You are on page 1of 13

Project-based Optimization Approach for Intermodal Transportation Safety Funding Allocation

Jiangbo Yu Department of Civil & Environmental Engineering, University of California, Irvine jiangby@uci.edu (213) 422-2837

Kate Amissah Los Angeles County Metropolitan Transportation Authority amissahk88@gmail.com (213) 309-0223

ABSTRACT: The aging transportation infrastructure and insufficient funding allocation makes it urgent to improve efficiency of funding allocation. This paper proposed an integer programming optimization approach to minimize the potential fatality with constraints of budget and other technical limitations using Federal Transportation Safety Improvement Program as an example. A stochastic modeling methods along with concepts of location set of project and project set of location are proposed to facilitate the consideration of potential multimodal program and correlation among projects in terms of working schedule, construction and maintenance cost, fatality reduction, and the uncertainties of funding obtainment. This paper selects projects listed in HSIP Cycle 5 and CPUC Railroad Crossing Funding Programs as an example to illustrate the advantage of the formulation. A brief sensitivity discussion and suggestions are also given.
KEYWORDS: Transportation Safety, Funding Allocation, Project, Optimization, Intermodal,

Stochastic

INTRODUCTION AND BACKGROUND Safety, as the first priority for transportation engineers, planners, and decision makers, is still an important issue for the efficiency of transportation and society (NHTSA). Due to the scale of the transportation infrastructure system and conventional dependence on intuition, the projects selected for improving safety (as well as efficient) are far from being optimized (FHWA Safety Report, SA-13-019). Current transportation safety improvement funding programs are widely separated by modes, purpose, and hierarchical governments. The total budget for improvement infrastructure and facilities is much less than actually needs (Jack Short, et al., 2005, Schofer, et al., 2010) ,due to the United States economics and other policy and political issues. Therefore, it is critical to allocate available resources in a more efficient and effective manner. The positive side of this issue is that funding allocator and decision makers have gradually realized it and tried to optimize it in the range they can control. For example, the Federal Highway Administration (FHWA) used three years to edit a technical report document on systemic safety project selection tool. It has managed to, in a sense, improve the state and local safety improvement project selection process. The Moving Ahead for Progress in the 21st Century Act (MAP-21) continues the Federal-level Highway Safety Improvement Program (HSIP), which, though remains to be federal-aid (Obligation Rates for Highway Safety Improvement Program, Federal Highway Administration, 2013), has experienced deregulation to become more data-driven and flexible since the issue of SAFETY-LU (23 U.S.C. 148 and 402). Aiming at significant reduction in traffic fatalities and serious injuries on public roads, the program allows multiple funding resources and gives various example of the usage of the funding (such as pavement and bridge repair, geometric modification, signal timing and device improvement, education program, policy/law enforcement, and potential application of advanced transportation technologies. Other programs such as High Risk Rural Roads Program (HR3) and Federal Safety Routes to School Program (SRTS) complete the coverage of safety funding in terms of modes. In addition to Highway, Federal Railroad Administration and state agencies (such as California Public Utility Committee) investigates and offers safety funding for railroad safety improvement (such as highway-rail grade separation). Although the funding receivers have more autonomy than ever before, funding allocation from different government levels might not in the most efficient shape in terms of reducing the accident cost on all transportation modes. In additional to the intermodal issues mentioned above, six more difficulties determining project to be funded included but not limited to (1) projects uncertainties caused by the funding fullness. More than often, a project attempts to receive funding from various resources (such as federal government, state and local agencies, as well as public private partnership) while proposing it to the HSIP, and uncertainties merge if the committee has to decide if allocate the funding before the rest of funding project is applying is complete. (2) The construction duration for a project is usually a distribution rather than certain. (3) The possibility of inter-project corporation and conflict should be considered. For example, if two projects at a location is installing guardrail and barriers and the other retrofitting pavement, then usually the local DOT would combine the two to reduce construction cost. On the other hand, two projects could be technically conflicted in terms of schedule (and therefore influence the complete time and potentially costs more.) (4) A project could be proposed at numerous adjacent locations and therefore sometimes cannot be treated separately when predicting accident rate. (5) The funding allocation decision has to take into account political and policy factors. A funding equity policy could set maximum for each region. A proposed project from one city could affect the territory 2

