A Multi-Layer Reinforcement Learning and Genetic Algorithm Control Scheme for Grid Stability and Dispatch Control in PV–Fuel Cell Dominated Distribution Microgrids
Fuente:
Zenodo
Gespeichert in:
| 1. Verfasser: | |
|---|---|
| Format: | Recurso digital |
| Sprache: | Altenglisch |
| Veröffentlicht: |
Zenodo
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866901359549218816 |
|---|---|
| author | Adel Elgammal |
| author_facet | Adel Elgammal |
| contents | <p>ABSTRACT: The recent proliferation of photovoltaic (PV) based generation and hydrogen-based fuel cell (FC) systems in smart distribution networks pose great operational challenges, such as intermittency, voltage instability, frequency deviations, and suboptimal dispatch efficiency. Classical deterministic or rule-based control algorithms are not capable of effectively dealing with the nonlinear, stochastic and highly interconnected behaviors of PV–FC based MG systems. In this paper, a new multi-layer intelligent control framework, which combines RL for real-time scenario adaptive decision-making and GA to achieve the multi-objective optimization on energy dispatch and grid stability indices is presented. The upper RL is capable of updating the optimal control policies based on an online interaction with the microgrid environment, considering the complexities such as irradiance variations, uncertainties in loads, hydrogen storage dynamics and FC ramping constraints. There, the lower GA layer coordinates PV generation with FC output and battery state of charge trajectories, subject to optimal grid-supportive actions such as reactive power compensation and harmonic mitigation.</p> <p>An extensive MATLAB/Simulink-Python co-simulation framework was implemented to study the proposed topology in the presence of fluctuating renewable profiles, sudden load changes, FC fuel starvation operation modes and islanded/connected configurations. Simulation results show that the RL–GA hybrid controller is in much better performance compared with conventional PI, model predictive control (MPC) and sole RL approaches. Significant results were found such as a reduction in the variation of 35–48%, an improvement in frequency stability of 28%, up to 22% reduction is hydrogen consumption, and an increase in dispatch efficiency ranging from 30 to 55% subject to operational constraints and extension of FC Stack life. Furthermore, the OC system is robust to prediction errors, and to communication delays as well as component uncertainties. Results demonstrate the promising prospects of scalable hierarchical AI-based control and operation, for stable, efficient, and autonomous performance in renewable-driven microgrids. This study paves the way for the development of intelligent controllers to use in distributed energy systems with potential high penetration levels of renewables and hydrogen-integrated configurations.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17927034 |
| institution | Zenodo |
| language | ang |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | A Multi-Layer Reinforcement Learning and Genetic Algorithm Control Scheme for Grid Stability and Dispatch Control in PV–Fuel Cell Dominated Distribution Microgrids Adel Elgammal Reinforcement Learning (RL); Genetic Algorithm Multi-Objective Optimization (MOGA); PV–Fuel Cell Microgrids; Dispatch Optimization; Grid Stability; Voltage and Frequency Regulation;; Hydrogen Energy Systems; Smart Microgrid Management. <p>ABSTRACT: The recent proliferation of photovoltaic (PV) based generation and hydrogen-based fuel cell (FC) systems in smart distribution networks pose great operational challenges, such as intermittency, voltage instability, frequency deviations, and suboptimal dispatch efficiency. Classical deterministic or rule-based control algorithms are not capable of effectively dealing with the nonlinear, stochastic and highly interconnected behaviors of PV–FC based MG systems. In this paper, a new multi-layer intelligent control framework, which combines RL for real-time scenario adaptive decision-making and GA to achieve the multi-objective optimization on energy dispatch and grid stability indices is presented. The upper RL is capable of updating the optimal control policies based on an online interaction with the microgrid environment, considering the complexities such as irradiance variations, uncertainties in loads, hydrogen storage dynamics and FC ramping constraints. There, the lower GA layer coordinates PV generation with FC output and battery state of charge trajectories, subject to optimal grid-supportive actions such as reactive power compensation and harmonic mitigation.</p> <p>An extensive MATLAB/Simulink-Python co-simulation framework was implemented to study the proposed topology in the presence of fluctuating renewable profiles, sudden load changes, FC fuel starvation operation modes and islanded/connected configurations. Simulation results show that the RL–GA hybrid controller is in much better performance compared with conventional PI, model predictive control (MPC) and sole RL approaches. Significant results were found such as a reduction in the variation of 35–48%, an improvement in frequency stability of 28%, up to 22% reduction is hydrogen consumption, and an increase in dispatch efficiency ranging from 30 to 55% subject to operational constraints and extension of FC Stack life. Furthermore, the OC system is robust to prediction errors, and to communication delays as well as component uncertainties. Results demonstrate the promising prospects of scalable hierarchical AI-based control and operation, for stable, efficient, and autonomous performance in renewable-driven microgrids. This study paves the way for the development of intelligent controllers to use in distributed energy systems with potential high penetration levels of renewables and hydrogen-integrated configurations.</p> |
| title | A Multi-Layer Reinforcement Learning and Genetic Algorithm Control Scheme for Grid Stability and Dispatch Control in PV–Fuel Cell Dominated Distribution Microgrids |
| topic | Reinforcement Learning (RL); Genetic Algorithm Multi-Objective Optimization (MOGA); PV–Fuel Cell Microgrids; Dispatch Optimization; Grid Stability; Voltage and Frequency Regulation;; Hydrogen Energy Systems; Smart Microgrid Management. |
| url | https://doi.org/10.5281/zenodo.17927034 |