A Predictive Model and Evaluation Framework for Smart Parking: The Case of ParkPGH Tayo Fabusuyi1, Robert C. Hampshire2, Victoria Hill3 and Katsunobu Sasanuma4
Abstract: ParkPGH is a smart parking system that uses historical parking and event data in a prediction model to provide real-time information on the availability of parking in eight parking facilities within the Pittsburgh Cultural District. The project’s unique characteristics include the collaborative efforts of funders, academia, nonprofit and for-profit entities that are involved in the project, the environment in which the product is deployed and the richness of the data from which the prediction model and the robust evaluation strategy draw upon. The paper describes the pilot phase of the project along with preliminary results, as well as the next steps to be taken for a full project implementation.
INTRODUCTION ParkPGH is a smart parking project designed to give real-time information on the availability of parking to patrons within the Pittsburgh Cultural District, a geographically-designated area within the city of Pittsburgh. This area is home to the arts and entertainment scene supported by the Pittsburgh Cultural Trust (PCT), a nonprofit arts organization established in 1984 to lead the cultural and economic development of downtown Pittsburgh primarily through the use of the arts. The organization was inspired by H.J. Heinz II’s vision for a world-class arts and entertainment center in Pittsburgh. The Trust presents performance art, owns and operates galleries, collaborates with other arts organizations, and is involved in real estate development within the Cultural District. Since its inception, the PCT has witnessed significant increases in attendance and patronage within the Cultural District. Attendance at performances within the District surged from 570,000 in 1990 to 2 million people in 2008, an increase of more than 250%. This development has placed considerable strain on the existing amenities within the District, particularly parking facilities, a situation further compounded by the scale of activities on the North Shore and the added demand for parking from sporting fans. A series of initiatives have been put in place by the PCT over the last few years to help alleviate this problem, primarily on the demand side for parking spaces but very little value added was observed from these interventions. These initiatives included promoting fringe parking on the North Shore and using shuttle buses to transport patrons to event venues. This concept was discontinued because very few patrons made use of the facility. The Trust also offers pre-paid parking to its high-level patrons although the pre-paid parking is not bound by contract and the provider of the service is not compelled to provide parking spots. As a result, the PCT is reluctant to widely publicize this service. To address this problem, PCT, with generous funding from the Benter Foundation5, initiated ParkPGH6, a smart parking, technology based, pilot program within downtown Pittsburgh. The program will enhance the existing off street parking facilities within the District by providing real time information using a host of 1
Lead Strategist at Numeritics,
[email protected] (corresponding author) An Assistant Professor of Operations Research and Public Policy, H. John Heinz College Carnegie Mellon Univeristy,
[email protected] 3 Principal and Research Scientist at Numeritics,
[email protected] 4 A Ph.D candidate at H. John Heinz College Carnegie Mellon Univeristy,
[email protected] 5 http://www.benterfoundation.org 6 http://parkpgh.org/ 2
1
information delivery methods that includes an iPhone application, traditional and mobile website, text messaging and an interactive voice response system. The primary goals of the program are to reduce search time and search time variability when finding a parking space within the Cultural District and to make the District a more desirable destination for patrons by reducing the anxiety and uncertainties related to parking issues. A secondary goal is to attract new patrons who were previously deterred by the uncertainty of parking availability. Currently, the pilot program monitors eight parking garages totaling 5000 parking spaces, representing approximately 20% of the total parking supply in downtown Pittsburgh and over 90% of the parking supply in the cultural district. This large market share of parking spaces in the Cultural District provides a unique opportunity to evaluate the impact of a smart parking system. In addition, the PCT possesses large amounts of demographic information about patrons, as well as ticket sales data to cultural events. This presents an unparalleled opportunity to conduct a robust program evaluation strategy. The PCT information also enables future enhancements to ParkPGH based on ticket sales data and cultural preference information. To the authors’ knowledge, neither of the two features mentioned above have been previously conceived or studied in the literature. The paper addresses these issues and also contributes to the body of knowledge on smart parking by addressing the unique environment within which the product is employed, the partnerships involved and the added complexity introduced by these elements, the use of an event-based parking prediction model for pre-trip planning; and developing a novel longitude program evaluation strategy for a smart parking system. The paper is organized as follows. The paper’s theme and contributions are placed squarely within the context of existing literature on smart parking systems in the next section. The third section gives a summary of a patron survey that shows the need for the smart parking system and presents baseline data obtained from the survey. The fourth section provides a detailed description of the unique environment in which the ParkPGH product will be deployed and enumerates further on the event-based parking prediction algorithm using a specific garage and the evaluation framework to be utilized. The final section presents our conclusions and a discussion of the results obtained.
