| Περίληψη: |
Ports increasingly compete on the performance of their hinterland connections, where inland waterway transport (IWT) can provide high-capacity, lower-emission container flows but faces commercial uncertainty driven by volatile demand and shared asset constraints. This paper studies weekly pricing for an anonymized industry-proposed “fast barge” service serving five inland corridors on the Mississippi River system. We formulate a finite-horizon Markov decision process with stochastic, price-sensitive demand, corridor limits, and a shared fleet-hours allocation problem. Shipment manifests calibrate seasonality but contain no transaction prices, so reference tariffs and elasticities are explicit modeling assumptions. We compare transparent tariff, myopic, one-period, and shadow-price policies with proximal policy optimization (PPO), advantage actor–critic (A2C), and soft actor–critic (SAC), using common continuous actions, ten training seeds, and 200 paired evaluation seeds. A loose reference case remains price-boundary driven. A prespecified 300-hour benchmark has an interior unconstrained price and a positive deterministic capacity dual in 48.1% of weeks. In final evaluation, PPO and SAC improve on the validation-selected transparent benchmark by 0.089% (USD 51,800/year) and 0.362% (USD 211,100/year), respectively, with positive hierarchical intervals. Neither meets the prespecified 0.5% primary adoption screen, and the transparent benchmark retains 99.6% of SAC’s mean profit. The economic case for RL is therefore conditional on whether its annualized implementation and governance costs fall below the measured incremental value, rather than supporting an unconditional replacement recommendation. [ABSTRACT FROM AUTHOR] |