View in EDS

Inertia Assessment Method for New Power Systems Based on Physics-Informed Reinforcement Learning.

Saved in:
Bibliographic Details
Title: Inertia Assessment Method for New Power Systems Based on Physics-Informed Reinforcement Learning.
Authors: Zhao, Yijie1 (AUTHOR), Shi, Linjun1 (AUTHOR) 19990041@hhu.edu.cn, Wu, Feng1 (AUTHOR), Wu, Chenyu1 (AUTHOR), Lin, Keman1 (AUTHOR)
Source: Electric Power Systems Research. Nov2026, Vol. 260, pN.PAG-N.PAG. 1p.
Subjects: Reinforcement learning, Generalization, Reward (Psychology), Prediction algorithms
Abstract: • Physics-informed deep deterministic policy gradient assesses power system inertia. • Two-level physics-informed strategy integrates physics into state and reward. • Distribution similarity describes probabilistic consistency. The inertia assessment of new power systems is confronted with insufficient accuracy and weak generalization capability, primarily due to strong nonlinearity and dynamic behavior. To address these challenges, a physics-informed deep deterministic policy gradient (PI-DDPG) method is used. First, the physics-based inertia estimated value is added to the state space as an auxiliary feature, providing the agent with an interpretable search starting point. Second, a reward function incorporating a hyperbolic tangent saturation accuracy reward and a disturbance compensation term is designed, which enhances training stability and incentivizes the agent to improve assessment performance under challenging conditions such as small disturbances. On this basis, the deep deterministic policy gradient algorithm is employed to train the agent to learn an inertia prediction strategy that follows physical principles while adapting to data complexity. This integration of physical information into reinforcement learning via feature concatenation and reward shaping aims to improve accuracy and robustness. Comparative experimental results demonstrate that the PI-DDPG method achieves higher accuracy and distribution similarity than pure DDPG, outperforms reward function variants lacking disturbance compensation, and shows better generalization on extreme test set compared to LSTM. [ABSTRACT FROM AUTHOR]
Copyright of Electric Power Systems Research is the property of Elsevier B.V. and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:• Physics-informed deep deterministic policy gradient assesses power system inertia. • Two-level physics-informed strategy integrates physics into state and reward. • Distribution similarity describes probabilistic consistency. The inertia assessment of new power systems is confronted with insufficient accuracy and weak generalization capability, primarily due to strong nonlinearity and dynamic behavior. To address these challenges, a physics-informed deep deterministic policy gradient (PI-DDPG) method is used. First, the physics-based inertia estimated value is added to the state space as an auxiliary feature, providing the agent with an interpretable search starting point. Second, a reward function incorporating a hyperbolic tangent saturation accuracy reward and a disturbance compensation term is designed, which enhances training stability and incentivizes the agent to improve assessment performance under challenging conditions such as small disturbances. On this basis, the deep deterministic policy gradient algorithm is employed to train the agent to learn an inertia prediction strategy that follows physical principles while adapting to data complexity. This integration of physical information into reinforcement learning via feature concatenation and reward shaping aims to improve accuracy and robustness. Comparative experimental results demonstrate that the PI-DDPG method achieves higher accuracy and distribution similarity than pure DDPG, outperforms reward function variants lacking disturbance compensation, and shows better generalization on extreme test set compared to LSTM. [ABSTRACT FROM AUTHOR]
ISSN:03787796
DOI:10.1016/j.epsr.2026.113379