Event-Driven Replanning in a Two-Echelon Transport and Warehousing Network of a Timber Enterprise: Model, Algorithm, and Computational Analysis

Introduction. Timber transportation at forestry enterprises follows a two-echelon scheme: from harvesting sites through intermediate hubs to receiving depots. Route plans in such networks are regularly disrupted by weather-induced road closures and equipment breakdowns. Two-echelon routing models (2E-VRP) have been developed mainly for static settings. Operational replanning mechanisms that simultaneously control replanning frequency and computational budget for forest two-echelon networks remain poorly studied. The objective is to propose an Event-Driven Receding-Horizon Replanning (ED-RHR) algorithm for such networks and to characterize its computational properties and its effect on service continuity. Materials and Methods. The network is modeled as a discrete-time controlled system on a directed graph with explicit state, control, and disturbance components. The ED-RHR algorithm operates on a rolling horizon: at each event-triggered step, 30 disturbance scenarios are generated and solved in parallel across 9 cores, with the static 2E-ACO ant colony algorithm used as the horizon subroutine. The experiment involved 2670 runs across 7 network classes (ranging from 17 to 339 vertices) with 4 control strategies, all performed as a single series in one version of the software implementation. The experiment comprised 2,670 runs across seven classes of networks (ranging from 17 to 339 vertices) using four control strategies, and was performed as a single series within one version of the software implementation. Results. The computational experiment yielded the following results The average decision-making time per cycle did not exceed 1.82 s, and solutions for all classes were produced within the computational budget. The growth in time was sublinear: a 20-fold increase in the number of vertices led to a 6.5-fold increase in time. For the M–XXXL classes, parallel efficiency ranged from 82% to 89%. The stability constraint R̄ = 2 reduced the number of reconfigurations by 13% without affecting service continuity. Replanning did not improve service continuity: none of the three replanning strategies, under any operating mode, outperformed the static plan in a statistically significant way. In the nominal mode, the static plan provided the highest level of service continuity for 5 out of 7 classes, while ED-RHR performed significantly worse than the static plan for 5 out of 7 classes. Discussion. The solution time is an order of magnitude below the computational budget, enabling real-time application of the loop on networks with more than 300 vertices. The decline in service continuity under frequent replanning is presumably attributable to changes in the set of serviced nodes given a limited fleet of vehicles, to the sensitivity of the metric to short-term service interruptions, and to selecting a solution for a single scenario rather than evaluating expected cost. Conclusion. ED-RHR is computationally feasible on networks of up to 339 vertices, and the stability constraint reduces route changes without compromising service quality. In the formulation studied, replanning did not improve service continuity. Future research will focus on aggregating solutions by expected cost, incorporating resource constraints into replanning, and moving toward a full flow model.