of another city. Also, in some cases, a project is submitted by two local agents together (therefore the funding equity constraint might arise). (5) Project effectiveness estimation and accident prediction is still a highly discussed topic, and therefore, it is important to leave enough space for the model to accommodate various model and assumptions. (6) The unspent funding (carryover) could be beneficial to the next cycle and therefore should be considered in the objective functions. The objective function minimizes the fatality (or equivalent) using a linear integer programming approach proposed in this paper in an attempt to solve the issues mentioned above. The first part reviewed literatures with regard to accidents data collection and analysis (high-risk location), fatality modeling and prediction, and determination of engineering and planning counter measurements. The second part explains the variables, objective functions, and constrains for the proposed model in this paper. The third part demonstrates the model using a case extracted from HSIP and UCTC data.

LITERATURE REVIEW Although the methods varies for selecting projects by say Highway Safety Improvement Program, so far all they have been used can be categorized into benefit-cost ratio method (i.e., comparing all the benefits associated the a countermeasure to the cost of implementing the countermeasure. However, the Highway Safety Improvement Program Manual gives a vague description assuming that the effectiveness difference between two alternatives solving a safety keeps the same without considering the relative locations (Lord, D., et al., 2010), future demand change and so on. Golob T., et al. compared the characteristics (such as mean volume and variation) of traffic flow with different types of accidents on freeway with the aid of inductive loop detector data, and demonstrate a non-linear, non-monotonous relationship. In addition, the accident rate, type of accidents (Ma, J., et al, 2008) are also related to the composition of flow (Kahane, C., 2003), existing features of a location (e.g., Interactive Highway Safety Design Model, FHWA Research and Technology), pedestrian (Retting, R., 2003) and bicyclist, and even local economics (Noland, R., et al., 2004). Accident data available (such as the two national computer record systems are FARS and GES, and each state as well as many large municipalities and counties maintains a computerized accident record system) and well-recognized guidebooks (such as ITE Traffic Safety Toolbox and Safer Roads: A guide to Road Safety Engineering) can assist in identifying the hot spots, possible causes, and corresponding countermeasures (such as exposure control, accident prevention, behavior modification, policy, and so on). Currently, different state and local agencies choose different categorization method for location/project. For example, the St. Louis County, MN, classified as S (segment), I (intersection), and HC (horizontal). These countermeasures or improvements can be proposed to corresponding safety improvement programs (usually need to be permitted by State Transportation Improvement Program, STIP, first). Traffic crash analysis to identify high crash locations are performed using methods such as frequency, crash-rate, rate quality-control. Highway Safety Improvement Program Users Manuel also presents standards and countermeasures. Some typical techniques for more efficient safety funding allocation have been proposed by Brown et al. in 1980 (used dynamic programming concept) and Farid et al. in 1994 (considered incremental benefit cost analysis). In 1985, Sinha et al. used the expected number of accident as 3

the measure of effectiveness in their binary integer model. Besides, the method proposed also considered the possible carry-over of unspent funds from previous fiscal year. Uncertainties and possible countermeasures with regard to estimating model parameters are also proposed. Pal and Sinha in 1998 continued the previous work by considering the changing of effectiveness for each projects selected and by allowing for flow variations in the future. In an attempt to simplify the problem by classify projects as geographically discrete and continuous and by, assuming, to a degree, homogeneity of the highway segment, a mixed integer knapsack LP model was introduced (Mechachrinoudis, E., Kozanids, G., 2002). Some more practical and simplified analysis conducted (Kar, et al., 2004) using concept of Safety Improvement Index (SPI) and priority ranking, though much more assumptions (such as project independence) are made. In this study, a relatively comprehensive statewide traffic-crash analysis was performed with crash and exposure data. Pal R. and Sinha K. in 2007 developed a procedure with the object of minimizing the total number of crashes (in a narrow sense) during a study time and improved the flexibility by considering the effectiveness of the improvements in the future years.

