A Multi-Agent and Constraint-Validation Framework for Sustainable Yellow River Water Allocation in Henan Province, China

Reliable and sustainable Yellow River water allocation in Henan Province, China, requires the coordinated use of heterogeneous evidence, sequential allocation calculations, operational constraints, and ecological-management requirements. Existing approaches often lack explicit coordination among demand forecasting, supply calibration, intake-gate allocation, and constraint-triggered feedback. This study proposes a constraint-gated multi-agent framework to support regional Yellow River water allocation. The framework integrates multi-source evidence retrieval with a sequential demand–supply–intake computational chain and sustainability-oriented constraint validation. Regional demand is predicted using a quota-prior-guided neural network (QPG-NN) that incorporates quota-derived demand information as both an input feature and a soft consistency constraint. Supply is calibrated using historical trends and demonstrated supply capacity, while intake-gate allocation combines historical trends, recent operational inertia, and regional supply adjustment. Ecological-water satisfaction, regional supply reliability, and sectoral satisfaction balance are embedded as auditable screening criteria. Experiments using 2015–2023 data from 13 prefecture-level regions in the Henan Yellow River supply area showed that QPG-NN achieved a mean RMSE of 0.736 × 108 m3 and an R2 of 0.976. Supply-calibration MAPE decreased from 10.05% to 5.76%, while intake-allocation WAPE decreased from 15.75% to 3.8%. The proposed framework obtained the highest mean expert score of 4.52 among the evaluated methods. In four synthetic drought scenarios, all cases triggered supply recalculation and passed a second constraint-validation evaluation after one automatic recalculation cycle. These results support the framework’s ability to coordinate traceable numerical computation, constraint-triggered feedback, and sustainability-oriented screening within an auditable water-allocation workflow.