The capacity evaluation of the merge sections in urban freeways considering deterministic and stochastic approach: Example of İstanbul

Traffic jam is generally caused because of unmatched demand over the road segment. The case of higher demand over the capacity, is observed especially at the bottlenecks of urban freeways and morning/afternoon peaks when the demand is relatively high. This study involves observations on capacity and traffic jams at the on-ramp showing typical characteristics of bottlenecks of urban freeway segments. Investigated section is the O2 Freeway which provides the merge from the Ümraniye district. The O2 Freeway is one of the components of E-80 Freeway in the European Transit Network and located in the Asian side of Istanbul. In the merge section, the deterministic capacity values for the freeway merge sections of Highway Capacity Manual (HCM) 2010 and calculated the capacity values by stochastic approach are compared. The dataset which is necessary for this study and represents the vehicle traffic, is obtained from the Metropolitan Municipality of Istanbul Traffic Control Center (MMI-TCC). This dataset reports the date, time, avarage flow speed, total vehicle volumes and the volumes of heavy vehicles per lane, which recorded by the RTMS (Remote Traffic Microwave Sensor) stated on all freeways by the MMI-TCC. In the scope of this study, one-year sampled dataset of 2012 is used that was provided from 3 different RTMS; number 324, 325 and 326, which are reported the data from the O2 Freeway - Ümraniye Merge Section. 4 upstream lanes of O2 Freeway are obtained from the RTMS no 324, 2 lanes of ramp roadway from Umraniye to O2 Freeway are obtained from RTMS no 325, 4 downstream lanes O2 Freeway are obtained from RTMS no 326. The dataset which is provided from RTMSs consists 2-minute time intervals. Initially during the process of this datasets arrangement, the missing measurements are recognized from all 3 RTMS. For this purpose, daily time streaming of the dataset were controlled and determined missing measurements including different time intervals. The blank rows by 2-minute time intervals were added for all missing time intervals to the dataset. In addition, during the process of controlling of the time streaming, some irregular measurements are recognized and deleted. Instead of these measurements, again the blank rows by 2-minute time intervals were added to the dataset. Another defective measurement was determined for RTMS no 324 and 325. On 3 months (october, november, december) it has been determined that those 2 RTMS datasets were not recording 2 minute intervals in every 2 hour period. It was decided not to use this 3 months-data because they couldn't be used in aggregating to 5-minute time intervals. The last defective measurement were determined for RTMS no 326. After July 24th, the data of the 4th lane in this section weren't recorded by the RTMS. Due to the weather conditions and some other effects some more defective missing and defective measurements were recognized during data arrangement process. At the end of these it was decided not to use all these defective data and only use the reported data of the first 7 months of the year. Some days, including these kind of data and 4 national holidays in these 7 months, remained out of study. As the result of arrangement of datasets, blank rows have been produced approximitely %2 of the datas recovered from all three RTMSs belonging to first 7 months of the year. In case of blank rows coinciding with the breakdown interval and and current value being less than %50 of the daily required number, data of whole day have been excluded from the study. As result of these exclusions, respectively 187, 180 and 177 daily data are used in this study from RTMS no 324, 325 and 326. At the next stage, the data which is provided from RTMSs and represented by 2-minute time intervals, were aggregated to proposed 5-minute time intervals for the calculation of stochastic capacity. During this process, 2-minute time intervals contaning missing data rows are not included to 5-minute data aggregation. After that, volume heavy vehicles represented per lane, transformed into passenger car volume. At the next stage, these volumes were transformed to the equivalent hourly flow rates per lane. Meanwhile, avarage speeds per lane are used to calculate volume-weighted avarage flow speed of the road section. With the help of these organized speed and flow rates, time-speed and time-flow rate diagrams are produced for every day. These scattered time-speed and time-volume diagrams were used to determine the breakdown phenomena. The "optimum speeds" were observed as 50, 50 and 70 km/h respectively for RTMS no 324, 325 and 326 by with the help of flow rate-speed diagrams. As a result, breakdown flow rates and time intervals are determined. Afterwards, the probabilities of breakdown flow rates (capacity values) were calculated by using the direct breakdown probability estimation method and the estimation of probability distribution function method. Additionally, the estimation of breakdown probability distribution function is calculated with two different methods as parametric and non-parametric. In the non-parametric method, Kaplan-Meier (product limit) method was applied by using of survival function, while in the parametric method, Weibull probability estimation function was applied with the estimating of scale and shape parameters by using of "maximum likelihood" method. At the end of the study, the capacity values calculated by the stochastic methods are compared with each other and with the deterministic capacity values determined by HCM 2010. The flow rates have the capacity range between 15%-85% probability values calculated by Weibull probability distribution were calculated as 1577-1732 pc/h/ln for freeway upstream, 1858-2086 pc/h/ln for ramp road, 2081-2270 pc/h/ln lane for freeway downstream. The maximum flow rates observed in these check points are 1828, 2157 and 2366 pc/h/ln accordingly. The values determined by HCM 2010 for the same points are as 2300, 2200 and 2300 pc/h/ln accordingly. As conclusion, when the results of direct probability estimation and probability distribution function estimation (stochastic approach) methods of determined breakdown flow rates shows that probability distribution function estimation gives more detailed results. To be able to produce probabilities for only determined flow rate groups and not being able to complete probability distribution function of all 3 RTMSs (being highest probability value of 0.9 at RTMS no 325) have been identified as weak points of the direct probability estimation method. Two different methods (parametric and non parametric) were applied for probability distribution function estimation. For nonparametric method; Kaplan-Meier method using "survival function", for parametric method; "Weibull probability estimation function" using "maximum likehood" method by calculating scale and shape parameters were applied. With Kaplan-Meier method, as being more detailed than the direct probability estimation method, a breakdown probability was calculated for all the flow rates. But it has been concluded that this method also was inferior to Weibull probability distribution function as it couldn't produce a flow rate for every probability value and giving the same probability value for two consecutive flow rates if second one has no observation. Lastly, the capacity values, even having the highest (%100) probability that calculated with Weibull probability distribution method, seems to be at a low capacity rate for observed freeway except the downstream flow and especially upstream flow. In this case, it is conceivable that behaviors which we can see frequently in the drivers, like lane breaches, frequent lane changing, behaving impatient, can cause capacity loss by causing distortions and fluctuations to the flow.