MODEL FORMULATION The model proposed in this paper applies linear integer programming techniques to minimize the fatality (or the equivalent) in the study area over analysis time period, in subject to budget and integer constraints. The notations used in the program are as below: , location set for project n, it stands for all locations that would be impacted by project n. For incidence, indicates that the project #3 would have impact on location #1, #6, and #7. , proposed project set for all projects that could be implemented or has impact on location. We can obtain it through the Project Sets in the first step using computer if sets are plenty. For incidence indicates that location #5 are targeted by projects 2, 5, and 8. One location could be a link (such as highway, railroad, airport runway, Positive Train Control System (all track likes and at-grade crossing and turnouts nodes should be considered, etc.) or a node (such as bus and subway station, toll station, at grade crossing etc.) in the network systems. , the set of all possible combination of project(s) that are proposed to implement at location l. For example, if project 5 and 18 are proposed on location 3, then = {00, 01, 10, 11}. , the set of the all possible project combination in which the corresponding position of the project N is 1 (rather than 0). In the above case, = {10, 11} and = {01, 11}. , binary variable, it represents whether at location l, a certain combination of projects would be implemented or not. () stands for a binary full permutation in the project set of location L. For example, given that , . As we can see, in this set, only one of the elements can and have to be 1, the others are constrained to be 0s. Fortunately, more than usual, one location has zero or one proposed project and therefore the amount of variables is very unlikely to increase exponentially. To obtain the binary variable of whether project is selected or not, simply

add all

at location l where project N is proposed on or has effect on. If this

summation is 1, then N is implemented. If 0, not. , the expectation of the predicted (equivalent) fatality (using crash modification factors, CMF) in the do nothing alternative at location l during time t. As mentioned in the literature review, there has been numbers of prediction model for this parameters. Common variables include but not limited to historical data at the location or similar (relative) location, traffic flow, geometric design, existing condition of the facilities, and social economic conditions. To simplify the process in the sensitivity analysis, a SpatialTime Poisson Distribution can be assumed. i.e., . The parameter can be the function of factors such as traffic flow patterns and facility conditions at the time predicting. The term A(S) is for cases where a location can be assume homogenous and the accidents can be estimated by multiplying the length or area of it. Based on Poisson distribution feature, the expected fatality at year y would (y), and the 95% confident interval can also be computed. Noted that should considered the accumulative effect for projects proved yet havent been implemented (so cannot be considered using real field survey or statistic data). , the predicted effectiveness at time y if the proposed project (combination) at location l was completed in time t. The estimation of EF can share the same model used for or simply base it on similar project in the past. If both data lacks, a logistic function, , is recommended. M here is used as a function shift to consider the delay from the approval to construction or to work finish. For example, a project at location j is approve in 2016 and the project needs 1 year to be fully funded (from multiple sources) and 0.5 year to construction, then M = 1.5. If the construction phase has high risk (in terms of fatality), a piecewise function can be proposed so that the potential fatality during the construction can be considered. For example, assuming that the effectiveness of a project is independent of the future traffic flow and also approximately logistic curving shape, for event that a project (or project combination) at a certain location is complete by time t=4 and t=6, the EF can be ensemble the figures shown below.