LITERATURE REVIEW A host of initiatives exist that seek to alleviate parking problems. These initiatives range from zoning changes and tax code modifications to pricing changes and the provision of information. The initiatives involving information technology are often called smart parking solutions. A large academic literature exists for this set of systems. For the purpose of this paper, we organize the relevant literature along two dimensions of smart parking solutions: parking guidance systems, and real-time vs. prediction information. The papers cited below are not meant to be exhaustive, but merely representative of the streams of literature in existence. The first dimension of the literature is concerned with the design and behavioral responses to parking guidance systems (PGS) which use variable message signs (VMS) to inform drivers about available parking spaces. These systems typically provide information about spaces inside parking facilities. Much of the literature on parking guidance systems is concerned with transit and park-and-ride lots (1). The system described in Orski (2), which is based on the approach utilized in Cologne, Germany, is representative of parking guidance systems connected with transit. In the U.S., parking guidance systems have been deployed in many places including Chicago and the Washington D.C. area to alert drivers of available spaces in parking facilities located at transit stations (RITA Report). Another stream of work on PGS explores their use inside of parking facilities ((3) and (4)). ParkPGH is distinct from the parking guidance
2
system literature in two ways: ParkPGH is not coupled with transit, and it does not employ VMS. The parking availability information is only available through mobile devices, interactive voice response (IVR) and the Internet. The other stream of research literature related to this paper examines the display and use of information for finding parking spots, using either real-time or prediction information. This literature is mostly concerned with providing drivers with real-time information about parking availability. The literature on parking prediction models is divided broadly into parking prediction algorithms for finding parking during a trip and parking prediction for use before the trip begins. The works of Caliskan et al. (5) and Teng et al. (6) are examples of the former that provide parking prediction models based on information exchanged between wirelessly connected vehicles for use during a trip. The prediction model described in this paper is an event-based parking prediction model for use before a trip begins. The prediction model uses historical parking and event data to predict future parking availability. These predictions have been shown to reduce the uncertainty often related to parking in downtown areas and central business districts (7). For example, an individual coming downtown to watch a performance could establish, with some degree of certainty, the probability of finding a parking spot assuming the Pittsburgh Penguins are playing the same evening. Drivers may subsequently incorporate these parking predictions into their pre-trip planning.
BASELINE DATA A needs assessment was conducted to determine who the target audience is to ascertain what issues and concerns presently exist surrounding parking within the Cultural District, and in consultation with stakeholders, to determine what the ideal state should be. The process revealed that the patrons of the PCT are the intended beneficiaries and that an appreciable number of these patrons are dissatisfied with the current parking situation. Key program objectives include reducing parking search time and search time variability, decreasing the incidence of late-coming as a result of difficulties with finding a parking spot and improving patrons’ perceptions about parking downtown. It is also expected that a reduction in uncertainty about finding parking will attract new patrons to the Cultural District. Baseline data obtained with respect to the primary objectives are presented below. Table 1: Baseline Data on Key Program Objectives Program Objectives Parking search time Search time variability Late coming incidence Patrons' perception about parking (% surveyed without a positive response ) Parking satisfaction Ease of finding a parking space Overall parking experience
Data Obtained 6.6min 6.0min 26.6% 25.7% 21.9% 24.3%
Data for this phase was collected through a combination of in-person and online surveys administered to patrons attending Pittsburgh Cultural Trust events. In all, a total of 736 individuals were surveyed about their perceptions on parking within the Cultural District in the time period between September 18 th 2010 and January 23rd, 2011. Parking search time is measured by the mean search time while Search time variability is measured by the standard deviation from the mean. A low standard deviation number indicates less
3
variability, whereas a high standard deviation number shows that the data points are more dispersed around the mean. More than one out of every four respondents said they have been late for a Cultural District event because they had difficulties finding a parking spot. The same ratio also reported not having a positive experience with overall parking when coming for a Cultural District event. The needs assessment phase also revealed that patrons are reluctant to use fringe parking lots partially because of security issues and because of the long walk during the winter season. Seven out of every 10 persons surveyed revealed their preference for closer proximity as compared to lower price. We also carried out analyses on subsets of the data collected to ascertain if there are distinct trends that could be localized to these cohorts of observation. Among the subsets of the population analyzed, noticeable differences were observed for Search time and Search time variability. For example, average Search time ranges from a low of 6 minutes for the in-person cohort to 8.1minutes for patrons arriving close to curtain7. In contrast, the highest magnitude for Search time variability of 6.2 minutes was observed for the population surveyed in-person compared to only 4.9minutes for the online cohort.
