Cover Image

Not for Sale



View/Hide Left Panel
Click for next page ( 20


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 19
19 radio, in which case it can also capture operator sign-in and Communication is also needed from the central computer operator-initiated events. Several AVL-APC systems feature to the on-board computer, whether for occasional software this arrangement, which helps facilitate matching and yields a upgrades or daily schedule updates. In general, the communi- richer database. On-board integration with the radio control cation method used for upload is used for download as well. head is recommended even if there is no real-time AVL. Off-line, radio-based event records also can conceivably be 2.3.3 Exception-Only Data Recording integrated with records uploaded from the on-board com- puter. However, the researchers are not aware of this design Some AVL systems send timepoint messages only if a bus is being used. off schedule, partly to limit radio traffic, and partly because controllers are usually interested only in service that is off schedule. For example, Tri-Met's AVL-APC system transmits Interstop Records and Detailed Tracking only exception messages by radio, while saving a full set of With on-board data collection, data can also be collected records on board for later analysis. between stops. One data collection mode is to make records However, if exception data is all that is available for off-line very frequently (e.g., every 2 s); this mode uses buses as GPS analysis, analysis possibilities become severely limited. For probes, which can enable such special investigations as study- example, if only off-schedule buses create timepoint records, ing the bus's path through a new shopping center or studying running times can only be measured for off-schedule buses. bus movements in a terminal area. Frequent interstop records Researchers at Morgan State University tested the feasibility also offer information on speed and acceleration. of data analysis using exception data from bus routes in Bal- An alternative data collection mode is to write event records timore, MD (13). Because they had records only on buses that for defined events that can occur between stops, such as cross- were outside an on-time window, they focused on next-segment ing a speed threshold. To measure delay, Eindhoven's system running time for buses that arrived at a timepoint early or records whenever a bus's speed rises above 5 km/h or falls late. If the bus reached the next timepoint "on time," the below 4 km/h. (Using differing thresholds prevents oscillation researchers had to guess when within the on-time window when the bus is traveling at the threshold speed.) Records for a the bus arrived. Also, because the system archived only variety of speed thresholds could be useful for analyzing speed exception messages, it was impossible to know whether a profiles. Tri-Met's system tries to capture maximum speed missing record meant that a bus was on time or that the between stops, both as a measure of traffic flow and as a safety radio system had failed. In an interview, the Morgan State indicator. It tracks speed continuously, storing the greatest researchers stated that while the Mass Transit Administration speed since the last stop in a temporary register; then at each (MTA) had asked them to systematize the way they trans- stop, maximum speed since the previous stop is recorded. formed the raw data into records that would support analy- ses of running time and schedule adherence, they felt it was impossible given the frequent need for assumptions to make Data Uploading up for missing data. When data is recorded on vehicle, there has to be a system Fortunately, advances in radio technology have reduced the for uploading the data from the on-board computer to the pressure to limit data collection to exceptions. As an example, central computer. Newer systems usually include an auto- CTA's real-time AVL system, specified in 1993 to provide loca- matic high-speed communication device through which data tion data only if buses were off schedule, was modified in 2002 is uploaded daily when buses are fueled. Older systems such so that buses transmit location data regardless of whether they as Tri-Met's rely on manual intervention, such as exchanging are off schedule. data cards or attaching an upload device, which adds a logis- tical complication. 2.4 Data Recovery and Sample Size The absence of an effective upload mechanism can render an otherwise promising data collection system useless for off-line Automatic data collection systems do not offer 100% data data analysis. One transit agency has a new stop announcement recovery. Traditional APCs have the worst record; a 1998 sur- system that records departures from every stop. However, the vey found net recovery rates for APCs ranged from 25% to data is overwritten every day, because the data logging feature 75%, with newer systems having better recovery rates (14). In was intended for system debugging, not data archiving. To 1993, the Central Ohio Transit Authority (COTA) reported that, make it an archived data source, the agency would have to with 11.9% of the fleet APC equipped, it netted on average five either exchange cassettes nightly, something it deemed imprac- usable samples per assignment each quarter (15). That rep- tical, or invest in a high-speed data transfer link, something it resents a 6.1% sampling rate, for a net recovery rate of about found too expensive. 50%. Only a small part of that loss was due to mechanical

OCR for page 19
20 failures, as mechanical reliability was reported to be well scheduled weekday trips, and at least three valid observations above 90%. for 83% of its scheduled weekday trips. On the weekend, Less is known about data recovery rates for radio-based AVL 85% of scheduled trips had at least one valid observation. In systems. King County Metro recovers AVL data from about 2002, King County Metro reported that its recovery rate had 80% of its scheduled trips. However, with the entire fleet improved to the 60% to 70% range. instrumented, data recovery rates are not so important with In general, a small sample size is sufficient to reliably esti- AVL unless there is systematic data loss in particular regions. mate the mean of a quantity with low variation such as run- Inability to match data in space and time is the single ning time on a segment or demand on a particular scheduled most cited reason for rejecting data. There are other reasons, trip. In contrast, a large sample size is needed to examine vari- including malfunctioning on-board equipment, data being ability or extreme values, such as the 90-percentile running out of range, radio failure, and so forth. Imbalance in pas- time or load. A very large sample size is needed to accurately senger counts, another common problem that forces data to estimate proportions, such as proportion of departures that be rejected, is covered in Chapter 8. are on time (16). The effective sampling rate of an AVL-APC system is the fleet Analyses that aggregate scheduled trips into periods, or penetration rate multiplied by the data recovery rate. If 10% of that aggregate routes, have the advantage of a larger sample the fleet is equipped, and data is recovered from 70% of the size than analyses of individual scheduled trips. For this instrumented vehicles, scheduled trips will be observed, on reason, many customary analysis methods, developed in a average, 10% * 70% = 7% of the number of times they are oper- limited-data environment, find results for the period rather ated. In a 3-month period containing 65 weekdays, 13 Satur- than the trip, and for the system rather than the route. days, and 13 Sundays, average observations per scheduled A large sampling rate allows more timely analysis of data. trip will be 4.5 for weekday and just under 1 for Saturday and Over a long enough period of time, even a small sampling Sunday trips. rate will yield a large number of observations. However, for An average sampling rate can mask significant variations management to react promptly to demand and performance across the system. When fleet penetration is small, logisti- changes, or measure the impact of operation changes, analysts cal difficulties coupled with the vagaries of data recovery need recent data. Therefore, tools used in the active manage- failure often result in some scheduled trips being sampled ment of a dynamic system will benefit from the high sampling well more than the average number of times, and others rate that follows when the entire fleet is equipped. perhaps going completely unobserved. Management of the Because leader-follower or headway analysis requires valid rotation of the instrumented vehicles, then, becomes another observations of consecutive pairs of trips, the number of valid important factor in determining whether needed data will headways one can expect to recover is proportional to the be available. square of the data recovery rate and the correct assignment In an example drawn from the project's case studies, an audit rate. For example, if the data recovery rate is 70% and if a of King County Metro's Fall 1998 sign-up (a 4-month period) request to instrument a given route results in 90% of the trips found on average six valid APC observations per weekday on that route having an instrumented bus, one can expect to scheduled trip, for a sampling rate of 7.5%. With about 15% of observe only (0.7)2 * (0.9)2 = 40% of the day's headways on that the fleet instrumented, that represents a data recovery rate of route. If there is a realistic possibility of trips overtaking each 50%. Coverage across the schedule was variable. King County other, having anything less than data from all the operated trips Metro recovered at least one valid observation on 97% of their casts some doubt on calculated headways.