The Use of Optimization Techniques in Parking Facility Siting at Stanford University I BY DENNIS G PERKINSON S tanford University’s extensive acreage affords a luxury rare to most university campuses — parking for all faculty, staff, students, and visitors who wish to drive to the campus. As in any community, however, parking at Stanford is a complex and sometimes volatile issue. Parking is essentially an infrastructure element, serving the needs of academic and residential facilities. Like any infrastructure system, parking must be responsive to facility plans, and parking concerns are sometimes dwarfed by more important considerations. Parking, however, differs from most infrastructure systems, such as underground utilities, in one important respect: The decisions that are made about parking are perceptible to all members of the community. The supply, location, design, and operation of parking facilities affect the convenience of daily users, the safety of drivers and pedestrians, and the ambiance of the campus environment. Regional planning, a relatively new approach to planning at Stanford, provides a context in which parking can be coherently planned (Figure 1). Through the regional planning process, long-range parking issues can be addressed in their institutional perspective – serving academic needs while also attempting to strike a balance among a variety of potentially conflicting concerns. Until the 1960s, surface parking lots had been interspersed among academic buildings throughout the campus. Since then, this pattern has changed. Interior spaces in the central campus have been converted to pedestrian malls and courts, sometimes displacing parking areas. Simultaneously, the need for academic facilities has grown and building sites have become scarce. As a result, many of the new buildings are located on former parking lots. The resulting shortage of land that can be used for parking in and near the central campus has led the university to plan multilevel parking structures that will maintain or increase parking supply while minimizing site coverage. Parking is essentially an infrastructure element, serving academic and residential facilities. As a result of the new constraints and costs, parking questions have recently become increasingly complicated and critical. Parking planning has thus become more sophisticated and more formalized. Whenever possible, parking plans are now developed in conjunction with regional plans; formal funding sources for parking facilities have been adopted, and greater expertise has been brought in-house to enhance the university’s planning and management capabilities in these areas. Model Methodology Data and The university’s Office of Transportation Programs has developed a computer-based optimization model to examine the distribution of current and projected parking spaces on campus and to determine the level of service for each planning region. Because the program goal is to meet the complex needs of employees, commuting students, and visitors, the model focuses on those users. Other parking categories (e.g., handicapped, student resident, and service vehicle) represent local needs and are addressed based on local demand and actual use. The Stanford parking model assesses the adequacy of existing and proposed parking supply by analyzing the parking supply in each planning subregion, the employee population within the subregion, and the walking distances between parking lots and buildings across subregional and regional boundaries. The model improves regional planning by allowing planners to examine hypothetical future scenarios (e.g., the effects of parking demand that might result from regional population changes or the parking impacts of a proposal to site a building or parking facility at a specific location). Technically, the model is a general linear optimization model that takes advantage of certain characteristics of the “transportation problem.” This problem consists of m origins (with origin i possessing a, items), n destinations (with destination j requiring b, items), and the sum of the a, values equal to the sum of the b, values. The costs associated with moving a single item from any origin to any destination are given. The objective of the model is to empty the origins and fill the destinations in such a way that the total cost is minimized. The problem ITE JOURNAL . APRIL 1989029 can be stated and solved as a linear programming problem whose variables are the mn individual allocations, with the summed cost as the