«BASIC HUMAN DECISION MAKING: An Analysis of Route Choice Decisions by Long-Haul Truckers John Holland Knorring Advisor: Professor Alain L. Kornhauser ...»
sky. The best part about the system is that it is available to all users at a cost equal to the cost of a receiver. There are no direct annual fees or charges to the users, as the system is funded by federal tax dollars. The Department of Defense started the system in the 1960’s to improve on the accuracy and availability of the Loran system. There are essentially three types of equipment that make up the GPS system: the network of satellites (Figure 3-4), GPS receivers, and land based control units.49 Figure 3-4: Image courtesy of the U.S. Department of Defense The space-based portion of GPS consists of 24 full time satellites plus 3 extra back up satellites. While in orbit, the satellites function as the transmitters and generate a pseudo random stream of numbers that is used as a timing signal for the receiver portion of the system. Because it takes a certain amount of time for the signal to get to the surface of the earth from space, there is a lag in the data stream. The satellites are
aligned so that there are 6 orbital planes equally spaced 60o apart in reference to the prime meridian. Each orbit plane contains four satellites that are spaced such that a satellite passes a point in the orbital plane every 6 hours. The radii of the orbital planes are 20,200 km. This organizational pattern ensures that a receiver will be able to pick up a signal from at least 4 different satellites anywhere on the surface of the earth at all times.
Since the satellites’ orbits are less than the required distance to make them geosynchronous, they are in constant motion relative to the earth. The ground based control units monitor and update this motion.50 The land-based portion of the GPS system is responsible for monitoring the status of the space-based portion in addition to broadcasting updating signals to the constellation. The terrestrial portion consists of six remote monitoring stations and four ground antennas located around the world in addition to a master control station located at Falcon Air Force Base in Colorado Springs, CO. The master control station is responsible for generating the orbital model and clock correction parameters for each satellite as well as relaying the correction information to the ground antennas so that the updates can be sent to the constellation.51 The receiver portion of the system allows for the signals from the satellites to be collected and decoded so that its location can be determined. The hand held receivers gather at least four signals from different transmitters. From there, they calculate the time delay in the signals. The receiver generates the same pseudo random data stream as the satellites transmit and calculates the time differential between when the signal is received and when it was sent. The problem with this method though is that the satellites
use highly accurate atomic clocks to generate their timing signals while the receivers us less accurate quartz clocks. The less accurate clocks result in greater errors in time estimation. To get around this limitation, GPS incorporates a fourth satellite signal that is used to constantly update the clocks in the receivers. Using the time delay information the receiver determines its position relative to the in-view satellites by multiplying the time delay by the speed of light as the timing signals are electromagnetic radio signals that travel at the speed of light. Additionally, the receivers store the locations of the satellites in their memory, so that they always know where the satellites are and the receiver can then generate its position relative to the satellites.52 The limitations to the accuracy of the GPS system on its own are minor, which accounts for its popularity. The errors in the accuracy of GPS come from a few sources.
First, as stated earlier, the clocks in the receiver units are only quartz clocks, which are significantly less accurate than atomic clocks and account for a loss of precision in time delay analysis. Next, electromagnetic signals travel through a vacuum at the speed of light. However, to get from space, a relative vacuum, to the earth, the signal must travel through the atmosphere where the signal slows down and varies depending on the medium that it is traveling through. The result is the precision of system is reduced.53 3.1.4 Inertial Navigation Systems Inertial Navigation Systems or INS are quite different from GPS and Loran in terms of how they calculate location information. Instead of relying on timing signals
In essence, an INS is just a piece of equipment with three highly sensitive accelerometers attached to measure accelerations in the three dimensions. The INS collects and records the outputs of the accelerometers based on the equipment’s set sampling rate. Armed with this information, along with the starting location for the system, the INS calculates the current position.
The INS records the accelerations and the direction that they occurred in. It also records the amount of time that the INS was accelerating. Taking the acceleration in the z direction as an example, one can multiply the acceleration over time of the INS times the length of time squared to obtain a distance segment in the z direction. Combining the z segment with the x and y segments, one can determine distance and direction that the INS currently is relative to the starting position.
Like GPS, there are problems with INS systems. First, noise in the sensors limits their ability to estimate parameters with a high level of accuracy. Next, INS system accuracy is limited to the rate at which new information enters the system. For example, if an INS samples acceleration once every 10 seconds, and in between the two sample times a large acceleration occurs which the INS does not pick up, the INS will not know that it has changed direction and will think that it is someplace that it is not. In addition,
the state estimation accuracy is directly related to the motion of the system. Lastly, the reliability of the system is highly dependent on obtaining external position updates.54 3.1.5 General Comments on Accuracy of Navigation Equipment The collection of position and navigation information is always an inexact science. While the accuracy of the collection equipment is not exact, for this study, they all have a sufficient level of accuracy. Many of the tools used for navigation rely on sensing of fields that are external to the actual equipment. Compasses rely on the sensing of magnetic fields that are not constant. Also, GPS and Loran sense electrical fields that vary with time and for a variety of reasons are also not always available. INS systems, though they do not require the sensing of fields outside their self-contained equipment, also have fundamental problems relating to accuracy. Going forward however, increasing accuracy and precision of the equipment can be accomplished by augmenting the primary collection method with additional alternative collection methods.
