AN INTEGRATED MODEL FOR AI-BASED FORECASTING OF TECHNICAL SERVICE DEMAND AND OPTIMAL RESOURCE ALLOCATION IN AGRO-INDUSTRIAL COMPLEXES

This paper proposes an integrated decision model for forecasting technical-service demand and allocating limited maintenance resources in agro-industrial complexes. The forecasting layer represents daily, intermittent service requests by zone and service category through a Poisson gradient-boosting model with calendar, weather, fleet, workload and lagged-demand features. A split-conformal procedure converts point predictions into empirical uncertainty margins. The prescriptive layer is formulated as a mixed-integer linear program that assigns team-days and service jobs subject to skill compatibility, regular and overtime capacity, spare-parts availability and shortage penalties. A reproducible controlled experiment with 21,600 zone-category-day observations is used to test the computational procedure. The HGB–Poisson model reduces MAE and RMSE relative to seasonal-naive and Croston–SBA benchmarks. In a calendar-defined 14-day workload window, budgeted-robust AI planning improves realized service continuity compared with a historical seasonal plan. The results support a rolling-horizon software architecture in which forecasts and allocation decisions are updated from completed work orders.