objective function and with rn+n-1 independent constraints. A program to establish the feasibility of conceptualizing campus parking in this way was developed by a team of graduate students as a project for the course Operations Research 280 (under the guidance of the Office of Transportation Programs). The Office of Transportation Programs then reformulated the program to better reflect real world parking problems, as well as to make it more capable of analyzing various scenarios. In the application, the origins are analogous to supply (parking spaces in 25 supply regions), the destinations are analogous to demand (people wanting to park as near as possible to one of the 25 population subregions), and cost is analogous to distance walked to a parker’s destination. No monetary value is assigned to distance walked or to walking time. The Stanford parking pricing and allocation policy is the same for faculty, staff, and commuting students. Thus, the assigned value of time is the same for all parkers in the aggregate and is not included in the model. The benefits of this formulation, aside from the computational benefits already noted, are conceptual clarity and utility of the output. The solution algorithms also produce whole-unit answers. Thus, we do not have to deal with fractional parkers or parking spaces. Specifically, the model provides measures of the marginal cost (in distance walked), of changes in supply in any subregion and the Al .4 FIGURE 1. Figure 1. Map of Stanford University regions. 30. ITE JOURNAL . APRIL 1989 marginal cost (in distance walked), and of changes in demand in any subregion of campus. The model assumes that parking behavior in the aggregate is rational and the dominant goal is to minimize the total distance walked. Individuals, however, attempt to optimize their own situation, rather than the campus-wide situation. Two constraints are sufficient to model these undesirable conditions in the a~regate. For regions where parking demand is less than or equal to 100 spaces, all of that demand is assigned to that region, up to the available supply in the region. For regions where parking demand is greater than 100 spaces, 50% of that demand is assigned to the region, up to the available supply in the region. No adjustments are made for regional variation in the mix of permit type or for the proportion of the population of a region (faculty, staff, students, and visitors) requiring parking. Campuswide averages are assumed in both cases. The mix of permit type is based on actual sales. The ratio of parking demand to total population is based on the observed ratio of actual parking use to population. The parking supply of each region is taken from a university parking inventory created and maintained for this purpose. Data on use are based on the surveys conducted by the university’s Public Safety Department and the Office of Transportation Programs. Subregional population figures are taken from Stanford University Facilities and Engineering Database (SUFED), maintained by Facilities Project Management. The output of the model is a universitywide distribution of parkers under current conditions and for any specified future scenario, which includes potential changes in population and parking supply. This formulation provides a campuswide perspective in that it allows the planners to minimize the total walking distance of all campus parkers and to rank subregions by the average walking distance. The ratio of parking demand to total population is based on the observed ratio of actual parking use tO pOpUldiOflm Stanford’s parking program, based on this procedure, helps to prioritize future parking needs and benefits by subregion. It can also determine the number of spaces needed to meet those needs, while identifying the most efficient parking configuration from the perspective of the entire campus. Where other land uses conflict with preferred parking locations, the cost of these conflicts in extra walking distance is available to decision makers. An Example Following the development of the optimization model, Stanford University began construction of an administrative office complex on an existing parking lot. This facility, located in the Serra Triangle region of campus (see Figure 1), displaced 390 parking spaces during construction; however, the project resulted in a net increase in the number of parking spaces in the region and throughout the campus. Our surveys showed that the users of this parking area had destinations in several other regions. The impact of this facility, during construction and at completion, was forecast using the techniques described in this report. The case during construction was relatively because several simple underutilized lots were available near the construction site. The model allowed us to identify the “best” alternate sites (least walking distance) for a given office location for those displaced by construction. Region- specitic notices, naming specific alternative parking locations, were prepared and distributed. This was one of the university’s least disruptive major construction projects. Of more interest is the use of optimization to predict parking behavior upon completion of the administrative office complex. The region-by-region cost matrix for the entire central campus, listing walking distances in hundreds of feet, is shown in Figure 2. The matrix is not symmetrical because parking supply centroids and population centroids are often at different points within a region. The row and column of interest in this example is labeled SER for Serra Triangle region. Table 1 is a portion of the allocation matrix STAFF,FACIJLTY AIIDVISITDAPARKING BY REGIcIN COSTMATRIX (IN HUNDREDS OF FEET TO DESTINATION) AA PO Occ AVU II LA AVVCC ML0121z:0 R E G 1 8 NNNJ UWUOMMEEE RCC PALM ARB OVAL CIV1 NuC1 NUC2 WC3 JORD sLmcl 3LMC5 NEC1 NEC2 NEC3 NEC4 FRO ATHE1 ATHE2 ATw3 BON RESU RESE COilL $ 10 20 19 30 26 28 40 36 9 21 16 ;: 2! 12 1 6 22 15 12 18 13 221181585 12 8 12 23 18 2622222620914161 18 19 24 32 28 8162131261919919 22 13 10 14 14 30 22 14 10 12 15 10 12 20 20 25 17 16 20 20 12 13 19 28 26 18 16 18 24 23 22 23 29 36 36 3021151510891813 35 33 37 40 42 38 30 24 23 20 40 34 28 26 30 44 32 26 21 25 z524242#a9%3tt Uc 32 35 27 32 26 32 32 36 IJUNNNNTT 22 30 15 18 1 10 22 12 30 22 11 8 15 23 5109 1 10 10 1 16 18 10 12 22 24 24 28 29 30 41 18 21 20 27 18 18 23 25 25 22 24 25 36 34 30 34 31 36 36 37 68 48 44 15$18 41 36 38 32 42 25 41 39 26 36 22 26 14 14 56 13 49 46 3043 26 32 15 20 19 24 19 24 6 12 RRc EEOSS Ssui?vw UEL U2C EFHHB CREEO CCC 151 234012N 18 8 22 26 16 28 19 40 13 32 60 14 16 48 22 18 48 18 10 38 21 12 48 30 1 1 33 1 1 27 33501108 39 34 10 29228181 37307138 31 24 !5 35 28 10 43 36 23 24 24 27 50 45 28 27 31 33 52 47 20 49 45 18 +f % 15 13 20 36 16 26 42 30 38 22 10 24 20 27 34 39 34 12 20 10 20 24 1.S 18 31 29 22 2512182235384044 233235 26 ?4 22 21 32443644 364048 17 13 1623333034 32 & 2732 2028243640232621 @32 36 28293047 L8J5413S3b 1520 2325243644 183632 301924 25 22 25 3442 25 41 39 %19 24 363637485613494643612 3731 35 43502752 49 kf 13 16 3024 283645 37 4745 %20 26 7 15 1023 2833 20 78 f% 3642 17 12 203944 1 18 13 24 18303332 8 761825362555 923642 112 51621401718 T4246 704044 24 7 12 1 71322422930 5 7 77319422123 64267 18 6 30 18 ~6 13 13 1 12 5423 31 %52 52 103634 $? 1920 22 26 2932326350 1602030 $6164 33 25 21221912 146434$1624 32 36 4042425460 1 6 t85550 17 25 172921232046 6 1 2%5256 12 25 183023313043 77*6W $4fJt$z9 7M% m ?? ~34 39 36 4240425261165552 44 42 46444750642450565441 .. Figure 2. Cost matrix: staff, faculty, and visitor parking by region (in hundreds of !eet to destination). Table 1. Base Case Allocation Matrix (Partial) supply (No. Lot Use Lot Utl – – – – -lo -12 -13 -5 – – 1.00 1.00 1.00 1.00 Dual Spaces) ATHEI NECI NEC3 NE(2I — — 90 251 20 861 187 1400 86 0 0 101 560 2000 4 11 1 219 290 1300 0 240 0 0 1200 0 0 19 541 ● ● ● ● ● ● ● ☛☛☛☛ ● ● ● ● 9 ● ● 999** ● ● ● ● 9 9 ● ● **** ● Demand Dual (ft] ATHEI NEC3 NEC4 SER TOTAL 560 ● 00 ** . . . 9** ● ● ● ** ** 90 251 20 861 8,623 10,307 ITE JOURNAL oAPRIL 1989.31 that shows the distribution of parking origins and building destinations before construction. This is part of the output of the base case optimization model. The row labeled demand represents the demand for parking within each region. This demand is estimated from parking origin and destination surveys conducted periodically by Stanford’s Office of Transportation Programs. The row labeled dual represents the improvement (reduction in total walking distance) wrought by adding a parking space in the region represented by that column. The remaining rows represent the various parking regions. The first column represents each region. The second column gives parking supply (in number of spaces); a 5 % vacancy factor is included in that number. Actual supply is 5 % higher than shown, because in terms of parker behavior and perception, a lot is functionally full when 95% of the spaces are occupied. The next group of columns represent the various demand or destination regions. The last three columns are the change in the solution for adding a unit of demand to the corresponding parking region row (dual), lot use in spaces occupied (lot use), and lot utilization as a percentage of spaces filled (lot utl). The cells of the matrix represent numbers of parkers. Note that because the Serra Triangle Region (SER) had no buildings before construction and was therefore not a destination from a parking lot, it is not represented by a column. We can see from Table 1 that before construction, the Serra parking region was fully utilized by three destination regions: Near East Campus 1 (NEC1), NEC4, and Athletics East 1 (ATHE1). Table 2 predicts the situation in these selected regions after completion of Table 2. Post Construction Aiiocation construction and occupancy. The Serra (SER) region is now a demand region with destinations, as well as a supply region, as shown by the addition of the column labeled SER. The parking capacity of the region has also increased from 861 to 915, as shown by the row labeled SER. The region is now fully utilized by NEC1, NEC4, and itself. The increase in the dual column for NEC4 shows that the benefit of an added space in NEC4 has increased considerably. Comparison of the values in the dual row shows that, of the regions shown, additional demand in the Serra region carries the least negative impact, 700 ft of additional total walking distance, as opposed to 1,200 to 2,100 ft. Of the regions shown, additional growth in Serra would have the least impact on parking campuswide. Changes in demand caused by the relocation of personnel to the new administrative office complex are also reflected in the corresponding columns. This analysis allowed the Office of Transportation Programs to anticipate the impact of this new administrative office complex on campus parking during construction, and more importantly, it allowed us to predict the combined impact of this project and other speculative future projects. Finally, the anafysis confirmed that the SER region more than adjacent regions, could support additional growth in terms of impact on campuswide parking. This method of forecasting parking behavior has successfully predicted the aggregate behavior of parkers for two major changes in parking supply on the Stanford campus: the opening of a major parking structure of Matrix NEC4 (Partiai) Dual SER ATHEI NECI NEC3 Demand Dual (ft) ATHEI NEC3 NEC4 – – 55 251 20 20 1200 20 0 0 530 2100 35 67 1 200 500 3t)t) ● *O 1400 0 184 0 915 0 134 0 700 0 0 t) 300 ● SER 1300 0 0 19 481 ● ● ● ● ● ● ● 9 ● . ● ● ● ● ● ● ● ● ● ● ● ● *O 00 ..* ● 0. OO ● *** ● *** 9*** ● Lot Use 32. ITE JOURNAL – – – 55 251 – 1.00 1.00 -55 20 1.00 .6 915 _ 1.00 ● ● ● ● ● ● 8,766 . APRIL 1989 Lot Utl – -11 -13 _ 10,419 Acknowledgments The author wishes to thank Professor Alan Manne, of Stanford University’s Operations Research Department, and the six students in his OR 280 class of 1985-1986 who helped establish the feasibility of this approach to university parking. Bibliography Hillier, Frederick S., and Gerald J. Lieberman. Introduction to Operations Research, 3rd ed. Oakland, CA: Holden-Day, 1980. IBM. An Introduction to Linear Programming. White Plains, NY International Business Machines Corporation, t964. Marine, Alan S. GAMS/MINOS: Three Eramples. Stanford, CA: StanfordUniversity, December 1986. SAS Institute, Inc. SA.!VOR User’s Guide: 1983 Edition. Cary, NC: SASInstitute, Inc. 1983.■ Summary SupPly TOTAL over 1,000 spaces and the displacement of almost 400 spaces because of the construction of an administrative office complex. This approach now forms the basis of Stanford University’s campuswide parking program that will guide the planning for all future parking facilities at the university. As with most quantitative tools, calibration and adjustment are ongoing. Model results are not used blindly, but help guide the planners’ thinking about the problem. Lknrds G. Avkin.son is currently director of planning for VIA Metropolitan Transit in San Antonio, Texas. At the time this article was written, he was managing director of transportion programs at Stanford University. In that capacity, he was responsible for Stanford’s campus transit system, which carries over 420,000 riders per year, the 15,0G9space parking program, and the university’s campuswide commute alternatives effort. Perkinson holds undergraduate and graduate degrees in sociology fi-om the University of South Florida and an M. S. in transportation from Northwestern University He is an Associate Member of ITE.