Abstract:

The uncertainty in the long-term growth of renewable energy generation output and load has heightened the complexity of power grid planning. Conducting an uncertainty analysis on the long-term scale of renewable energy output and load is of significant importance for the planning and construction of the power grid. To address this issue, a multi-stage scenario tree generation method for wind-solar load based on complex feature extraction and Sinkhorn distance was proposed. Firstly, to enhance the clustering efficiency of wind-solar load scenarios, a method based on stacked sparse autoencoders for feature extraction of wind-solar load scenarios was introduced. The feature set of wind-solar load scenarios was clustered by using an improved affinity propagation algorithm based on density peak, and typical curves of wind-solar load were obtained as the root nodes of the scenario tree. Subsequently, by considering different growth rates in load, a yearly generation of wind-solar load scenario trees was performed, and a scenario tree reduction method based on Sinkhorn distance was proposed to reduce the size of the scenario tree. Finally, a simulation example showed that the proposed method had high calculation efficiency, and the generated multi-stage scenario tree for wind-solar load can reflect the uncertainty of wind-solar output and load growth.

Key words: scenario tree, stacked sparse autoencoder, improved affinity propagation algorithm, Sinkhorn distance, load uncertainty