Courier Route Optimization - Traveling Salesperson Problem With Time Window(TSPTW)

Summary This thesis implemented optimization algorithms and heuristics for solving the travel salesperson problem with time-window (TSP-TW) to optimize the routes and reduce all associated costs for a courier delivery company that dominates the Jordanian and middle eastern markets. New clusters were conducted using the k-means algorithm based on historical geo-coordinates; each cluster contained less than 100 stops; thus, the suitable model for each cluster was (TSP-TW), which keeps the solving time feasible for the NP-hard TSP-TW. The complexity order for TSP-TW is O((b+2)2). Implementing the TSPTW algorithm by (Bektas et al., 2005) and (Joubert et al., 2006) is not workable for the actual operations of the fleet, as this model assumes that the starting depot and ending depots are two different nodes; I have changed the algorithm to accommodate for one depot of ending and start of the trip, still while reserving the sub-tour eliminating constraint (SEC). Before running the model with actual data, all algorithms were tested using benchmark data (Langevin et al., 1993). The research's effort also entailed implementing a construction heuristic (i.e., nearest neighbor). In case business needs arise for N-hard routes that will have infeasible solution time when done on the TSPTW algorithm but will have shorter run-time on the nearest neighbor heuristic but with a compromised solution quality (less than near-optimal). The clustering and K-means implementation are based on Maptitude mapping software (i.e., GIS); while the TSP-TW, I constructed using IBM ILOG CPLEX Optimization Studio; finally, the heuristic is JAVA-based. As a result, the newly constructed clusters have increased the throughput, delivering more stops and reducing costs. Also, assigning one courier for bank deliveries with time windows is feasible and eliminates the percentage of failure to deliver within the allowed time window. K-means clustering resulted in a new demarcation for routes, which distributed the 1,500 nodes (i.e., stops) in Amman into clusters, and each cluster is served by one vehicle. The new demarcation of territories suggests operating 18 couriers instead of 35, and the clusters have grown in terms of daily stops, increasing throughput per cluster (i.e., courier). Based on the optimal schedules resulting from solving the TSP-TW, the test run in real life excelled the KPI by 0% failure to deliver on time by the courier (in compliance with the promised time window) and a shorter total travel time per route. The results are promising, and we highly encourage expanding this into specific services that would be a core competence for the courier company, where the customer has to provide a reliable address and Geo-coordinates of the designated delivery/pickup. The implementation has been preliminarily studied, and we suggested manually feeding our TSP-TW with time matrices for each courier to yield a daily master schedule per each cluster. An API connection with Google Maps obtains the latter matrices.