Figure 1. Example of EF for certain project (project combination) at location l. This logistic form can be more detailed and precise if the Idea direct- or cross- effect of a safety improvement project on a certain location is considered. This concept is based on a fact that accidents could be due to different reasons and an improvement could be 5

more effective to reduce certain type(s) of accidents. If there are mainly 3 reasons in total (say alcohol (DUI), inadequate/missed civil facilities, or signal), and we know that at location l, 80% is caused by DUI, 15% civil facilities, 5% signals, then two current available proposed signal timing improvement projects might have less reduction than expected. The FHWA systemic safety project selection tool also recommends this concepts and demonstrate it in the way of Crash Tree Diagram. , the cumulative probability of project or project combination completed at location l by time t. It can also be written as Prl(T<=t). Treated the time analyzed as discrete (separated by cycle or fiscal year), Prl(t=T)=Fl(t )Fl(t-1)=Prl(T<=t)- Prl(T<=t-1). For instance, if N5={l3,l4} (e.g., signal retiming and pedestrian exclusive phase addition), N19={l3, l7} (e.g., speed limit and bike route signage repair and update) is proposed at l3 and l7. L3={N3, N19}. If assume (based on convention or pass similar projects) project N19 job time is beta distribution with parameter and (>0, >0), project 19 is uniform distribution from a to b (a>0, b>a), >b, <a and N5 and N19 are independent, then

, in which { {

The cumulative distribution needs not necessarily to obtain for the analysis. Sometimes, for preliminary purpose, only expectation and variance can be used for the formulation. Pfl, the probability of the project or project combination will be able to receive enough funding to proceed. According to 23 USC 148 Sec (d) 2B and Sec (e), Usually, a project applying for funding of HSIP can in the same time apply for other resources. Assuming a project applying for 3 funding programs, and the joint distribution of the three random variables (M1, M2, M3) is PrM1,M2,M3(m1, m2, m3), with covariance matrix 3X3, and the project can only start with more than M, the probability becomes Pr(M1+M2+M3>=M)=1-Pr(M1+M2+M3<M). T0N=a random variable indicating the time point completing the project N. E(T0N) is the expectation of T0N. MN, regular maintenance cost during time period [t-1,t] by corresponding agency. MNp, regular maintenance cost in time period [t-1,t] by the program.

ON, regular operation cost during time period [t-1,t] by corresponding operator. ONp, regular operation cost during time period [t-1,t] by the program. SLN, service life time of the project N. Here we assume that for multi-location project, the service life time is the weighted expectation of all locations. SN, salvage value of the project N if the project (after funded) belongs to corresponding jurisdiction or entities. SNp, salvage value of the project N if the project uses States rightof-way (ROW). EN, the set of entities/agencies (players) involved. For example, the safety improvement program, state/local governments. , the responsible portion of the cost for project(s) settled by the agency/entity en. When en=p, it is the cost settled by the safety improvement program. When=I, the cost settled by the state or local government the location is in. Ben, budget available for any entities in the set EN. The budget usually is composed of newly received project for current fiscal year and the carryover from last one. Ben is the equivalent present worth. The usage of this budget constrains also the case in which a funding applicant submit more project proposal than they can financially hand if all of them are permitted. , the maximum funding that can be allocated to the region. This is for the funding equity constraints. R is the set of all the feasible combinations of locations and projects in the region R. The objective function is (1) In which, EFT is the (equivalent) fatality and can be obtained by { [( ) ( )]} (2)

is the estimated dual price (potential fatality reduction) if increase 1 unit of funding. This can be obtained simply from the past allocation experience. And SF is the surplus funding for the current fiscal year. (3) If not consider the potential collaboration among projects (for constructions, maintenance, and operation), the total cost C can be obtained through: ( ) (4)

The first term in the objective function, (1), is essentially formulated based on probability multiplication theorem and the definition of the expectation of discrete variable, E(A)=E(A|BN)P(BN), where B1, B2,, BN,, Bn are collectively exhaustive and mutually exclusive. Noted that the EF has subscripts of location, l, project set at this location, , the time the project(s) start to be effective, t, and the time after the project to be effective, y. Another advantage is that EF term can include the potential increase due to construction and so on by simply setting EF <0. EF (t<=T0) and EF (t>SL) can be set at 0. For instance, the (1-EF) term 7

of a certain combination of projects, , completed in 2 years (here assuming F(2)-F(1)=1) with one year construction, at location l, over seven years, then it becomes formulated as 5.4(10X)+5.6(1+0.09X)+5.7(1-0.3X)+5.8(1-0.29X)+5.9(1-0.28X)+6.0(1-0.27X)+6.1(1-0.29X). The second term of the objective function is for considering the potential for the leftover that could have better use if save it for some better or more cost-efficient projects in the future. The conversion parameter (for converting monetary value to fatality equivalent value, the estimation of which can be based the historical average utility of the each monetary unit saving life or equivalent life (say if saving 10 slight injuries is equivalent to 1 death and $1.0M accidents cost saving due to the program can be treated as one human life and 5 life or dears can be treated as one life). Behind the conversion parameter, is essentially the sum of the net present worth of all project (belongs to federal or state government) using economic evaluation methods. In the equation (4), is the cardinality of the set safety improvement program) of the project N. The program budget constraint ( ) (5) and is the funded cost (by the federal

The funding receiver budget constraints below are for case that funding applicants submit more project proposals than they can financially maintain. Please noted that an improvement could cover or affect more than one cities or counties. { [
[ ( ) ]

]}

(6)

Funding Equity/Policy Constraints ( ) (7)

Mutually exclusive constraints at location l Consistency constraints for project N

(8)

(9)

Binary Integer Constraints (10) (11)

ABOUT SENSITIVITY (POST-OPTIMALITY) ANALYSIS Although the model considers the stochastic nature of project funding fullness as well as construction delays, the uncertainty of the coefficients in the objective functions and constraints such as potential effectiveness of improvement or improvement combination, accident rate, probability distribution, budget increasing, and so on, should also be analyzed in the post8

optimality phase. If, for example, in the sensitivity analysis, the upper limit (for basis change) of a certain variables coefficient is lower than the, say, 95% percentile, we can increase sample space, or try to adjust the basis so that the analysis can provide more useful information to the decision makers. CASE STUDY The case assumes 5 proposed projects and 7 locations affected. The data refers to the HSIP from the California Department of Transportation and the Railroad crossing funding list from the California Public Utility Committee. Table 1. Projects, locations, integer constraints, and project data
N1 L1 L2 L3 L4 L5 L6 L7 Project Constraints C, $ (No considering collaboratio n) M+O, $ SL, yrs Pf 0/1 0/1 0/1 0/1 No constraint 0.50M X1,1+X4,1= 0 or 2 2.25M X2,01+X2,11+X5,1= 0 or 2 3.65M 0/1 X3,1+X7,1=0 or 2 3.25M No constraint 1.75M N2 0/1 N3 0/1 0/1 N4 N5 Constrains X1,0+X1,1=1 X2,00 +X2,01 +X2,10 +X2,11 =1 X3,0+X3,1=1 X4,0+X4,1=1 X5,0+X5,1=1 X6,0+X6,1=1 X7,0+X7,1=1 -8.0M

5,000 15 0.95 --

10,000 20 0.70 --

7,000 20 0.95 --

9,000 15 1.00 --

8,500 10 0.90 --

---0.00035

Table 2. The project effectiveness given the time the project (combination) complete (the time t is the inspection cycle which is usually twice every fiscal year)
EF at Li if project complete at time of corresponding E(T0) N1 N2 N3 N4 N5 t=1 t=2 t=3 t=4 t=5 t=6 t=7 t=8 t=9 t=10

