RSE-AS：再エネマイクログリッドエネルギー管理のための研究コード、学習済みSACモデル5種、評価結果

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:蓄電池制御への強化学習適用における因果観測設計と分布外ロバスト性評価の手法を参考にできる。

🏢実務担当者:蓄電池・マイクログリッド運用のコスト最適化にRLを使う際の現実的な削減幅(約2%)と限界を把握できる。

🏛政策担当者:需給調整・蓄電池活用の政策設計において、運用最適化の効果量と統計的頑健性の限界を踏まえるべき。

This repository accompanies the research study entitled “Causality-Aware Risk-Sensitive Economic Action Selection for Battery-Constrained Deep Reinforcement Learning in Renewable Microgrid Energy Management.” It provides the research code, five trained Soft Actor-Critic (SAC) models, and experimental results for the proposed Risk-Sensitive Economic Action Selection (RSE-AS) method for battery energy storage management in renewable microgrids. Repository contents Research notebooks: Two Google Colab notebooks covering the research workflow and dataset preparation. Trained SAC models: Five saved policies corresponding to random seeds 42, 43, 44, 45, and 46, each with 100,000 recorded training steps. Chronological evaluation: Results for validation in 2021, full-year testing in 2022, and partial-year testing in 2023. Out-of-distribution evaluation: Results under increased electricity demand (+20%), reduced photovoltaic generation (−30%), increased electricity tariffs (+50%), and combined perturbations. Ablation analysis: Comparisons of RSE-AS with Fixed, Adaptive, Economic-Only, and unmodified policy actions. Statistical analysis: Per-seed results, paired cost differences, confidence intervals, and exact two-sided sign tests. Documentation: README, software requirements, reproducibility notes, data-source attribution, and licensing information. Experimental configuration The microgrid includes a 200 kWh battery with a maximum charging/discharging power of 50 kW and state-of-charge limits of 20–90%. The SAC policy receives a 12-dimensional causal observation vector and produces a one-dimensional battery-control action. Key results For the 2022 test period, RSE-AS achieved a mean adjusted electricity cost of USD 40,954.93 compared with USD 41,760.10 for the Fixed baseline, representing a mean saving of USD 805.17 (1.93%) across five trained policies. Cost savings relative to Fixed remained positive across all four out-of-distribution scenarios, ranging from 1.46% to 1.76%. The exact two-sided sign test for five improvements across five seeds yields p = 0.0625; therefore, statistical significance at the 0.05 level is not claimed based on this test. Reproducibility All five saved SAC models were successfully loaded and passed basic deterministic-inference checks in a compatible Python environment. The saved evaluation tables and paired statistical summaries were also checked for numerical consistency. The notebooks retain research-development stages rather than providing a fully automated, single-command reproduction pipeline. Independent end-to-end reproduction of the complete experimental workflow has not been established. Associated dataset The processed causal microgrid dataset is published separately on Zenodo: https://doi.org/10.5281/zenodo.23638036 Licensing The researchers' original software is provided under the MIT License. The researchers' model artifacts and numerical results are provided under Creative Commons Attribution 4.0 International (CC BY 4.0), subject to the rights notices and third-party licensing information included in the archive.

• Zenodo https://zenodo.org/records/23638735first seen 2026-10-26 04:13:27