کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Deep Clustering for High-Fidelity Scenario Reduction in Stochastic Microgrid Optimization

نویسندگان :

Mahmoud Mollayousefi Zadeh¹ Fatemeh Aminizadeh² Mohammad Hossein Alizadeh Roknabadi³ Seyed Mohammad Taghi Bathaee⁴

1- دانشگاه صنعتی خواجه نصیرالدین طوسی 2- دانشگاه صنعتی امیرکبیر(پلی تکنیک تهران) 3- دانشگاه صنعتی امیرکبیر (پلی تکنیک تهران) 4- دانشگاه صنعتی خواجه نصیرالدین طوسی

کلمات کلیدی :

Scenario Reduction،K-means،Deep Clustering،Distance Metrics،Unit Commitment

چکیده :

Stochastic optimization for microgrids requires scenario reduction to ensure computational tractability, yet conventional methods such as k-means, k-medoids, and forward selection often rely on fixed distance metrics that fail to capture nonlinear temporal dependencies in renewable generation data. This paper introduces a deep clustering framework that leverages autoencoder-based representation learning to project high-dimensional scenarios into a nonlinear latent space, enabling more faithful clustering without predefined similarity measures. Numerical experiments on real-world wind data demonstrate that the proposed method achieves superior preservation of statistical features, with less than 0.5% deviation in expected operating cost compared to the full-scenario benchmark—significantly outperforming traditional techniques that introduce notable distortions. In terms of reduction quality, the framework improves Energy Score by up to 0.05 absolute units and maintains competitive computational efficiency, requiring only 0.02 seconds per reduction versus 25.58 seconds for backward elimination. A stochastic unit commitment case study confirms that higher-fidelity uncertainty representation directly enhances operational decision-making, yielding more robust and cost-efficient dispatch strategies. Overall, deep clustering establishes a scalable and statistically grounded paradigm for scenario reduction, advancing the reliability and economics of microgrid optimization under high renewable penetration.