Title: S3TS: Stochastic Scenario-Structured Tree Search for Advanced Planning Under Uncertainty

Abstract:

Ensuring the stable operation of electrical grids and their associated infrastructure relies heavily on efficient energy sector scheduling. This involves critical tasks such as optimizing the dispatch of storage systems and generation units. To be effective, a planning strategy must satisfy two key requirements: (a) it must leverage the abundant data available in modern grids by incorporating advanced, potentially non-linear system models, and (b) it must explicitly manage uncertainties, such as those introduced by renewable energy integration. Currently, while methods like Monte Carlo Tree Search can handle non-linearity and stochastic mathematical optimization can address uncertainty, no single technique successfully tackles both issues concurrently.

To close this gap, we introduce the Stochastic Scenario-Structured Tree Search (S3TS) algorithm. This approach utilizes scenario trees to explicitly represent uncertainty, thereby allowing for the seamless integration of sophisticated non-linear models. We assessed the performance of S3TS using a simulated problem involving the publication of demand response signals, which closely resembles the imbalance settlement mechanism used in Belgium. Our findings indicate that in linear environments where analytical solutions are feasible, S3TS achieves near-optimal results, with costs staying within 14% of the mathematically optimal solution conditioned on the scenario trees. In highly non-linear contexts, S3TS demonstrates a substantial advantage over baseline methods, delivering cost reductions of up to 51% compared to a myopic algorithm and 5.4% compared to deterministic MCTS.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC