Research on a high-reliability-guided hierarchical dynamic optimal scheduling method for photovoltaic systems

In hierarchical scheduling of active distribution networks (ADNs), conventional methods often struggle to maintain the consistency of nodal power fluctuation characteristics while enforcing hierarchical forecast coherence. To address this issue, this paper proposes a high-reliability-guided hierarchical dynamic optimal scheduling method for photovoltaic (PV) power. First, Crossformer is employed to construct a nodal power forecasting model, and frequency-domain features are introduced to construct a forecast reliability proxy. On this basis, a dynamic hybrid reconciliation strategy is designed, which adaptively switches between bottom-up aggregation and top-down decomposition according to the reliability assessment results. In the aggregation path, high-reliability nodes are used to correct forecast deviations. In the decomposition path, a constraint-based neural network, namely ConsNN, is introduced to reconstruct nodal forecasts while strictly satisfying the aggregation-consistency constraint between system-level and node-level forecasts. Simulation results on the IEEE 33-bus system show that the proposed method can effectively balance system security and scheduling efficiency.