PARKPGH DESCRIPTION The ParkPGH project has a number of unique characteristics. The project includes ten different stakeholders ranging from local foundations, to academia, to public sector agencies and privately owned firms. In addition, the parking assets that are featured in the pilot program have different management and ownership structures. The parking facilities are owned and operated by a wide variety of entities including public, nonprofit and private sector entities. This fragmented ownership adds to the challenges of creating a robust strategy to achieve the program objectives. For example, the city-owned garages that are managed by the Pittsburgh Parking Authority (PPA) employs unionized workers with specific workplace rules as compared to the Alco Corporation owned and operated garages where this is not the case. Added to the complexity of the project are significant issues with the parking garages themselves. The different ownership and/or management structure makes it extremely difficult to design a standard approach that will be amenable to all the garages. When the project was conceptualized, it was thought that there was a standardized method of calculating the number of currently available parking spots in the garages, along with a way of determining when the garage was identified as being “full.” However, each parking garage has its own “culture” of determining how and when to identify the garage as being “full.” Variables that factor into that decision include the number of leased spots to hold open, which can vary by time of day, number of events being held that day, at what point of capacity does the “full” sign go up. In fact, some of the garages even distinguish from a “hard full” and a “soft full.” This lack of standardization has made for significantly increased complexity in the algorithms that are used in the ParkPGH applications.
Usability and Delivery Channels In order to address usability issues, we have embraced traffic sign colors in providing information to patrons looking for parking spaces. The green and yellow color coding will be complemented with a numerical figure that shows the available number of parking spots, except in cases where the garage is deemed low/full, in which case the text will be superimposed on the red color coding. The delivery channels that will be used to relay this information include the following: This cohort was arrived at by culling email addresses of patrons whose tickets were scanned both close to curtain time and shortly after the performance has begun. 7
4
iPhone: A free iPhone app is available at the iPhone App Store. The app features a scrollable view, listing each available parking facility and its parking space availability. The application will be periodically updated once every minute. In the app, clicking on the parking lot will reveal more information, including the facility address, map and pricing. Patrons are able to view the map, space availability as well as the user’s current location if GPS is enabled. This will help visitors to pick the closest parking lot to their location or their destination. ParkPGH.org: The website, ParkPGH.org, offers a web page with a list of parking facilities and their parking space availabilities. Clicking on a facility will provide the user with an address, map, and options to retrieve directions. Pricing information will also be displayed. In addition to parking garage information, popular destinations will also be displayed so that visitors can locate their targeted destination and find the closest available parking. The information on the main page will be periodically updated so the end-user can leave it open for a prolonged amount of time. A snapshot of the website showing Pittsburgh downtown, destinations within the Cultural District, garages and the available spaces is provided below.
Figure 1: ParkPGH representation of Pittsburgh’s downtown map with available parking spaces
The website will also allow users to subscribe to parking alerts. For example, a user could elect to receive a text message notification if a specific parking garage fills up before a chosen time on a
5
certain day. One situation that these alerts would be useful for is if a person is attending a 7 p.m. show and wants to be alerted if the lot they planned to park in fills up before 6:30 p.m. The alert will also suggest nearby garages with available parking. Mobile Website: A mobile version of the website, m.parkpgh.org, provides the same information as the traditional website but will be optimized for mobile devices, focusing on the Blackberry and iPhone.
Figure 2: ParkPGH Smart phone App
Figure 3: ParkPGH Smart phone Detail Screen
Figure 2 shows a screenshot from the mobile website with the available capacity superimposed on the color coded boxes. The reader may observe that the exact available number of parking spots is not provided when the garage is deemed nearly full. Double clicking on any one of the garages, in this case, the 6th and Penn garage, produces more detail on the garage chosen as could be seen in Figure 3. Information provided includes rates and the theaters in close proximity to the garage. SMS: Visitors can text PARKING to 412-423-8980 to obtain a list of downtown parking lots in the order of available space. Voice: Similar to the SMS offering, visitors may call 412-423-8980 to receive a list of parking lots with available space. A text-to-speech system will announce the parking lot names and the
6
percentage of available spaces for each lot. The call can be interrupted at any time by pressing the number of the parking lot (1, 2, 3. etc.) to get location and pricing information for the respective lot.
PREDICTION MODEL In this section, we present a parking model for the Theater Square garage in downtown Pittsburgh. The model predicts parking vacancy based on events such as theater performances and sports games held in Pittsburgh’s Cultural District and is served for those who are planning to drive and park to watch the events. We predict the number of vacancies Y(t) at time t with ten minute interval at Theater Square garage using a multiple regression model: , where X(t) is the predictors (a set of events at Cultural District) at time t, is the coefficients of predictors (an impact on the number of vacancies by the corresponding event) at time t, and e(t) is the error term. is estimated following the procedure of OLS (ordinary least squares): First, we define RSS(t) (residual sum or squares) as a function of . This is an indicator for the level of prediction errors. . is minimized when ̂ Here, ̂ is the OLS estimator of ̂ ̂ become and
(
)
.
. Using this ̂ , a prediction of Y(t) and the minimum RSS(t) ̂ )( ̂ ), respectively. (
The goodness of fit of the OLS model we used can be evaluated by determination): ̂
(coefficient of
,
where SYY(t) is sum of squares for the yi’s at time t.
Parking Data The parking data set was provided from Alco parking Theater Square garage. The number of available parking spaces, which we call Y(t), is collected with 10 minute interval for 24 hours (144 data per day) from 11/9/2008 to 7/10/2010 (609 days or 87 weeks). We dropped two day data (9/24/2009 and 9/25/2009) from the data set since almost nobody could use the Theater Square garage during G-20 Summit period in Pittsburgh, hence, the 144x607 parking data matrix. Each element of the matrix Y ranges from 0 to 691, where 0 and 691 correspond to no parking availability and full parking availability, respectively. We first
7
split the data set into each day of the week and holidays.8 The average numbers of the available parking spaces (or vacancies of parking spots) corresponding to each day of the week and holidays are as follows:
Figure 4: Day of Week Mean Parking Availability
Monday-Friday average show a similar pattern: the availability of parking spaces become very low between 10am and 3pm, although we can also observe some differences, such as there is no drop after 5pm on Monday but a huge drop after 5pm on Friday. Compared to weekday average, weekend and holiday average are consistently higher from midnight to around 5pm, but there seem to be no common pattern among Saturday, Sunday, and holidays averages. Part of the reason for this variation is that weekend/holiday parking availability is affected significantly by various events held on weekend/holiday in downtown. We aggregate all weekday data together and all weekend and holiday data together. The following figures are the average available parking spaces for weekday and weekend/holiday and their corresponding variances.
Figure 5: Mean (left) and Variance (right) of the number of vacancies on weekday and weekend/holiday
A huge drop in the number of available spaces observed between 10am and 3pm on weekdays is considered to be work related because the number of spaces is more or less stable (low variance), where we can use a historical data to predict the available spaces. In contrast, a drop at around 3pm on weekend and drops at around 8pm on weekend and weekday are considered to be event related because the number of spaces 8
There are 18 federal holidays included in the data set.
8
fluctuates heavily (high variance) depending on the occurrences of events. In fact, variances have three clear peaks at 3pm on weekends and at 8pm on weekdays and weekends. Our task is to predict weekday night and weekend day/night parking availability when their variances are high and the parking spaces become scarce.
Predictors X and their OLS Coefficients ̂
The ̂ can be interpreted as an impact of the corresponding predictor X(t) on the number of vacancies Y(t). We used the event data set from Pittsburgh Cultural Trust. We picked only major events operated in Benedum Center, Byham Theater, O’Reilly Theater, Heinz Hall, Pirates baseball game, Steelers football game, University of Pittsburgh football game, and Penguins ice hockey game.9 We split all events into 1) morning event (-12:00pm), 2) day event (12:10pm-4pm), and 3) night event (4:10pm-). (Note that there are no events for some combinations. For example, there are no Pirates/Steelers/Penguins games before noon.) We also include rain precipitation and amount of snowfall in the predictor X, which are measured in inch10 using a gauge. Each element of coefficients ̂ corresponding to each predictor can be interpreted as an impact of the predictor on the total number of available parking spaces. For example, if an element of ̂ is -200 at 8pm, this means that the corresponding event reduces the total available spaces by 200 on average. The following six graphs show the change of OLS coefficients ̂ over 24 hours. The left three graphs show the impact of day events, and the three graphs on the right show the impact of night events. These results show that an impact by events is negligible on weekday daytime and non-negligible on weekday night and weekend day/night, and their impact on the available spaces are similar regardless of time when it is nonnegligible. For example, the peak impact of Benedum day event on the weekend is around -250 at 3pm, which is similar to the peak impact of Benedum night events on the weekend/weekday at around 8pm.
