Enhancing power peak shaving with cascade hydropower: A buffer for wind and solar power uncertainty by deep learning-based data-driven approach

Dramatic increases in variable renewable energy (VRE) necessitate power grid flexibility to accommodate steeper fluctuations in net load, posing dramatic challenges for peak shaving. Hydropower can be a leading and renewable role in enhancing peak shaving, but only if operated according to adequate strategies considering VRE uncertainty and nonlinear operational characteristics of hydropower. Here, we present a deep learning-based data-driven framework designed to enhance peak shaving with cascade hydropower by adopting generative approaches and stochastic optimization. VRE uncertainty is captured by a hybrid approach that couples Wasserstein Generative Adversarial Networks and Variational Recurrent Autoencoder. The generated scenarios are then integrated into a stochastic optimization approach to derive peak shaving operational decisions. The proposed nonlinear model is transformed into a tractable mixed integer linear programming model using linearization techniques. Case studies are conducted for a provincial power grid in Southwest China. Results indicate that the proposed framework can effectively enhance power peak shaving with cascade hydropower, providing valuable insights for the operation of hydroplants in grids with high VRE penetration.