Scenario-recognition-based reinforcement learning dispatch with dynamic preference regulation for power grid supply assurance

Renewable power shortfalls can threaten supply adequacy during extreme peak demand when upward reserve and stored energy are limited. Fixed dispatch preferences may be overly conservative when response capability is sufficient yet provide inadequate support when it is scarce. This article proposes a reinforcement learning dispatch framework based on scenario recognition and dynamic preference regulation. Historical forecast-error scenarios provide the expected renewable availability and a conservative estimate based on a lower quantile. A supply adequacy risk index relates renewable uncertainty to load pressure, upward reserve, net load ramping, storage energy, and routine supply capability. A proximal policy optimization policy uses this information to select the operating mode, the relative weights of economic and supply adequacy objectives, and the renewable dispatch reference coefficient. Constrained scheduling allocates resources before renewable realization, and real-time corrective dispatch responds to the observed shortfall. The formulation enforces aggregate power balance and resource operating constraints. Offline learning reduces repeated online searches over dispatch preferences while retaining constrained physical resource allocation. In an aggregate multi-resource simulation based on the generation and load parameters of the IEEE 30-bus system, all 10 independently trained policies achieve zero energy not served within numerical tolerance on 200 common test trajectories. Compared with conservative dispatch, mean operating costs decrease by 5.05%, and renewable accommodation increases from 89.71% to 92.05%. The proposed method achieves similar mean operating costs to risk-index rule dispatch with deterministic reference optimization while reducing mean online computation time by 82.08%, from 38.64 ms to 6.92 ms per interval.

Authors

Institutions

Publication Details

Journal
Frontiers in Energy Research
Published
2026-11-19
DOI
https://doi.org/10.3389/fenrg.2026.1951801
Primary Topic
Electric Power System Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Scenario-recognition-based reinforcement learning dispatch with dynamic preference regulation for power grid supply assurance

Liang Lu, Duo Xu, Hong Zhou, Xin Li et al.
Frontiers in Energy Research
Electric Power System Optimization
article

Scenario-recognition-based reinforcement learning dispatch with dynamic preference regulation for power grid supply assurance

Liang Lu, Duo Xu, Hong Zhou, Xin Li, Yiyu Wen, Li Shen
article en

Abstract

Renewable power shortfalls can threaten supply adequacy during extreme peak demand when upward reserve and stored energy are limited. Fixed dispatch preferences may be overly conservative when response capability is sufficient yet provide inadequate support when it is scarce. This article proposes a reinforcement learning dispatch framework based on scenario recognition and dynamic preference regulation. Historical forecast-error scenarios provide the expected renewable availability and a conservative estimate based on a lower quantile. A supply adequacy risk index relates renewable uncertainty to load pressure, upward reserve, net load ramping, storage energy, and routine supply capability. A proximal policy optimization policy uses this information to select the operating mode, the relative weights of economic and supply adequacy objectives, and the renewable dispatch reference coefficient. Constrained scheduling allocates resources before renewable realization, and real-time corrective dispatch responds to the observed shortfall. The formulation enforces aggregate power balance and resource operating constraints. Offline learning reduces repeated online searches over dispatch preferences while retaining constrained physical resource allocation. In an aggregate multi-resource simulation based on the generation and load parameters of the IEEE 30-bus system, all 10 independently trained policies achieve zero energy not served within numerical tolerance on 200 common test trajectories. Compared with conservative dispatch, mean operating costs decrease by 5.05%, and renewable accommodation increases from 89.71% to 92.05%. The proposed method achieves similar mean operating costs to risk-index rule dispatch with deterministic reference optimization while reducing mean online computation time by 82.08%, from 38.64 ms to 6.92 ms per interval.

Frontiers in Energy ResearchVol. 14
State Grid Corporation of China (China) (CN), Sichuan University (CN)
Electric Power System Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon