Cover Image

Not for Sale

View/Hide Left Panel
Click for next page ( 26

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 25
25 CHAPTER 4 Uses of AVL-APC Data This chapter includes traditional analysis methods as well they open the door for new analysis methods that can be as new tools developed by agencies using archived AVL-APC used to improve monitoring, planning, performance, and data. In addition, part of this project's survey/interview management. process included asking for suggestions of how AVL-APC Transit agencies--such as Tri-Met, King County Metro, OC data might be used, and the researchers also developed some Transpo, and HTM--that have good, automatically collected concepts. operational data are finding more and more uses for it. At first, To a large extent, performance analysis can be driven by a agencies may look to an automatic data collection system only performance-monitoring protocol that specifies measures or to provide the data needed for traditional analyses. But, once indicators that are to be reported. Transit agencies have exper- they have the larger and richer data stream that AVL and APCs imented with a large number of performance measures. AVL offer, they think of new ways to analyze it, and they want more. or APC data apply to many of the performance measures listed Eventually, their whole mode of operation changes as they in TCRP Report 88: A Guidebook for Developing a Transit become data driven. One APC vendor explains, "We're selling Performance-Measurement System (26). TCRP Report 100: an addiction to data." Transit Capacity and Quality of Service Manual (27) also sug- Five trends in data use have emerged from the paradigm gests several AVL- and APC-related measures for use as service shift from data poor to data rich: quality indicators. A list of current and potential uses of AVL-APC data is given Focus on extreme values in Table 4. They are discussed in detail beginning in Section Customer-oriented service standards and scheduling 4.3, following general discussions of data uses in an era of Planning for operational control automatic data collection (Section 4.1) and of data needs for Solutions to roadway congestion different analyses (Section 4.2). Section 4.3 includes several Discovery of hidden trends examples of analysis reports; many more can be found in the accompanying case studies. Both TCRP (14) and the Cana- 4.1.1 Focus on Extreme Values dian Urban Transit Association (17) have published synthe- ses of practice that include numerous samples of reports Traditional methods of scheduling and customer service generated from archived AVL and APC data. monitoring generally use mean values of measured quanti- ties, because mean values can be estimated using small sam- ples. However, many management and planning functions 4.1 Becoming Data Rich: are oriented around extreme values and are, therefore, better A Revolution in served by direct analysis of extreme values such as 90th- or Management Tools 95th-percentile values. These extreme values now can be esti- The transit industry is in the midst of a revolution from mated reliably because of the large sample sizes afforded by being data poor to data rich. Traditional analysis and deci- automatic data collection. Three examples follow: sion support tools required little data, not because data has little value, but because traditional management methods Recovery time is put into the schedule to limit the proba- had to accommodate a scarcity of data. Automatic data- bility that a bus finishes one trip so late that its next trip gathering systems do more than meet traditional data needs; starts late. Therefore, logically, scheduled half-cycle times

OCR for page 25
Table 4. Decision support tools and analyses and their data needs. Record Type Function Tool/Analysis Data Detail or Analysis Capacity Needed Sampling Rate Needed Needed Targeted General view Any Incident codes, control messages. 100% Investigations Trip recorder view (speed, acceleration) Interstop Maximum speed;records every 2 s or more 100% (complaints, disputes, to measure acceleration/deceleration rates. incidents) Analyzing and Route and long segment running time and Timepoint Endpoint data needed for route running time. 10% for mean running time; Scheduling scheduling Both arrival and departure times are helpful. 100% for analysis and Running Time Stop-level running time and scheduling Stop scheduling based on Running time net of holding time Stop Door open and close times, incident codes, extreme values control messages, on and off counts. Speed and traffic delay Stop, interstop 10% Dwell time analysis Stop Door open and close times, on and off 10% counts, farebox transactions, incident codes. Schedule Adherence Schedule deviation Timepoint Both arrival and departure times are helpful. and Long-Headway Waiting time (long headway) Stop Waiting 100% Connection protection Timepoint Arrival and departure times, control messages, farebox transactions. Headway Analysis Plotting successive trajectories Timepoint and Short-Headway Route-level headway analysis Timepoint Waiting 100% Waiting time (short headway) Stop Bunching (load/headway) analysis Stop Current on and off counts. Route Demand Demand along a route Stop On and off data. 10% for mean values; Analysis 100% for analysis and Demand over the day and Stop On and off data. scheduling based on headway/departure time determination extreme crowding Passenger crowding Stop On and off data. Near 100% Pass-ups, special uses, and fare categories Stop Incident codes, farebox transactions. 100%

