Integrating TMS with AVL Systems

Edward Sitarski, Vice President of Technology (e.sitarski@schedule-masters.org)

Introduction

Advances in the technology of Global Positioning Systems (GPS) have now allowed GPS receivers to be cost-effectively placed on transit buses. GPS systems receive signals from an array of satellites positioned worldwide so that at least three satellites are in direct line of site at all times. By triangulating from the signals transmitted from these satellites, GPS receivers can determine their latitude and longitude location to an accuracy of a few meters. The GPS satellite location network was created by the US military and is available for general civilian use.

Automatic Vehicle Location (AVL) systems use GPS technology. They take the coordinates from the GPS systems and use them in conjunction with Geographic Information Systems (GIS) to determine the travel time of vehicles between any locations in the street network.

For many years, transit properties have been using Computer Aided Transit Scheduling (CATS) systems to assign vehicles and drivers in the most cost-effective manner. These systems use sophisticated techniques and user interfaces to help schedulers find the least-cost solution while meeting all service requirements. CATS systems have been shown to dramatically reduce costs and improve asset utilization when compared to manual scheduling methods.

However, CATS systems rely on basic data which describe the running times of the buses on the street. In the past, this data was collected manually timing vehicles with stop watches. Because this is an expensive, tedious and error-prone process, it is not frequently performed for all times of day, for all routes and in all seasons and weather conditions. The running time data used for scheduling was only a rough approximation of the real travel times. This can cause poor schedule adherence and excessive customer waiting, missed connections, loss of customer confidence in the transit system and reduced revenue.

Studies have shown that a key factor in increasing transit ridership satisfaction is for the buses to run on time. This requires that accurate times be used to create vehicle schedules.

Unfortunately, actual vehicle running times depend on many dynamic factors such as traffic levels, season and vehicle occupancy. These factors can and do change over time. A transit property should adapt its schedules and services to meet the needs of the changing ridership as closely as possible. This will increase customer confidence in the published schedules, increase customer service and ultimately increase ridership and revenue.

AVL systems are a cost-effective way to automate the collection of true running times on a scale that could never be achieved before with manual methods. These true running times can be used by the CATS system to create much more accurate schedules.

This paper describes features of an integrated AVL and CATS system to maximize the value they add together to a transit property including:

• using AVL data to determine more accurate running times for scheduling
• using passenger level data to do service planning
• using AVL data to determine optimum trip layover times to guarantee on-time service objectives

 

Processing the Data

AVL Data: AVL systems provide vehicle running time data between any two time points. This data includes the time of day, route, vehicle number and the direction of travel.

Actual running times depend on many complex, dynamic and interdependent factors including:

• season
• service day (for example, weekday, weekend or holiday service)
• current weather conditions
• current traffic conditions
• current vehicle occupancy
• loading/unloading patterns of passengers

This variability makes it essential to use statistical analysis to determine average running times useful for planning and scheduling. Multiple measurements between the same time points and direction need to be statistically combined to determine both the average running time and the expected variation. Data for many days of the same service (for example, all weekdays) also needs to be included.

It is important to note that we can never really know the "natural" or "real" average running time, we can just estimate it by repeated measurements. The more measurements we take, the more accurately we can approximate the "real" time.

Since it is likely that traffic conditions, vehicle capacity and passenger movement on a service day are likely to vary with the time of day, it makes sense to group running time measurements into data sets defined as follows:

• same season
• same service day
• same route
• same direction
• same hour of the day (for example, measurements taken between 7:00am and 8:00am. This interval could be increased or decreased depending on the amount of data available.)

Each of these data sets would have a statistical mean (average) and variance. The mean value would be considered the expected running time of a vehicle on that service day, that route, that direction in that hour. The variance is measure of how predictable the running time is. A small variance indicates that the actual running time is usually close to its average value. A large variance indicates that large running time differences are common and that the average time is not a good predictor of the actual time. The variance is a very important measure that can be used to predict how likely the vehicles will be able to keep to the schedule.

The remainder of this document focuses on how these data sets can be used to improve the scheduling process.

CATS Data:

Modern transit scheduling systems describe running times with rules. These rules work on a general-case/exception-case basis. For example, a general case rule might be "the running time between A and B is 10 minutes". A more specific rule might be "the running time at 7:00am to 9:00am between A and B is 15 minutes". When the scheduling system needs to determine a running time, it uses the most specific rule that applies. In the previous example, these two rules mean that for all service days, all routes, all patterns and all directions, at all hours of the day the running time between A and B will be 10 minutes except between 7:00am and 9:00am, when it will be 15 minutes.

The Master Scheduler allows further criteria to specify running time rules including:

• service
• time of day
• route
• pattern
• direction

These exception categories allow times to specified down to an individual trip on a particular route if necessary. However, if this level of detail is not required, schedulers do not have to be burdened with entering and maintaining all this low-level data and can quickly enter a general rule.

