three models, and four other variables were only

significant in one or two models. The ADT-based, AHT-based, and five-minute-based

models were used to predict safety conditions at different levels: total, hourly, and five-minute

intervals. The results indicated that the AHT-based crash-estimation model

performed the best in predicting total and hourly crash frequency, and that the real-time

crash prediction model was the best in identifying crash events for dangerous segments

at five-minute intervals. The AHT was recommended for future long-term traffic safety

analysis, and traffic at five-minute intervals was suggested for the implementation of

active traffic management (ATM).

Since the existence of weaving segments was found to be significantly related to

crash potential in all three crash analyses models, crash-contributing factors of weaving

segments were further studied using real-time safety analysis, which implements traffic

at five-minute intervals to predict crash potential. This study presents a logistic

regression model for crashes at expressway weaving segments using crash data,

geometric data, traffic data at one-minute intervals, and weather data. The results show

that the speed difference between the beginning and the end of the weaving segment

and the logarithm of volume have significant impacts on the crash risk of the following

five to ten minutes for weaving segments. The configuration is also an important factor.

The weaving segment, in which there is no need for on- or off-ramp traffic to change

lanes, presents a high crash risk because there are more traffic interactions and greater

speed differences between weaving and non-weaving traffic. Meanwhile, weaving

influence length, which measures the distance at which weaving turbulence no longer

has impact, is found to be positively related to the crash risk at the 5% confidence

interval. In addition to traffic and geometric factors, the wet pavement surface condition

significantly increases the crash risk since vehicles are more likely to be out of control

and need longer braking distances on wet pavement. Once the crash mechanism of

weaving segments was found, the safety condition of weaving segments could be

estimated using traffic, geometry, and weather factors at five-minute intervals.

This study focused on the safety of a congested weaving segment, which has a high

crash potential. Various ATM strategies were tested in microscopic simulation (VISSIM)

through the Component Object Model (COM) interface. The strategies included ramp

metering (RM) strategies, variable speed limit (VSL) strategies, and an integrated RM

and VSL (RM-VSL) strategy. Overall, the results showed that the ATM strategies were

able to improve the safety of the studied weaving segment. The modified ALINEA RM

algorithms, which took both lane occupancy and safety into consideration, outperformed the traditional ALINEA algorithm from a safety point of view but at the expense of

average travel time. The 45 mph VSLs, which were located at the upstream of the

studied weaving segment, significantly enhanced the safety without notably increasing

the average travel time. In order to reduce the average travel time of the modified

ALINEA RM and maintain its impact on safety, the modified ALINEA RM was adjusted to

control queue length and was integrated with the 45 mph VSL strategy. The simulation

results have proved that the consolidated RM-VSL approach yields slightly lower total

crash risks, but provides much lower average travel times than the modified ALINEA.

Overall, the existence of a weaving segment would significantly increase crash

potential, and the traffic at five-minute intervals was suitable for the implementation of

ATM. Based on these two findings, a congested weaving segment was chosen to test

the impact of ATM on safety in real time through microscopic simulation. The result

showed that ATM was able to significantly improve the safety of the studied weaving

segment.