Bi-Level Optimal Sequence and Schedule for Multiple Workzone Projects using WISE and Traffic Simulation By Sabya Mishra, Khademul Haque, Mihalis Golias, and Tao Ma(University of Memphis) Brad Freeze (Tennessee Department of Transportation) Presentation at 2017 planning applications conference, Raleigh, May 16, 2017

Overview Motivation Goals and Tasks Rationale Pilot Project Experience Challenges and Limitations Initial Recommendations

Motivation to WISE (1) Significant projects which are anticipated to cause sustained work zone impacts Optimal multiple project coordination helps to reduce work zone related crashes Obtain three pillars of benefit Social Economic Environmental

Motivation to WISE (2) Need for coordinating work zone (WZ) projects Type: construction, maintenance, utility, etc. Pass through state EDC-3 Smarter Work Zones as a initiative to help stimulate and support the improvement of work zone planning. In TN Approximately 500 WZ/ year on interstate/state routes Highest type of WZ are construction Include a map…..of TN

Goals and Tasks Goals of the pilot project Assessment of WISE to identify limitations Recommendations to improve WISE Tasks of pilot project include Collect network and WZ data, Build Planning (sequencing) Operational (mesoscopic simulation for traffic diversion) Model calibration and validation Use WISE to optimize sequence of WZ projects WZ data includes data such as start and end date, spatial location, hours of operation, work zone type

Rationale Work Zone Sequencing 1. Decision Maker Objective: e.g. Minimize Cost 2. Users Objective: Minimum cost path choice 3. Input: # of WZs, and their characteristics, Budget and other constraints 4. Input: Network, Origin-Destination, and Traffic Control 6. Output: Link flow, path flow, travel time, queuing 5. Output: Sequence of WZ activities Planning Operations

Pilot Project – Shelby County, Memphis, Tennessee Shelby county: Located in Memphis Metropolitan Area SOURCE: Memphis Metropolitan Planning Organization (Interactive Map)

Pilot Project Network Overview Total Length of Roadway Segments (miles) Rural Interstates 2.1 Urban Interstates 60.3 Rural Principle Arterials 28.6 Urban Principle Arterials 232.7 Urban Freeways/Expressways 35.5 Rural Minor Arterials 13.6 Urban Minor Arterials 513.2 Rural Major collectors 7.7 Urban Collectors 383.3 Rural Minor Collector 81.7 Local Roads 3183.5

NeXTA – Study Area Network Network Properties # Nodes: 26,348 # Links: 38,606 # O/D pairs: 1,267 Limitation: 23,000 nodes

Pilot Project Work Zones Work Zone Type Frequency (1992-2016) Completed/Ongoing 511 Proposed/Not Started 51 Total 562

Pilot Project Work Zone Characteristics (1)

Objective in WISE (Planning Mode) Find a feasible work zone schedule and sequence such that the overall cost (Z) on the network is reduced to the minimized, where: overall cost = user costs (traffic delay) + agency costs (Eq. 1) Subject to constraints including link flow conservation, path flow, and OD pair demand, etc.

Solution algorithm WISE Scheduling Engine Heuristic method, Tabu search to find optimal sequence of projects with defined starting month times, and construction mode (daytime, night time, or both) K-SP: Network Wide Travel Time k-shortest path (k-SP) algorithm to find alternative paths where travel time is less than the path going through the work zone *(b) Assign diverge traffic from the work zone path to K-SP by equilibrium method Evaluate network-wide travel time accounting for traffic diversion

Traffic Diversion Analysis (Operation Mode) Estimate traffic diversion due to the capacity reduction *(a) from the external DTA model, DynusT Isolates a single work zone project Runs DTA for diversion analysis with the work zone in place Operation Diversion = 1 - (((Base Volume - Project Volume)/(Base Volume)) X 100) Operational Diversion is made available to the user in the Project Information Section once the Base and Project Scenarios have been run for a given Project.