The advantage of this rule approach is that dramatically less information is required to describe all the running times with no loss of accuracy. Vehicles that travel the same "trunk line" and short-term routes can share the same running times from common rules. Fewer rules mean that schedulers can be much more responsive to systematic changes in running times. For example, if road work caused a detour on a main street, the scheduler would only have to update that one running time rule. All routes that matched that rule would then automatically get that new time. This exception-case philosophy allows the schedulers to be more responsive by quickly dealing with problem areas only.

From a scheduling perspective, using AVL data involves analyzing:

• that existing running time rules are accurate
• where and when new exception rules should be defined
• whether past exception rules should be removed
• how weather and season affects running times and what rules are needed to describe these times
• whether service levels should be increased/decreased based on vehicle occupancy levels
• what minimum layover times should be set for trips to meet statistical on-time service levels

Features of Integration

This section describes features which are possible from the integration of AVL and CATS systems. The added business value of each feature is also explained. The detailed technical design of these features is described in the document "AVL Integration with The Master Scheduler, Detailed Design Document", Edward Sitarski, 1996.

Travel Time Exception Report

This report shows the deviation of each AVL data set from the applicable CATS running time rule. The report is sorted so that the greatest deviations are listed first. The deviation is defined as the statistical probability that the CATS running time does not match the average running time of the data set. This is determined with the statistical "student-t" test. Schedulers and Planners analyze this report to determine what corrections and modifications (if any) are required to the running time rules. After the corrections are made, this report is generated again to verify that the changes achieved the desired result.

Value:

• verify that running time rules used for scheduling match AVL determined average running times
• indicate where quick adjustments should be made in scheduled running times to quickly follow the requirements of seasons, road work and ridership
• more accurate schedules; better schedule adherence

Vehicle Capacity Exception Report

In addition to vehicle location and running time data, some AVL systems can also track the number of passengers on the bus. This report shows when the average bus capacity of an AVL data set exceeds a maximum or minimum value at different times of the day. The report is sorted with the greatest violations first. Two reports are typically run: maximum violations and minimum violations.

Value:

• quickly determine if changes should be made to the frequency of service, both increasing and decreasing
• indicate where to cost-effectively deploy vehicles to best meet the needs of ridership
• improve customer service and convenience

Connection Time Rule Extraction

In addition to the Connection Time Exception Report, it is possible to derive all travel time rules automatically from the AVL data sets. This can be done without any intervention or data entry by the scheduler. This process uses sound statistical arguments to combine different AVL data sets together and describe them with one running time rule. The result is the minimum set of rules which described the data to a given accuracy. This accuracy is defined as a statistical "confidence level" that two data sets actually have the same average. For example, a 95% confidence level means that two data sets can be described by the same rule if we are 95% sure that the two sets have the same average running time. This can be determined with a form of the statistical "student-t" test. The set of derived rules ends up as the smallest possible given the given accuracy. Schedulers can be more responsive to changes in ridership if they have fewer rules to maintain. The report can be run at CATS implementation time if an AVL system had already been installed at the property and true AVL-based running time was available.

Value:

• more accurate rules - rules can be chosen based on statistical analysis and description accuracy
• better schedule adherence
• fewer rules to maintain mean less work for schedulers and faster response to running time changes

Minimum Layover Time Optimization

A trip consists of a set of locations that are strung together and driven by a vehicle. At the end of each trip, the scheduler can specify a minimum layover (waiting period) which is a time interval that the vehicle must wait before performing other work. One of the purposes of minimum layover is to allow the vehicle to "catch up" if it is running late. A large amount of minimum layover will improve schedule adherence but will reduce the utilization of the bus (more waiting); too little layover may decrease adherence (no catch up time) but better utilize the vehicle. The optimal minimum layover time meets schedule adherence policies and maximizes vehicle utilization. A schedule adherence policy can be defined as the percentage of trips that are expected to leave on time. For example, "It is policy to assign enough minimum layover time so that 95% of the time this layover is enough to absorb any delays in the trip". This policy would ensure that 95% of the trips leave on time (statistically). The variance computed from the AVL data sets can be used here. The sum of all the running times along a trip can be statistically analyzed with another form of the "student-t" test to deter-mine the minimum amount of layover required meet any schedule adherence policy.

Value:

• ability to guarantee statistical schedule adherence figures to the public. For example, "Our buses are always 95% on time"
• allows the property to explore different schedule adherence policies and their effects on driver and vehicle costs and utilization

Basic Integration

All of the scenarios described above rely on the AVL system to be able to import the timepoints to measure from when the vehicles pass. These timepoints would normally be specified and maintained in the CATS system.

Additionally, it must be possible to pass the route numbers from the CATS system to the AVL system.