OCR for page 25
Mapping Geocode stops, verify base map Stop 10% Map bus path through shopping centers, Interstop Interface for surveyor-initiated comments. 10% new subdivisions, etc. Miscellaneous Acceleration and ride smoothness Interstop Records every 2 s or more. 100% Operations Analysis Mechanical demand Stop, interstop 100% Terminal movement Interstop 100% Control messages Varies Incident codes and control messages. 100% Operator performance Varies 100% Higher Level Beforeafter study Varies As required by the type of analysis. varies Analysis Special event/weather analysis Trends analysis Aggregation and comparison over routes, Varies Multiroute analysis capacity including systemwide passenger-miles Transfer and linked-trip analysis Stop Farebox transactions with card IDs. 100% Shared route analysis, including headways Varies Data structures for shared routes. and load on a trunk Geographic demand and service quality Stop GIS with stop locations. 100% analysis Note: Items in italics are optional.

OCR for page 25
28 (scheduled running time plus recovery time) should be more than 5% of our customers should experience a bus based on an extreme value such as the 95th-percentile run- whose load exceeds x passengers." ning time. However, without enough data to estimate the As these examples show, a shift toward customer-oriented 95th-percentile running time, traditional practice sets it measures goes hand-in-hand with the ability to measure equal to a fixed percentage (e.g., 15% or 20%) of scheduled extreme values. running time. Yet, some route-period combinations need more than this standard, and others less, because they do 4.1.3 Planning for Operational Control not have the same running time variability. AVL data allows an agency to actually measure 95th-percentile running One of the questions posed by the explosion of informa- times and use that to set recovery times. A study for Tri-Met tion technology is how best to use information in real time to found that basing recovery times on 95th-percentile run- control operations, for example, by taking actions such as ning times would lead to substantial changes in cycle time holding a bus to protect a connection or having a bus turn that, compared to their current schedule, could reduce back early or run express. As agencies experiment with, or use, annual operating cost by an estimated $7 million (28). such actions, they need off-line tools to study the impacts of Passenger waiting time is an important measure of serv- these control actions in order to improve control practices. ice quality. Studies show that customers are more affected For example, AVL-APC data were used to determine the by their 95th-percentile waiting time--for a daily traveler, impacts of a Tri-Met experiment in which buses were short- roughly the largest amount they had to wait in the previ- turned to regularize headways during the afternoon peak in ous month--than their mean waiting time, because 95th- the downtown area (29). percentile waiting time is what passengers have to budget in their travel plans to be reasonably certain of arriving 4.1.4 Solutions to Roadway Congestion on time. Passenger crowding is also a measure in which extreme Transit agencies are more actively seeking solutions to values are more important than mean values. Although traffic congestion, such as signal priority and various traffic traditional planning uses mean load at the peak point to management schemes. They need tools to monitor whether set headways and monitor crowding, planners understand countermeasures are effective. For example, a Portland State that what matters for both passengers and smooth oper- University study done for Tri-Met using archived AVL-APC ations is not mean load but how often buses are over- data found that while signal priority reduced running time on crowded. Therefore, design standards for average peak load some routes, it had no positive effect on others (30). In that are set a considerable margin below the overcrowding particular study, only the overall effect on rather long seg- threshold. However, load variability is not the same on ments was analyzed by comparing before and after running every route. With a large sample of load measurements, times, making the results hard to correlate with particular headways can be designed and passenger crowding meas- intersections. For better diagnosis and fine-tuning of coun- ured based on 90th-percentile loads, or a similar extreme termeasures, agencies need tools to analyze delays on stop-to- value, rather than mean loads. stop, or shorter, segments. 4.1.2 Customer-Oriented Service Standards 4.1.5 Discovery of Hidden Trends and Schedules Behind a lot of the randomness in transit operations may AVL-APC data allows customer-oriented service quality be some systematic trends that can be discovered only with measures to replace (or supplement) operations-oriented large data samples. For example, by comparing operators service standards. For example, on high-frequency routes, a with others running the same routes in the same periods of traditional operations-oriented standard of service quality is the day, Tri-Met found that much of the observed variability the coefficient of variation (cv) in headway. Although such a in running time and schedule deviation is in fact systematic: standard may mean something to service analysts, it means some operators are slower and some faster. Exploratory nothing to passengers, and it resists being given a value to pas- analysis might also reveal relationships that can lead to bet- sengers (e.g., how much does it benefit passengers if the head- ter end-of-line identification, or to better understanding of way cv falls from 0.35 to 0.25?). With a large sample size of terminal circulation needs. headway data, one can instead measure the percentage of pas- It is the nature of exploratory analysis to not have a pre- sengers waiting longer than x minutes, where x is a threshold defined format. To support exploratory analysis, therefore, of unacceptability. Similarly, in place of average load factor as AVL-APC databases need to be open to standard data analy- a crowding standard, one could use a standard such as "no sis tools.