A hypergraph-based dispatchable potential forecasting method for virtual power plants under price incentives

Accurate prediction of the dispatchable potential of a virtual power plant is fundamental to its participation in electricity markets and ancillary grid services. Existing studies often overlook higher-order interactions among heterogeneous distributed resources and their coupled responses to price incentives. To address this issue, this paper proposes HGNN-Former, a hybrid forecasting framework that integrates a hypergraph neural network with a Transformer. The framework first constructs a function-driven hypergraph prior by defining three types of hyperedges for heterogeneous VPP resources, namely resource type, response characteristic, and source-load-storage coordination, thereby forming a topology capable of capturing higher-order correlations. A temporal-hypergraph collaborative fusion mechanism is then designed. The Transformer extracts price-driven temporal dynamic features and maps them to hypergraph nodes, while hypergraph convolution propagates and aggregates higher-order spatial information among functionally related resources. The post-response net load under different price incentive scenarios is predicted through point forecasting. The response quantity under each incentive scenario is then derived relative to the baseline load, and the incremental potential is further obtained from cross-scenario differences. These two quantities characterize the dispatchable potential from the perspectives of individual incentive scenarios and cross-scenario variation, respectively. Simulation results show that HGNN-Former achieves superior post-response net load forecasting performance compared with baseline methods, demonstrating its effectiveness in dispatchable potential prediction.