Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
CHAPTER 3 INTERPRETATION, APPRAISAL AND APPLICATIONS The findings reported in the previous chapter will have a significant impact on the analysis of delay and level of service at traffic-actuated intersections. A software has been developed, as descnbed in Appendix G. for implementation of the analysis procedures developed under this project. In areas where state or local governments have adopted the HCM as a standard for the assessment of level of service on their roadway system, the new procedure could have a strong impact on decisions related to roadway construction programs and growth management decisions. RELATIONSE[IP TO OILIER TRAFlIC-ACTUATED INTERSECTION MODELS The analytical models developed under this project must find their place among the current set of traffic engineering analysis tools for signalized intersections. Discussion of these tools should prove to be helpful at this point. Two other tools that perform functions similar to the proposed analytical model, both of which are simulation based, were examined in this study. The first is NETS=, a microscopic simulation program for general traffic control networks. The second is EV1PAS [10], a special purpose simulator dedicated to traff~c-actuated intersections. EV]PAS was investigated because it is the only model that purports to optimize traffic-actuated control settings such as unit extension and maximum green times. These models will be discussed separately. Relationshin to NETS~i NETSIM results were used in this project as a surrogate for field data in the evaluation of the proposed analytical model. NETSIM also provides an alternative to the proposed analytical model for estimating delay and level of service at intersections. For relatively straightforward situations involving signalized intersections, the analytical model will be the preferred choice of many agencies because it is simpler to apply, it offers intermediate computational outputs, and it performs direct estimates of capacity for each movement. For these reasons, it lends itself to adoption as an official standard much more readily than a simulation program. If a combination of analytical and simulation techniques is to be used, it is important to recognize some ofthe conceptual differences between NETSIM and the HCM type of models that will make direct comparison difficult. Several such differences have been identified. Delay Definitions NET SIM defines stopped delay in terms of the accumulated number of seconds at which a vehicle is actually travelling at zero speed. The HCM, on the other hand, computes stopped delay analytically NCHRP Project 3-48 Final Report: Page 29
by dividing the total delay (i.e., the area contained within the queue accumulation polygon plus the incremental delay correction) by a factor of I.3. This factor was developed previously as the average ratio of total delay to stopped delay. Direct comparisons of the analytical delay estimates against NETSIM would therefore be more of a test of the validity of the " ~ .3 " factor, than of the analytical mode! itself NETSIM measures stopped delay explicitly by examining microscopically the time-space trajectory of each vehicle passing through the system. The stopped delay is reported as the accumulated number of seconds at which a specific vehicle had no movement. Thus, any comparison between the stopped delays reported by these two methods must have considerable faith in the validity of the HCM def~ni- tion of stopped delay as a subset of total delay under all conditions. NETSIM also reports its own version oftotal delay. However, the NET SIM delay is referenced to the free-flow speed of the approaching vehicles instead of the cruise speed used by the HCM. So, the NETSIM total delay is, by definition, higher than the HCM total delay because it includes the "cruise delay," i.e., the delay resulting from cruise speeds that are lower than the free flow speed along the link. Among other things, this means that NETSIM's total delay depends on the length of the approaching link, whereas the HCM total delay does not. NETSIM also reports a value for "queue delay," which includes aD of the stopped delay plus a more or less arbitrary portion of the time spent in acceleration and deceleration. Again, it is not possible to relate this value by definition to any delay value computed by the HCM. The HCM total delay represents the delay actually caused by the signal. If there were no signal, the HCM total delay would be zero. The problem is that NETSIM does not report any term that isolates the signal delay. Therefore, it is not possible to compare delays estimated by HCM-type analytical models directly with any delay value reported by NETSIM. Distribution of Gaps in Arriving Traffic A traff~c-actuated controller terminates a given phase whenever a gap appears at the detector that exceeds the allowable gap specified for that phase. Therefore, the length of a phase will depend heavily on the probability of a gap of a certain length occurring in the traffic flow after the queue has been serviced. If the analytical and simulation models do not use exactly the same vehicular arrival distribution, then systematic differences will be built into any comparison that wait cause the results to diverge. The arrival distribution for the proposed analytical mode} is the bunched negative exponential mode! which is described in Appendix C to this report. An equation was described for predicting the expected time before a gap of a specific length occurs to terminate the phase. The results are a filnc- tion of the assumed distribution of gaps in the arriving vehicles. NCHRP Project 3-48 Final Report: Page 30
It is not possible in the current version of NET SIM to specif,r the characteristics of the arrival distribution. Vehicles are released into each external link from a hypothetical entry node with constant headways that fonn a uniform distribution. NETSIM's queue propagation algorithm causes some bunching of vehicles to occur as they proceed through the link. The actual shape of the arrival distribution at any point downstream depends on several factors, such as vanability of speeds among different vehicle types and the length ofthe link. It is very difficult to predict analytically, and even more difficult to produce a simulated gap distribution that matches the distribution used in the analytical model. The actual dispanty in the signal timing and delay estimated by NETSIM and by the analytical mode! In this study was relatively small. However, it is suggested that a significant por- tion of this dispanty was due to differences in the gap distributions used by the two models. Steacly-State Headway Resolution Another potential source of difference lies in the resolution of the steady-state headway values accepted as input data by NET SIM. This resolution currently stands at 0. ~ seconds per vehicle. For typical headway values In the range of 2 seconds per vehicle, this translates into a saturation flow rate choice of 1715, IS00, or ~ 895 vehicles per lane per hour of green time. The HCM Chapter 9 proce- dure deals in a continuum of choices in this range. This problem was solved in the studies described in this report by ensuring that the saturation flow rate was always specified at the IS00 level. This treatment wall not generally be quite as simple in practical applications. L`ink-Based Delay Assignment The analytical models include all of the delay due to the signal in a single computation. NETS]M, on the other hand, is a network mode! that uses a specified~stnucture of links and nodes. Vehicles passing through the signal that have not reached their filet speed wall therefore have some of their delay assigned to the link immediately downstream of the signal. In addition, some of the delay from the Intersection immediately upstream of the signal could be associated incorrectly with the subject intersection. This problem will be most significant under congested operation. Relationship to EV1PAS EV]PAS is an optimization/simulation mode! for actuated controlled, isolated intersections which is described in more detail in Appendix A. It is capable of analyzing and determining the optimal settings of timing parameters for a wide range of geometric configurations, detector layouts, and almost any phasing pattern available in a single or dual-ring NEMA and Type ~ 70 controllers. It will generate the optimized settings for controllers ranging from pretimed to volume-density actuated controllers. The optimum settings oft~m~ng parameters include minimum green time, maximum green time, unit extension time, minimum gap, time before reduction, time to reduce, added initial, and maximum initial for each phase. NCHRP Project 3-48 Final Report: Page 31
As such, EV1PAS offers a self-conta~ned alternative to the design and evaluation of a timing plan for a traffic-actuated intersection. Compared to NET Sew, it offers the advantage of a timing plan design, in addition to the evaluation. Studies reported in Appendix A indicate a high level of agreement in the timing plans produced by NET SIM and EV]PAS. Studies reported in Reference [l I] indicate that EVIPAS produced timing parameter settings that were generally superior to various analytical methods found in the literature. While it appears to offer some advantages, it suffers the same short- com~ngs as NETSIM with respect to the characteristics that make it suitable for adoption as a stan- dard for level of service analysis by public agencies. EVIPAS is a relatively new program that is not familiar to many users. It is in its final stages of development and testing and is being incorporated as a component in traffic mode} software inte- grators. These developments should promote improved acceptance and implementation in the future. LIMITATIONS OF TI1E TRAFFIC-ACTUATED TIMING ESTIMATION PROCEDURE The traffic actuated timing estimation procedure described in this report provides a reasonable approximation of the operation of a traffic-actuated controller for nearly all of the conditions encountered in practice. The results obtained from this method have correlated well with extensive simulation data and with limited field studies. However, the procedure involves a deterministic analytical representation of an extremely complex stochastic process, and therefore has some limitations which must be recognized. Some of the limitations result from unique situations that cannot be modeled analytically in a satisfactory manner, even by the filly methodology presented in Chapter 9 of the HCM. One example is the case of compound left turn protection with opposing shared lanes for leg turns and through movements. The HCM Chapter 9 methodology deals with this as a separate case (Case 6 in Table 9-12), and applies an empirical treatment to determine the saturation flow adjustment factor for leR turns. Simulation provides the only effective way to estimate the timing plan parameters for this case. The sample problem presented in Appendix C demonstrated the sensitivity of the procedure to the unit extension times set in the controller. As expected, longer unit extension times produced longer average Been times except when constrained by the maximum green time settings. Shorter extension times had the opposite effect. There is, however, a lower limit to the range of unit extension times that can be modeled realistically. It is well known in practice that when the unit extension times are too short, premature terminations of a phase may result due to anomalies in the departure headways created primarily by lapses in driver attention. The traffic-actuated control mode! described in this report assumes a constant departure headway, and does not, therefore, reflect this phenomenon. S~rnulation models introduce a stochastic element into the departure headways based on a theoretical distribution. They are therefore able to invoke premature phase terminations to some extent, but they do not deal with anomalous driver behavior. NCHRP Project 3-48 Final Report: Page 32
As a practical matter, unit extensions should reflect headways at least 50 percent longer than the expected departure headways. For example, assuming a 2 second average departure headway, unit extensions should accommodate a 3 second departure headway without terminating the phase. Assuming a detector occupancy time of 1 second. Two seconds of the headway will appear as a gap. So the minimum practical value for the unit extension would be 2 seconds. Smaller values may be appropriate in multiple lane cases in which average departure headways are shorter. The analysis of permitted left turns from shared lanes always poses special problems. The sern~- empirical treatment prescribed for shared lane permitted left turns in Chapter 9 of the HCM does not lend itselfto the iterative timing estimation procedure described In this report. An analytical approx~- mation of the shared lane model was therefore substituted to ensure stable convergence of the solution. This treatment produced timing plans that agreed well with simulation results, however, the analysis of delay resulting from the timing plan did not always agree with the results of the HCM Chapter 9 method. It appears that an iterative method of achieving equilibrium between the shared lane and the adjacent through lanes in the HCM Chapter 9 methodology is a prerequisite to the development of a satisfactory timing estimation procedure. This problem will require further research. When traffic volumes are extremely low the timing plan becomes somewhat of an abstraction unless the recall Unction is used for each phase. In the absence of any demand the green indication rests on the phase that received the last demand, and may do so for several minutes. This implies that very long red times will be displayed on some phases, however no delay will be associated with these red times, because no vehicles will be affected. The procedure described in this report will compute very short equivalent red times for these phases in an attempt to provide a signal timing plan that will produce realistic delays. While the procedure produces a reasonable approximation, design decisions based on comparison of the minimal delays encountered in the LOS A range should be avoided. NCHRP Project 3-48 Final Report: Page 33