-0.05 -0.10 0 -0.05 -0.05 0 0 -0.01

-0.09 -0.10 -0.05 -0.13 -0.09 -0.05 -0.04 0.25

0.30 0.35 0.29 0.37 0.25 -0.05 -0.05 0.25

0.29 0.35 0.29 0.36 0.24 0.21 0.21 0.24

0.28 0.34 0.29 0.34 0.29 0.21 0.20 0.24

0.27 0.33 0.28 0.33 0.27 0.21 0.20 0.23

0.25 0.31 0.27 0.31 0.25 0.18 0.19 0.23

0.22 0.29 0.27 0.29 0.23 0.18 0.19 0.23

0.19 0.28 0.27 0.28 0.21 0.16 0.18 0.23

0.18 0.27 0.27 0.27 0.18 0.16 0.18 0.23

Table 3. Predicated fatality rate in do nothing alternative


L1 L2 L3 L4 L5 t=1 7.11 2.79 5.17 2.71 6.14 t=2 7.12 2.85 5.52 2.82 6.16 t=3 7.14 2.89 5.82 2.85 6.18 t=4 7.17 2.94 6.25 2.98 6.20 t=5 7.18 2.97 6.33 2.93 6.22 t=6 7.19 3.11 6.54 3.12 6.25 t=7 7.20 3.20 6.62 3.19 6.27 t=8 7.22 3.26 6.73 3.25 6.30 t=9 7.23 3.32 6.79 3.32 6.33 t=10 7.25 3.40 6.87 3.39 6.36

L6 L7

8.21 4.37

8.52 4.52

8.64 4.82

5.66 4.25

5.68 4.34

5.69 4.53

5.71 4.62

5.72 4.72

5.73 4.77

5.75 4.81

Table 4. Accumulative probability distribution for each project (combination) at location l


F L1 L2 L3 L4 L5 L6 L7 t=1 0 0.1 0 0 0.5 0 0 t=2 0.6 0.7 0.5 0.4 0.9 0.2 0.3 t=3 0.9 1.0 0.7 0.8 1.0 0.5 0.7 t=4 1.0 1.0 0.9 1.0 1.0 0.8 0.9 t=5 1.0 1.0 1.0 1.0 1.0 1.0 1.0 t=6 1.0 1.0 1.0 1.0 1.0 1.0 1.0 t=7 1.0 1.0 1.0 1.0 1.0 1.0 1.0 t=8 1.0 1.0 1.0 1.0 1.0 1.0 1.0 t=9 1.0 1.0 1.0 1.0 1.0 1.0 1.0 t=10 1.0 1.0 1.0 1.0 1.0 1.0 1.0

Putting the data above into LINGO, N1, N2, N4 are selected. The reason that N5 is not selected (though enough funding) is that it has lower marginal fatality reduction effectiveness (than carry it over to the next budget). Also, the total reduction is 11% more than selecting the most seemingly effective four projects: N1, N2, N3, N5. This is mainly due to difficulties of considering such dynamic process and of the project overlapping effect. The study time duration can be varied to see if significant change of result happens. What needs to be noted is that this is just for assisting the final decision. Since the parameters and coefficients in the model are mostly based on historical data. It is possible to compare it with the sensitivity analysis result. For example, if the 90% confident interval of coefficient at time t at certain location is between [7.5, 9.6] with mean of 8.5 and if the corresponding upper limit is 9.0 (for changing the optimal solution), it might need further discussion (such as acquiring more data or using other qualitative or quantitative comparison). A more detailed and systematic sensitivity analysis will be left for future research.

CONCLUSIONS The paper developed an integer LP model for optimizing transportation safety funding allocation. The uniqueness of this paper is that (1) it includes the stochastic nature of project funding and construction, (2) it can be applied to a more general intermodal funding allocation process. What also is improved is the potential interaction and possible construction and maintenance collaboration among projects. Several directions of future work can be done to improve the model in this paper, such as the improvement effectiveness factor and parameter estimations, stochastic modeling of multi-source project funding application and allocation, some particular characteristics of certain type of projects (such as seaport improvement) that might not fit very well into the model proposed in the paper. A more comprehensive study on relationship between freight and passenger safety will also be beneficial.