Figure 6a Impact of Heinz Hall event during daytime (left) and night (right) 9
Event information of Convention Center is not included because those who participate at Convention Center mostly park at its on-site parking garage or at Strip District parking. Penguins game data is not included in the event set provided by Cultural Trust since the impact is considered to be small. This is because the event location(Civic (Mellon) arena) is a bit far from the Theater Square garage. We included Penguins events, but it turned out to be a small impact, as expected. 10 For example, one inch of rain fell in a 24-hour period would create a layer of water with one inch of thickness over the ground if water is not absorbed. Reference: http://geography.about.com/od/physicalgeography/a/precipitation.htm
9
Figure 6b Impact of Byham Theater event during daytime (left) and night (right)
Figure 6c Impact of Benedum Center event during daytime (left) and night (right)
Introduction of Interaction Terms Two or more major events sometimes occur at the same time. Even when they occur jointly, the number of available parking spaces must be non-negative. As a result, without considering interaction terms that account for joint occurrences, the estimated impact of each major event (coefficients of predictors) could be underestimated. We include interaction terms between major events in the same time slot and reevaluate the impact of each event again. Figure 7a shows the impact of a Steelers football game without interaction terms and Figure 7b shows the one with interaction terms.
10
Figure 7a-b: Impact of Steelers Games on Parking with and without interaction terms
Without interaction terms (Figure 7a), the impact of a Steelers’ night game on a weekend is smaller than the impact on a weekday. The impact on weekends is underestimated heavily because there exists many joint events on the weekend (there are 8 Steelers’ games which involved joint events during the weekend in our data set compared to only one joint event during the weekday). With interaction terms (Figure 7b), the impacts on a weekend and a weekday become similar to each other.
Explanation of Variance The analysis so far suggests that our model (which includes interaction terms) explains the variation in the number of parking spaces well on weekday night and weekend day/night. The R2 graph shown below confirms this. We have high (>0.8) R2 on weekday night and weekend day/night. The times coinciding with high R2 are the moments when the highest variation in available parking spaces are observed.
2
Figure 8 Hourly Evolution of R
A high R2 shows that the prediction model we developed is effectively used on weekend daytime/nighttime and weekday nighttime, when fluctuation is high and parking availability is low. The model can be applied to all other parking garages in the same district using the same event set.
11
OUTPUT, PROCESS AND OUTCOME MEASURES In order to improve on and ascertain the value added by ParkPGH, we are presently tracking a series of indicators. We are using count data to track output measures that include the weekly usage volume for each of the delivery channels used to provide information by ParkPGH. This includes iphone app, mobile and traditional website usage, number of text messages sent on request, number of automated phone responses, the average duration of page views for each of these sessions and the bounce rate for each web visit. A sample of the measures being tracked is shown below. Figure 9 shows the number of daily requests for selected delivery channels between January 1st 2011 and July 31st, 2011. The usage volume is typically higher during the weekdays compared to weekends except when events are scheduled. For example, the noticeable spike in usage on the weekend of June 3rd to the 5th is attributed to the Pittsburgh JazzLive International, a weekend of music that includes outdoor stages, visual art shows, musicians of international repute and a JazzLive crawl.
