Winning the L2RPN Challenge: Power Grid Management via Semi-Markov Afterstate Actor-Critic
Authors: Deunsol Yoon, Sunghoon Hong, Byung-Jun Lee, Kee-Eung Kim
ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | This paper provides a formal description of the algorithmic aspect of our approach, as well as further experimental studies on diverse power grids. In this paper, we further evaluate our approach using Grid2Op, the open-source power grid simulation platform used in the competition, by training and testing the agent in 3 different sizes of power grids. We show that the agent significantly outperforms all of the baselines in all grids except for the small grid where the task was easy for all algorithms. |
| Researcher Affiliation | Collaboration | 1Graduate School of AI, KAIST, Daejeon, Republic of Korea 2School of Computing, KAIST, Daejeon, Republic of Korea 3Gauss Labs Inc., Seoul, Republic of Korea |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The detail of implementation is provided in Appendix A.3 and the code is provided in https://github.com/sunghoonhong/SMAAC. |
| Open Datasets | Yes | In this paper, we further evaluate our approach using Grid2Op, the open-source power grid simulation platform used in the competition, by training and testing the agent in 3 different sizes of power grids. Our experiments are conducted on the 3 power grids, IEEE-5 (smallest), IEEE-14, and L2RPN WCCI 2020 (largest, used in the challenge), provided by Grid2Op. |
| Dataset Splits | Yes | Figure 3 shows the total average scaled score of evaluation rollouts on the 10 validation scenario set during training: the scores are scaled in the range [-100,100]... |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions software components like Grid2Op, GNNs, and Adam optimizer, but does not provide specific version numbers for these or other ancillary software dependencies. |
| Experiment Setup | Yes | We use Ls = 6 GNN blocks with embedding dimension ks = 64/128 for shared layers. For actor s head, we use La = 3 GNN blocks with embedding dimension ka = 64/128. For critic s head, we use Lc = 1 GNN block with embedding dimension kc = 64/128 followed by linear layers with kc+ n /4 hidden units. We use δτ = 0/0.1/0.15. Adam optimizer (Kingma & Ba, 2015) is used for training with 5e 5 learning rate and 128 batch size. |