3.2 Complications and Difficulties with Data Extraction There are a number of difficulties involved with the collection and use of this data set. A short list of the complications will be discussed below.
3.2.1 Origin-Destination Information One of the most important factors in analyzing a representative traveler in any study examining routing and travel demand is obtaining origin-destination (OD) information. It is necessary to have the origin and destination information of the drivers, so that one can determine where the truck started and where it stopped on any given trip.
This data set did not explicitly specify separate trips, rather, it contained all of the location information for all of the trucks as a single grouping of consecutive observations.
In other words, the data came as a single chunk of the entire 13 days for each truck.
Additionally, the data set does not divulge anything that occurred between separate location observations. The result is that it is impossible for the analyst to know if a driver was on a single trip, or had multiple trips linked together. Also, because the observations were generally separated by about 45 minutes, it was impossible to tell exactly what route the truck followed between consecutive observations. This will make a large assumption that the driver took the shortest path between separate observations. While this assumption could seriously damage the accuracy of the analysis, it appears as if it is a fairly safe assumption.
3.2.2 Non-Random Sample of Trucks One important consideration to take into account for this study is the possibility that the sample of trucks in the database is not a totally random sample of trucks in general. This is an important consideration because if the data is not totally random sample of the population of trucks, then the decisions shown in this representative sample could potentially be different from the decisions displayed by the rest of the truck population. For this study, it is likely that this is not a random sampling of trucks because the trucks in the data set had to have special position gathering equipment. The trucking company was required to purchase the equipment, and thus many truckers are excluded from the data collection.
3.2.3 Random Time Spacing of Observations One other factor that further complicated the analysis for this thesis was the lack of uniformity of time spacing of observations. Each truck had a different number of observations over the two week period that were not equally spaced apart in times.
Furthermore, the frequency of reporting of position information was also varying such that a truck could report five observations spaced 20 minutes apart, and then the next observation might be 4 hours later. This made the analysis very difficult to determine some of the route choice decisions being made. As will be shown later, the time spacing was also a large factor when analyzing some of the smaller sections on the US highway network. There was no guarantee that there would be an observation in the study area even if the truck did pass through the respective area. Another assumption was made that further complicated the accuracy of the analysis. This study assumes that the truck
traveled at a constant rate between two observations. This can potentially reduce accuracy of analysis because the interpolated speed of the truck might not be the actual speed of the truck. As a result, this study made an effort to reduce this problem by limiting the number of acceptable trips to those that occurred in a “reasonable” amount of time. The specific amount of time will be covered later in the discussion of the analysis.
3.2.4 Observation Reporting Error and Map Matching Complications As stated previously, neither INS, GPS, or Loran systems can determine exactly where the truck is. There is always some error. Depending on the system used, this error could be as little as 1 m or as much as 5000m. However, they do provide a pretty good estimate for the location of a truck. In this study, the position observations were gathered and matched to the U.S. highway network using a complicated algorithm courtesy of ALK associates Co Pilot® navigation system that would “snap” the observation to a specific link on the network. The problem with this method though is the location of the observation was not always on a highway or a road for that matter. Additionally, even if the observation snapped to a major highway, many highways have other roads that run parallel to the highway that trucks do drive on. If the matched location is not the actual location, it is possible for the accumulated errors to not accurately portray the decisions made.
3.2.5 Stop Determination One last factor that posed problems for the data analysis and data reduction was the determination of stops. Stop determination is very important because it signifies the beginning or the end of a trip. The data set did not contain any information that explicitly stated that the truck had made a stop. However, this thesis used a heuristic to estimate stops. There are a number of areas in the data set where there is a significant time differential in two consecutive time tags and only a small difference in the reported position of the truck. A stop was defined as any point in the data set where two or more consecutive observations were close enough together such that a truck traveling at 4 miles per hour could go between the two points in a 30 minutes or less. This heuristic was used in the data reduction stage of analysis, but it presents a problem. Because of the relative infrequency of observations, it is possible that a truck could have been at a terminal, left to make a delivery, and returned before any position observation was made.
This is important because many short trips are left out of the data analysis.
3.3 Additional Data Considerations As stated previously, there are many factors that contribute to the utility function of drivers. Many of those factors however are difficult to gather. Because most of those factors are not included in the data set, the precision of analysis is reduced. The following sections will cover a few of the more important factors in the decision making process that were not included in the data set.
3.3.1 Observed Traffic on Routes As stated previously, traffic is a very important factor that effects decision making by truck drivers. It is possible that the drivers had information about either their primary route or an alternate route that effected their route choice decisions. While it would significantly increase the complexity of the data analysis, it would be very helpful to have the traffic information for the area around the trucks. This information could be incorporated into the decision making model and could potentially dramatically increase the precision of the analysis.
3.3.2 Socioeconomic Factors Socioeconomic factors have been previously demonstrated to dramatically effect decision making. The lack of socioeconomic data in the database is therefore problematic for an analysis such as this. Not only does the lack of socioeconomic data limit the precision of the analysis, but also in applying the results of this study to the general population of truck drivers, it is difficult to make predictions of decision making