REFERENCE
Short, J., Kopp, A., 2005, Transportation Infrastructure: Investment and Planning. Policy and Research Aspects, Transportation Policy, 12: p360-367

10

Schofer, J., Evans, L., et al., 2010, Research Agenda for Transportation Infrastructure Preservation and Renewal: Conference Report, J. Infrastructure. Syst. 16: p228-230 FHWA Safety Program, 2013, Obligation Rates for the Highway Safety Improvement Program , Federal Highway Administration, http://safety.fhwa.dot.gov/hsip/gen_info/slorhsip/slorhsip20131001.pdf FHWA Safety Program, Dec. 2013, Planning: Project Periodization, Highway Safety Improvement Program Manual, http://safety.fhwa.dot.gov/hsip/resources/fhwasa09029/sec4.cfm FHWA Office of Safety, Jul. 2013, Systemic Safety Project Selection Tool, Report No. FHWA-SA-13-019 Lord, D., Mannering, F., 2010, The Statistical Analysis of Crash-frequency Data: A review and assessment of methodological alternatives, Transportation Research Part A, 44, p291-305 Golob, T., Recker, W., Alvarez, V., 2004, Freeway Safety as a Function of Traffic Flow, Accident Analysis & Prevention, Volume 36, Issue 6, p933-946 Kahane, C., 2003, Vehicle Weight, Fatality Risk and Crash Compatibility of Model Year 1991-99 Passenger Cars and Light Trucks, NHTSA Technical Report, National Highway Traffic Safety Administration, DOT HS 809 622 Ma, J., Kockelman, K., Damien, P., 2008, A Multivariate Poisson-lognormal Regression Model for Prediction of Crash Counts by Severity, Using Bayesian Methods, Accident Analysis and Prevention, 40, p964-975 Retting, R., Ferguson, S., McCartt, A., 2003, A Review of Evidence-Based Traffic Engineering Measures Designed to Reduced Pedestrian-Motor Vehicle Crashes, American Journal of Public Health, Vol. 93, No. 9 Noland, R., Oh, L., 2004, The Effect of Infrastructure and Demographic Change on Traffic-Related Fatalities and Crashes: A Case Study of Illinois County-level Data, Accident Analysis and Prevention, 36, p525-532 Pande, A., Abdel-Aty, M., 2007, Crash Data Analysis: Collective vs. Individual Crash Level Approach , Journal of Safety Research, 38, p581-587 Institute of Transportation Engineers, 1999, Traffic Safety Toolbox, ISBN: 0-935403-43-4 Ogden, K.W., 1995, Safer Roads: A Guide to Road Safety Engineering, ISBN-13: 978-0291398291 Wahs, M., 2006, Improving Efficiency and Equity in Transportation Finance, http://escholarship.org/uc/item/3sg2q6n2#page-1 Brown, D.B., 1980, Use of Dynamic Programming in Optimally Allocating Funds for Highway Safety Improvements, Transportation Planning and Technology 6, p131-138 Farid, F., Johnston, D.W., Laverde, M.A., Chen, C.-J., 1994, Application of Incremental Benefit-cost Analysis for Optimal Budget Allocation to Maintenance, Rehabilitation and Replacement of Bridges, Transportation Research Record 1442, p88-100 Pal., R., Sinha, K.C., 1998, Optimization Approach to Highway Safety Improvement Programming , Transportation Research Record 1640, p1-9 Melachrinoudis, E., Kozanidis, G., 2002, A Mixed Integer Knapsack Model for Allocating Funds to Highway Safety Improvements, Transportation Research Part A 36 p789-803 Kar, K., Datta, T., 2004, Development of a Safety Resrouce-Allocating Model in Michigan, Transportation Research Record: Journal of the Transportation Research Board, No. 1865 TRB, National Research Council, Washington, D.C., 2004, p64-71

11

Pal, R., Sinha, K., 2007, Optimization Approach to Highway Safety Improvement Programming , Transportation Research Record 1640, Paper No. 98-1114

12

You might also like