Traffic Diversion Analysis (Operation Mode) Projects: Lists all Projects Within the Planning Mode Project Modified: True if Project Information has been modified since Volume Data was last collected False if Project has not been modified since Volume Data was last collected. Base Volume: Nexta/Dynust Volume for Link without a Project: Null if Nexta/DynusT has not been run for Base Scenario Contains the Volume Collected from Nexta/DynusT when Base Scenario is executed Project Volume: Nexta/Dynust Volume for Project Link: Null if Nexta/DynusT has not been run forProject Scenario Contains the Volume Collected from Nexta/DynusT when Project Scenario is executed Run Selected Project Scenario: Runs Nexta/DynusT for a Single Selected Project in the “Projects” list and collects the “Project Volume” Data Run Base Scenario: Runs Nexta/DynusT for Base Scenario, not containing any Projects the “Base Volume” Data for all Projects in the “Project” list.

Overall algorithm procedure Step 1: Initialization Set each project an initial feasible schedule. Step 2: Neighborhood search For in each construction mode: daytime only, nighttime only, and both: For in each project: For in each month: If schedule project to start at month with a mode of is feasible: Do traffic diversion. Compute the objective of Eq. (1). If it results in a reduced objective, schedule project starting from month with construction mode , update the corresponding solution; Add triple into Tabu list, and restrict this solution in the next few iterations (user defined). Otherwise: continue Step 3: Check stop criteria Stop if a predefined maximal iteration number is reached, or in continuous 5 iterations it does not find a solution with improved objective. Step 4: Return the solution with the minimum objective that has been found. Why didn’t they put DTA in the loop? Need to redesign…

Complementary Comments Compare with other methods, such as (can be more than the list below) Distributed Genetic Algorithm (GA) Simulated annealing Particle swarm optimization Simulation based optimization What if *(a) and *(b) consider combinatorial projects in place? Improve the method computing traffic diversion; Improve K-SP; Automate scheduling and traffic impact assessing procedure, batch run DTA?

Complementary Comments Combinatorial work zone sequencing 1. Work zone set (One at a time) 1. Work zone set and sequencing algorithm (Potential combinations) 2. Network wide travel time 2. Network wide travel time 3. Sequencing after individual WZ are analyzed Improved, relatively complex and realistic Current, simplistic, and far from reality

WISE structure WISE interface Static Assignment Demand Converter Scheduling Algorithm DynusT NEXTA

Development environment Programming language: C# for Interface of WISE Python for modules including Demand Converter, Static Assignment, Scheduling Algorithm C++ for NEXTA, traffic network editor and visualizer Fortune for DynusT, dynamic traffic assignment engine

NeXTA – Simplified Study Area Network Network Properties # Nodes: 119 # Links: 348 # O/D pairs: 26

Scheduling Results WorkZone Link Capacity Reduction Original Speed New Speed Start Month Duration (months) From To 1 22 95 0.25 65 20 Apr-15 2 87 0.20 25 May-15 3 101 104 55 Mar-15 4 27 28 0.15 15 5 39 41 0.22 45 6 96 92 Jun-15 7 79 Jul-15 8 90 83 0.3 30

Challenges and Limitations Data Preparation Identify work zones to be analyzed State, MPO and City (all have different databases!) Sub-area selection if the network is bigger WISE uses DynusT for DTA Provide flexibility to include other software Significant effort needed mesoscopic model calibration Detail construction cost components not defined as input in WISE Labor, materials, tools, schedule conflict, and other components

WISE Software Review Summary Improvement in four dimensions Inputs and output data flow in NexTA Make it easier for practitioners Importing features from existing DOT/MPO data formats Standards for signal and control features Improved processes Engineering dimensions Algorithms Improved algorithms for both project sequencing, and traffic assignment Single level versus bi-level Consider combinatorial problems Consider larger number of work zones GUI User friendly GUI for practitioners

Q & A Sabya Mishra, Ph.D, P.E. Assistant Professor University of Memphis Email: smishra3@Memphis.edu Eric Hands back to Nicole.