.png){kind=icon}
.png){kind=icon}
Return
A Predictability-Guided Dynamic Reconciliation Architecture for Hierarchical Wind Power Forecasting
.png){kind=icon}
PRE
AI
DOI：10.1109/TSTE.2025.3627677.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Modern large-scale power systems typically operate under hierarchical dispatch, requiring multi-level wind power forecasting, especially in high penetration of renewable energy integration. These forecasts need to be reconciled to satisfy power balance constraints, commonly using empirical, optimal-combination, or neural network-based methods. However, these methods often degrade the accuracy of certain nodes to meet power balance constraints. To this end, we propose a predictability-guided dynamic reconciliation architecture for wind power forecasting. First, potentially high-accuracy forecasts are identified based on their predictability and further retained to guide forecasts reconciliation at other nodes through a hybrid bottom-up and top-down strategy, preserving the accuracy of each node. Second, a constrained neural network is designed to align child node forecasts with more predictable parent forecasts during the top-down process, leveraging the nonlinear modeling capabilities of neural networks to enhance child-node accuracy. Third, the method is extended to a dynamic and fine-grained architecture, better adapted to the time-varying attributes of wind power and mitigating the curse of dimensionality. Validation using realistic operational data from 212 wind farms demonstrates consistent accuracy improvements across levels and time horizons compared to benchmarks. Furthermore, the proposed architecture generalizes more effectively, as reflected in its enhanced accuracy across diverse base forecasting models.
Keywords:
Constrained neural network
frequency principle
hierarchical forecasting
reconciliation
predictability
Journal
I
IF:
10
Papers:
210
Citations:
0
Organization
View more organizations