Operational researchers have formulated a two-phase optimization framework that reduces total freight distribution expenditures by five percent across large-scale container transshipment networks while absorbing vessel schedule disruptions through distributionally robust modeling. Developed around the multi-tributary delta network of Guangzhou Port in southern China, the engineering investigation resolves the pickup-and-delivery vehicle routing problem with time windows for heterogeneous river barges connecting international deep-sea terminals with hinterland ports. Maritime logistics systems increasingly rely on container-on-barge movements along inland waterways as an energy-efficient, low-carbon alternative to overland trucking, yet fluctuating deep-sea liner arrivals routinely trigger port congestion, idle berthing, and late delivery penalties. To counter these systemic inefficiencies, the researchers combined an exact decomposition strategy with an adaptive metaheuristic and a statistical distance uncertainty framework, demonstrating that structured mathematical scheduling simultaneously lowers total operating costs and guarantees service reliability across complex estuarine freight hubs.
The study examines the dual-circulation transshipment architecture of the Pearl River Delta, where ultra-large container ships carrying up to 20,000 twenty-foot equivalent units berth at the deep-water facilities of Nansha Port. Upon discharge, quay cranes and automated guided vehicles transfer import containers to storage yards, from which logistics planners dispatch feeder vessels through the branching tributary systems of the Xi River, Dong River, and Bei River. In reverse, export containers converge from distributed manufacturing centers along inland waterways to dedicated barge berths at Guangzhou Port for consolidation onto outbound ocean liners. Conventional river dispatching enforces static departure patterns that assign vessels to predetermined destinations on fixed weekdays, an operational rigidity that frequently causes low capacity utilization and steep inventory holding expenses when ocean liners encounter voyage delays. By formulating this logistics flow as a variable weekly scheduling model, the study captures the bidirectional cargo exchanges between maritime trunk routes and shallow river ports across multi-tiered regional waterways.
To capture the operational realities of delta freight transport, the mathematical formulation integrates heterogeneous fleet configurations with strict spatial, temporal, and navigational boundaries. Inland river operations deploy five distinct container barge classes with capacities of 30, 50, 100, 150, and 200 twenty-foot equivalent units traveling at an effective navigation speed of 17 kilometers per hour following channel curvature adjustments. The mathematical objective minimizes cumulative system expenditure, encompassing fixed vessel deployment expenses, distance-dependent fuel and operating costs, and penalty charges incurred when urgent consignments require supplementary road haulage. Operating constraints enforce pickup and delivery deadlines, vessel payload limits, navigation continuity along sparse river branches, dedicated berth allocations at Guangzhou Port’s twelve terminal berths, minimum fleet capacity utilization thresholds that can exceed one hundred percent under multi-order rotations, and minimum port service reliability ratios that restrict reliance on auxiliary overland trucking.
Navigational geography introduces severe structural sparsity into the routing network, distinguishing waterway logistics from fully connected road-based vehicle routing models. Deep-sea containers cannot travel directly between arbitrary river terminals without traversing circuitous junctions, because navigating across unconnected tributaries causes excessive transit delays and unsustainable fuel burn. To render large-scale instances computationally tractable, the researchers instituted a two-stage topological preprocessing methodology that evaluated 54 regional port nodes. The Guangzhou hub was designated as root node 1, while 53 regional terminals were numbered sequentially according to geographic coordinates. Using graph-theoretic clustering across seven secondary river sub-basins, the team merged adjacent terminals located within waterway distances of less than 30 kilometers and integrated service points exhibiting greater than eighty percent weekly schedule overlap, successfully consolidating the regional network into 14 representative characteristic shipping corridors.
Data preprocessing further tackled cargo volume imbalances through dynamic programming knapsack procedures designed for consignment consolidation and deconsolidation. When single customer demands exceed the maximum capacity of any available 200-unit vessel, the algorithm partitions the order across multiple hulls using cost-minimizing knapsack recursions that prevent oversized shipments from monopolizing multiple partial holds. Conversely, smaller consignments originating within the same tributary cluster and sharing identical delivery deadlines are systematically consolidated into aggregated shipping manifests before algorithmic routing. This systematic reduction preserves strict client service commitments while stripping redundant nodes from the combinatorial search space, establishing a clean mathematical baseline for both deterministic exact solvers and stochastic heuristic evaluations.
Because the resulting pickup-and-delivery routing formulation with time windows represents an NP-hard combinatorial optimization challenge, the researchers designed a two-phase algorithmic solution that accelerates convergence. In the first phase, a subproblem isolates high-marginal-cost trucking assignments, using the commercial solver Gurobi to determine which orders must be transferred by road to prevent deadline violations. In computational trials containing 50 orders, Gurobi resolved this road-assignment subproblem in 40 seconds, compared with 3,430 seconds required when attempting the full unassisted problem. Once overland truck nodes are fixed, the second phase tackles barge routing and fleet assignment using a hybrid metaheuristic that combines adaptive large neighborhood search with variable neighborhood search routines to explore solution spaces without becoming trapped in local mathematical optima.
