Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna A Real-life Application of a Multi Depot Heterogeneous Dial-a-Ride Problem for Patient Transportation in Austria Patrick Hirsch and Marco Oberscheider Institute of Production and Logistics University of Natural Resources and Life Sciences, Vienna IN3-HAROSA Workshop
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Agenda Introduction Problem Description Method Numerical Studies Conclusion and Outlook
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Introduction Projects for Home-Health Care Public and Private Transport Rural and Urban Areas Daily and Weekly Scheduling Synchronized Tasks / Precedence Constraints Assignment Constraints (qualification, language,…) Time-dependent Travel Times (public transport) Scenarios with Natural Hazards Austrian Red Cross as Project-partner → Rich Vehicle Routing Problem (VRP)
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Problem Overview Optimization of Patient Transportation – No Emergency Services Austrian Red Cross Ex-Post Analysis Dial-A-Ride Problem Multiple Depots Pick-up and Delivery Locations Heterogeneous Car Fleet Aim get a schedule for a single day Implementation: Set Partitioning Problem Initial Solution Heuristic Tabu Search Metaheuristic P2 P P1 D1 P3 D2 P5 P4 D5 P8 D6 D7 P6 P7 D4 D3 D B B2 B3 B1 B4
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Problem Overview – Vehicles Auxiliary Ambulance „Casual“ car Transport of mobile patients One paramedic Patient Transport Ambulance Special car - equipment Stretcher, patient seat and wheelchair place Two paramedics
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Problem Overview - Model Objective: Minimize the operation time (= handling and driving time) of the used vehicles Constraints: Each request has to be served Time windows at pick-up and delivery locations Maximum ride times Given shifts and mandatory breaks The order to return to the home-depot if idle Capacities of the vehicles Exclusive use: e.g. radiation therapy or mental-health problems Auxiliary ambulance: Up to three mobile patients Patient transport ambulance: Two patients allowed – only one stretcher available
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Problem Overview – Possible Solution B2 B1 P2 P1 D1 P3 D2 P6 P7 D4 D3 P8 D6 D7 P5 P4 D5
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Example → 50 Minutes → 45 Minutes 1-1 → 25 Minutes 2-2 → 30 Minutes 3-3 → 15 Minutes → 55 Minutes → 90 Minutes …
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Example → 50 Minutes → 45 Minutes 1-1 → 25 Minutes 2-2 → 30 Minutes 3-3 → 15 Minutes → 55 Minutes → 90 Minutes …
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Example → 45 Minutes 1-1 → 25 Minutes 2-2 → 30 Minutes 3-3 → 15 Minutes → 55 Minutes → 90 Minutes …
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Example → 45 Minutes 3-3 → 15 Minutes
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Metaheuristic Solution Approach (1) Tabu Search Algorithm based on the Unified Tabu Search method from Cordeau et al. (2001) task moves with local reoptimization – insert the task at the cost-optimal position in the new tour fixed tabu durations – depending on the number of tasks and vehicles aspiration criteria – attribute related intermediate infeasible solutions penalization of worsening candidate solutions by adding costs which are dependent on how often an attribute was used in a solution diversification strategy “Standard” Tabu Search (TS) implies the whole neighborhood of a solution time-consuming and not suitable for large problem instances
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Method Metaheuristic Solution Approach (2) Tabu Search with Alternating Strategy Static (TSAS-stat) (Gronalt and Hirsch, 2007) motivated by “Granular Tabu Search” (Toth and Vigo (2003)) concentrates on “bad” connections in current solutions sort the links according to their duration in a descending order select a predefined number of links starting from the one with the longest duration only these links are chosen to be removed in neighbor solutions – other links can only be modified if a task from a removed link is inserted after a certain number of iteration steps with a limited neighborhood an iteration step with full neighborhood search is set Tabu Search with Alternating Strategy Dynamic (TSAS-dyn) (Hirsch, 2011) if there is no improvement in the solution quality for a predefined number of iteration steps → change the neighborhood structure automatically an iteration step with full neighborhood search is set after a predefined number of iteration steps
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Parameters (1) Real-Life Data - Three Scenarios (days) 24 Hours Eight Hour Shifts 30 minutes time windows Allowed maximum ride time depends on shortest path SP < 10 min → 10 min 10 min ≤ SP ≥ 30 min → 100 % SP > 30 min → 30 min 10 % exclusive transports Scenario# Patient Transports# Vehicles# Depots Maximum Medium Minimum
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Parameters (2) Manipulation time depends on vehicle type mobility of patient (stretcher, wheelchair or mobile) hospital/ward two patients having the same pick-up or delivery location (parallelization possible?) Manipulation times are based on statistical analysis of > 80,000 patient transports Driving speed of vehicles: Interstate highways Limited access highways Other highways Arterial roads Other streets 100 km/h85 km/h60 km/h45 km/h30 km/h
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Map (1)
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Map (2)
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Initial Solution (1) Small dataset (221 patient transports) tested yet With given parameters the used routing would not be feasible Shifts → manually altered Combinations (TW, MRT) Comparison only possible to a certain degree Initial solution heuristic uses two versions to determine the best vehicle for the next task: Vehicles have to return to their base if idle G…total driving time = dt(depot,pick-up) D…total driving time = dt(delivery,depot) + dt(depot,pick-up) Example: If G → Red vehicle will perform 3 If D → Green vehicle will perform min 10 min 20 min
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Numerical Studies – Initial Solution (2) NE…No exclusive transports 40…Time windows of 40 minutes +5….Extension of maximum ride times of 5 minutes Version# VehiclesTotal Driving time [min.] Empty Driving time [min.] Working time [min.] Waiting time [min.] Real598,7804,02111,02317,837 V_G529,0653,94511,38217,478 V_D579,2033,83111,19817,662 V_G_NE508,7923,78911,11017,750 V_D_NE579,2143,68810,89017,970 V_G_NE_40468,3713,45310,51018,350 V_D_NE_40558,6303,45810,52518,335 V_G_NE_+5488,3483,34510,39918,461 V_D_NE_+5568,2343,21110,23718,623 V_G_+5508,6303,81111,06717,793 V_D_+5568,6493,48910,69118,169
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Conclusions and the Way Forward Combinations as input for Set Partitioning Problem Formation of Tasks Manipulation times are dependent on the transported patients Feasible combinations are strongly dependent on Length of Time Window Maximum Ride Time Short computation time to get a feasible result with initial solution heuristic (~ 1 second) TS and TSAS (static and dynamic) implementation work in process the two different initial solution heuristics indicate the potential for improvement heuristics
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna Thank you for your attention!
Institute of Production and Logistics – University of Natural Resources and Life Sciences, Vienna References Cordeau J.-F., Laporte G., Mercier A., A unified tabu search heuristic for vehicle routing problems with time windows. Journal of the Operational Research Society 52, Gronalt M., Hirsch P., Log-Truck scheduling with a tabu search strategy. In: Doerner, K.F., Gendreau, M., Greistorfer, P., Gutjahr, W.J., Hartl, R.F., Reimann, M. (Eds.), Metaheuristics - Progress in Complex Systems Optimization, 65-88; Springer, New York. Hirsch P., Minimizing empty truck loads in round timber transport with tabu search strategies. International Journal of Information Systems and Supply Chain Management 4(2), Toth P., Vigo D., The Granular Tabu Search and Its Application to the Vehicle Routing Problem. INFORMS Journal on Computing 15(4),