Ultra-short-term Interval Prediction of Wind Power Based on Graph Neural Network and Improved Bootstrap Technique

Reliable and accurate ultra-short-term prediction of wind power is vital for the operation optimization systems. However, volatility intermittence pose uncertainties to traditional point prediction, resulting in an increased risk system operation. To represent uncertainty power, this paper proposes a new method interval based on graph neural network (GNN) improved bootstrap technique. Specifically, adjacent farms local meteorological factors are modeled as form from graph-theoretic perspective. Then, convolutional (GCN) bi-directional long short-term memory (Bi-LSTM) proposed capture temporal spatial features between nodes graph. obtain high-quality intervals (PIs), technique designed increase coverage percentage narrow PIs effectively. Numerical simulations demonstrate that can spatiotemporal correlation graph, results outperform popular baselines two real-world datasets, which implies high potential practical applications