The hybrid metaheuristic incorporates five specialized destruction operators and four greedy repair mechanisms calibrated specifically for river shipping dynamics. Destruction operators dynamically remove nodes based on uniform randomness, order expenditure contributions, calculated delivery time pressure, barge capacity utilization levels, or combined pickup and delivery volume weights. Complementary repair heuristics reinsert removed orders through random assignment, time-efficient path extension evaluations, payload matching into minimally sufficient vessel hulls, or cost-optimal insertion points calculated via the Bellman-Ford shortest-path algorithm operating in polynomial time. To balance broad exploration with intensified local search, operator selection probabilities are updated dynamically through a Softmax function governed by an adaptive simulated annealing cooling schedule, while an auxiliary variable neighborhood search engine executes order insertions, two-opt edge reconnections, and route cross-exchanges whenever consecutive iterations fail to yield structural cost reductions.
Controlled computational benchmarks on an AMD Ryzen 7 8845H processor demonstrated clear operational thresholds between exact mathematical programming and the proposed hybrid metaheuristic across 25 small-to-medium test instances and 25 large-scale network configurations. In small problem instances with fewer than 20 orders, the exact Gurobi solver converged more rapidly than the iterative heuristic. When order counts ranged between 10 and 30 across fleet sizes of 5 to 25 vessels, the hybrid algorithm achieved identical objective costs to exact mathematical benchmarks. However, as problem dimensions expanded to 50 orders, the exact solver struggled to identify optimal configurations within a one-hour computational ceiling, allowing the hybrid algorithm to produce superior solutions that reduced total expenses by 7.32 to 10.21 percent relative to Gurobi’s truncated search.
In large-scale computational evaluations containing 50 to 250 customer orders and 25 to 125 heterogeneous barges, the algorithm was evaluated against three benchmark neighborhood search techniques under a strict 600-second execution window. Across the 25 large-scale scenarios, the hybrid algorithm lowered total logistics costs by an average of five percent compared with standard simulated annealing, large neighborhood search, and variable neighborhood search implementations. For complex instances exceeding 100 orders, the hybrid approach achieved superior objective values in 95 percent of test cases. The numerical experiments also revealed non-linear cost escalation when order volumes surged against fixed vessel availability, driven by unavoidable reliance on high-cost trucking; in a stress case featuring 250 orders and 50 available barges, the schedule deployed all 50 vessels alongside 76 supplementary road trucks to preserve delivery integrity.
To resolve the vulnerabilities of deterministic dispatching against maritime schedule disruptions, the authors formulated a distributionally robust optimization model based on Kullback-Leibler divergence ambiguity sets. Rather than relying on rigid moment matching that can misrepresent asymmetric vessel delays, the statistical distance framework constructs probability ambiguity sets around empirical arrival distributions derived from historical order transit logs. Using sample average approximation over 1,000 synthetic sample scenarios, the formulation solves for minimal operational expenditure under worst-case probability distributions within bounded divergence distances, translating chance-constrained arrival guarantees into linear 0-1 mixed-integer inequalities through binary search optimization.
Empirical evaluation of a representative scheduling scenario encompassing 20 orders, 10 heterogeneous barges, and a 95 percent confidence threshold revealed the direct cost of operational resilience against voyage disruptions. Under container arrival uncertainty, overall fleet expenditure increased by 13.95 percent, rising from 29,400 yuan in the deterministic baseline to 33,500 yuan, while required vessel deployments expanded from seven to eight barges. Schedule adjustments concentrated primarily on overloaded routes, splitting a high-utilization vessel itinerary into two independent voyages while preserving stable trajectories on routes possessing adequate berth buffer intervals. Sensitivity assessments confirmed that lowering tolerance for delivery failure drives non-linear cost escalations toward infinity as guarantees approach absolute certainty, providing terminal managers with an explicit quantitative framework to balance barge fleet expenditures against guaranteed on-time delivery thresholds.
Subject of Research: Optimization of heterogeneous barge scheduling and inter-waterway transshipment under vessel arrival uncertainty
Article Title: Heterogeneous barge scheduling optimization considering inter-waterway transshipment
Article References: Qu, Z., Zhu, S. X., & Zhou, S. (2026). Heterogeneous barge scheduling optimization considering inter-waterway transshipment. Management System Engineering, 5(1), Article 12. https://doi.org/10.1007/s44176-026-00064-7
Image Credits: AI Generated
DOI: 10.1007/s44176-026-00064-7
Keywords: barge scheduling, inter-waterway transshipment, heterogeneous fleet, container-on-barge, pickup and delivery problem, time windows, adaptive large neighborhood search, variable neighborhood search, distributionally robust optimization, Kullback-Leibler divergence, Pearl River Delta, Guangzhou Port, sample average approximation, multimodal logistics, berth allocation, mixed integer programming, knapsack problem
News Source: Denise Maddox. (November 19, 2026). Hybrid Algorithm Optimizes Heterogeneous Barge Scheduling Under Maritime Transshipment Uncertainty. Scienmag.