Figure 9: Number of daily requests for selected channels
In addition, process measures are being utilized for formative evaluation purposes. Information obtained from these measures are used to make modifications to the smart parking project. Ease of use, difficulties with design and accuracy of the information provided are some of the process related measures currently being tracked. A negative response to any of these measures prompts an open ended question that allows the respondent to provide detailed information as to the nature of the problem being encountered. Such information is subsequently relayed to the development team. Table 2 shows the data obtained for some process related measures for a total sample size of 43 respondents. Table 2: Process related measures PROCESS MEASURES Usability Ease of use % of respondents with positive response % of respondents that are indifferent % of respondents with negative response Difficulties with Design % of respondents that experienced difficulties with product design Information Accuracy % of respondents that said information provided was not accurate
%
77.1 20.0 2.9 8.8 8.6
While only 3% of respondents reported having difficulties using ParkPGH, approximately 1 out of every 12 reported experiencing difficulties with the design or with the accuracy of the information
12
provided. Some of the problems users experienced include difficulties telling a landmark apart from a parking structure; recorded messages not being clear; the cursor hovering over the icon and sending a text request too early. Outcome measures that document the impact of ParkPGH are provided in Table 3. Approximately, 1 out of every 2 respondents reported that the application has reduced the time it takes them to find a parking space. The magnitude of the reduction in search time ranges from as little as a minute to more than 6 minutes with individuals reporting a 4-6minute reduction in search time being in the majority. Table 3: Outcome related measures DOCUMENTED IMPACT ParkPGH has made finding parking spaces easier % of respondents with positive response % of respondents that are indifferent % of respondents with negative response Specific reduction in search time % of respondents saying there is no reduction in search time % of respondents with 1-3min reduction % of respondents with 4-6min reduction % of respondents with more than 6 min reduction in search time
% 57.2 42.8 0.0 51.4 17.1 22.9 8.6
CONCLUSION AND DISCUSSION OF RESULTS Since its inception in 1984, the PCT has witnessed a significant increase in patronage within the Cultural District, a development that has placed considerable strain on the existing parking facilities within the district. ParkPGH was designed specifically to address this problem. The project will provide real-time information about the availability of parking in facilities in close proximity to events in the Cultural District. This paper describes the pilot application of the project that currently includes eight garages with a capacity of 5000 parking spaces or approximately 20% of the total parking supply in downtown Pittsburgh. ParkPGH has a number of exceptional features. Apart from the myriad stakeholders involved in the project, the prediction model employed is unique in its use of historical data as well as actual event data. Secondly, the PCT possesses a rich set of demographic information about patrons, ticket sales and cultural preference information that presents an unparalleled opportunity to conduct a robust program evaluation. The program also has distinct challenges that are peculiar to the environment in which the product is deployed. The fragmented ownership of parking assets and the absence of a common operational standard for the garages added to the project’s complexity. Information on value added by the pilot product is also provided. It must be understood that, while this paper describes the ParkPGH program in its concept, the pilot prediction model has been implemented using a single parking facility in Pittsburgh’s Cultural District. Future directions include applying the predictive model to the other parking facilities in the Cultural District.
13
ACKNOWLEDGEMENTS The authors would like to thank Marc Fleming and John Mumper of the Pittsburgh Cultural Trust for guidance and Merrill Stabile, Jim Funovits and Don Levkus from Alco Parking for providing us with parking data set, events calendars and many valuable suggestions. Thanks are also due to Hajra Iftikhar, Collins Siu, Srinath Sinha and Christopher Loncke for providing research assistance. Finally, we would like to acknowledge Deep Local for building the ParkPGH application. This work was supported by a CMU initiative called Traffic 21 and the Benter Foundation.
REFERENCES 1. Smart Parking Linked to Transit: Lessons Learned from Field Test in San Francisco Bay Area of California. Shaheen, Susan and Kemmerer, Charlene. 2008, Transportation Research Record: Journal of the Transportation Research Board, Vol. 2063, pp. 73-80. 2. Best Space Scenario. Orski, K. 2003, Traffic Technology International, pp. 54-56. 3. Real-time parking information management to reduce search time, vehicle displacement and emissions. Caicedo, Felix. 4, 2010, Transportation Research Part D: Transport and Environmen, Vol. 15, pp. 228-234. 4. The use of space availability information in 'PARC' systems to reduce search times in parking facilities. Caicedo, Felix. 1, 2009, Transportation Research Part C: Emerging Technologies, Vol. 17, pp. 5668. 5. Predicting Parking Lot Occupancy in Vehicular Ad Hoc Networks. Caliskan, M., et al. April 2007. Vehicular Technology Conference, . VTC2007-Spring. IEEE 65th. pp. .277-281. 6. Parking difficulty and parking information system technologies and costs. Teng, H., Qi, Y. and Martinelli, D. R. 2008, Journal of Advanced Transportation, Vol. 42, pp. 151–178. 7. Modeling Effect of Travel Time Uncertainty and Traffic Information on Use of Park-and-Ride Facilities. Bos, Il, Ettema, D. and Molin, E. 2004, Transportation Research Record: Journal of the Transportation Research Board, Vol. 1898, pp. 37-44. 8. Parking Search Time and Information Identification for Off-Street Parking in New York City. Teng, H, Qi, Yi and Yi, Ping. 2002, Transportation Research Record: Journal of the Transportation Research Board, Vol. 1800, pp. 44-52.
14
