Analysis of Demand Variability and Robustness in Strategic Transportation Planning May 25, 21 Cambridge, MA Ahmedali Lokhandwala Candidate for M.S. in Transportation & M.Eng. in Logistics and Supply Chain Management, 21 Thesis Advisor: Dr. Chris Caplice Overview Trucking Industry Trucking is the largest mode of commercial freight transportation in the US, contributing $66 billion in revenue and hauling 68.9% of all freight All carriers in the industry need to develop a strategic transportation policy to anticipate and satisfy customer requirements at lowest costs and with high service quality Criticality of Transportation Planning Most traditional off-the-shelf planning software packages use average demand per week occurring on transportation lanes as the basis for planning Supply/demand/geographical variability over the transportation network results in sub-optimal execution of the plans Need of the Hour Planning mechanisms need to incorporate variability Robust planning methods must be easy to execute at an operational level 2 1
Planning versus Execution Long Term Planning Fleet Acquisition Allocation of fleet capacity Capital investments Hiring Drivers Relationships with For-Hire Carriers Gap Demand Variability Complexities in execution Short-term Planning Allocating fleet capacity locally to optimal routes Emergency shipments of unmet requirements 3 Freight Network Optimization Tool (FNOT) FNOT is a Large-Scale Linear Optimization Model created by the MIT-CTL Freight Lab team FNOT uses stochastic methods for allocating weekly demand over a transportation lane to an optimal mix of private fleet and for-hire carriers FNOT also creates optimal tour routes for hauling demand on the private fleet Thus, FNOT creates a planning strategy by optimizing load movements over the distribution network 4 2
FNOT Demand Allocation Vendor Warehouse Tour 2 Tour 3 Store Tour 1 Warehouse Store Vendor Warehouse Store Loaded Leg Empty Leg 5 Need for Simulations FNOT is a steady state model Real-time application of the model requires operational flexibility to re-optimize the network on a weekly basis this adds a lot of complexity during execution Important to simulate weekly demand scenarios over the network for a year, and test the plan created by FNOT to see if it is robust 6 3
Planning Hierarchy FNOT Planning Model Long Term Annual Planning Methodology for creating Plans Stochastic Plan Deterministic Plan Operational Flexibility in Annual Plans Complete Operational Flexibility Zero Operational Flexibility (Not Addressed in this Thesis) Variations in Annual Plan through Heuristics Minimum Volume Assignment Maximum Volume Assignment 7 Plan of Action Complete Flexibility Volume allocation reoptimized every week Best-case scenario v/s Zero Flexibility Volume allocated to stochastic plan tours only Excess demand goes to for-hire 8 4
Metrics used for Analysis Network Level Costs, Volume Allocation, Average Cost per Load, Average Cost per mile Facility Level Tour Level Lane Level Costs, Volume, Distance, Loaded miles, Tour details, Driver utilization Tour Occurrence, Tour Fickleness, Demand Statistics Lane Fickleness, Demand Statistics 9 Network & Facility Level Results Minimum Volume Assignment scenario is within 6% of Total Costs for Complete Flexibility Scenario Standard Deviations for metrics in the Min. Vol. Assgnmt. case is lower indicating tighter bounds on weekly operations Facility Level metrics mirror the results observed at the Network Level indicating no loss of generality in behavior of metrics 1 5
Tour Level Analysis Frequency 25 2 15 1 5 Histogram & CDF of All Tours in Complete Flexibility Operational Plan 1 2 3 4 5 1 15 2 25 3 35 4 45 5 52 No. of weeks Frequency Cumulative % 1% 8% 6% 4% 2% % Consistent Tours! Cumulative %age of Occurrence 11 Tour Level Analysis Frequency 6 5 4 3 2 1 Histogram & CDF of Stochastic Plan Tours 1% 8% 6% 4% 2% % 5 1 15 2 25 3 35 4 45 5 52 No. of Weeks Greater Percentage of Consistent Tours!! Frequency Cumulative % Cumulative %age of Occurrence 12 6
Tour Level Analysis No. of Tours 2 15 1 5 Break-down of the No. of weekly generated tours that consist of Stochastic Plan Tours Ad-hoc tours generated, that were not part of the Annual Plan (29%) No. of tours generated, that were part of the Annual Plan (71%) 1 2 3 4 5 Operational Plan Simulation Week No. of Tours in Full Flexibility Operational Plan No. of Tours from Stochastic Plan 13 Lane Level Analysis Lane Volume Assignment to Private Fleet 16 1% No. of Lanes 14 12 1 8 6 4 2 81% of these lanes are Outbound (DC-S) Lanes 86% of these lanes are Inbound (V-DC) Lanes 8% 6% 4% 2% Cumulative %age %.5.1.2.3.4.5.6.7.8.9.95 1 Lane Fleet Fickleness Factor Frequency Cumulative % 14 7
Lane Level Analysis Frequency 16 14 12 1 8 6 4 2 Lane Volume Assignment to For-Hire Carriers 82% of these lanes are Inbound (V-DC) Lanes 78% of these lanes are Outbound (DC-S) Lanes.5.1.2.3.4.5.6.7.8.9.95 1 Lane For-Hire Fickleness Factor Frequency Cumulative % 1% 8% 6% 4% 2% % Cumulative %age 15 Conclusions The Stochastic Plan created by FNOT is operationally viable using the Minimum Volume Assignment scenario It handles 71% of Demand Variability It is within 6% of the Optimal Cost Solution It is easy to execute, since it requires zero operational flexibility on a weekly basis It reduces the chances of sub-optimal local decisionmaking and maintains the network at close to systemoptimal levels It reduces the gap between planning v/s execution 16 8
THANK YOU QUESTIONS? 17 9