Improving the efficiency of using the transport fleet in conditions of demand volatility

the article discusses the increase in the efficiency of using a vehicle fleet with short-term demand volatility. An integrated contour is proposed, which includes: probabilistic short-term load forecasting; capacity planning at the upper level; operational dynamic routing in the sliding horizon mode; performance monitoring. At the forecast stage, SARIMAX, gradient boosting (CatBoost/LightGBM) and recurrent neural networks are used, taking into account calendar, weather and spatial factors.; accuracy is assessed by MAPE/WAPE with rolling origin validation. The capacity plan is formalized as a mixed-integer task with a choice of outsourcing share by zones and time intervals, taking into account fixed and variable costs, as well as SLA penalties. Operational dispatching is solved as DVRPTW with repeated recalculation in case of significant events and compliance with shift restrictions, time windows and capacity. Uncertainty is taken into account in a scenario-based and robust manner: through time buffers, penalties for under-fulfillment of SLA, and the CVaR risk component in the objective function. The target criterion is to minimize the total logistical costs at a given level of service. Computational experiments and sensitive analysis to forecast errors show a reduction in empty mileage and stabilization of on-time deliveries with a moderate share of outsourcing. DVRPTW settings, sliding horizon mode, and the use of CVaR are described in the current literature and engineering documentation on which the work is based.

dynamic routing, DVRP, VRPTW, demand forecasting, outsourcing, subcontracting, robust optimization, SLA, CVaR