Global Journal of Engineering and Technology Research (GJETR)

Intelligent Self-Evolving Neural Control Using Multi-Objective Swarm Optimization and Reinforcement Learning for Dynamic Energy Management in PV-Driven Smart Cities

Elgammal, Adel

22 November 2025 · Vol. 1, Issue 3, pp. 96-108

DOI: 10.65150/EP-gjetr/V1E3/2025-03

Abstract

In this paper, an adaptive self-evolving neural control strategy based on the integration of Reinforcement Learning (RL) and Multi-Objective Particle Swarm Optimization (MOPSO) is proposed for dynamic energy management in Photovoltaic PV-based smart cities. The motivation behind this system is to cope with complexity of modern energy networks, uncertain large-scale solar generation and stochastic urban demand in combination with dynamic grid conditions. We adopt a self-evolving neural network (SENN), being the adaptive decision-making core, that dynamically tunes its parameters by reward-driven learning and Pareto-based optimization. The RL agent continuously controls power flow, energy storage distribution and load scheduling and MOPSO adjusts the multi-criterion objective indices including PQ improvement, voltage stability index, total cost of electrical system and carbon dioxide emission reduction. A hierarchical control architecture is proposed to integrate on-line operation optimisation of the microgrid while considering long-term city-scale prediction. The potential of the framework is demonstrated on a realistic urban PV-grid model including residential, commercial, and EV charging nodes with dynamic weather and demand. Simulation results show that the proposed SENN–MOPSO–RL controller outperforms the traditional Model Predictive Control (MPC), Proportional-Integral (PI) and Fuzzy Logic controllers. In particular, the total power loss is mitigated by 18.6%, voltage deviations by 27.4%, and the REs consumption improved about 22.9% respectively. Additionally, the controller kept grid stability for partial shading, demand surge and communication time delay, with good robustness and self-adaptation. The results validate that integrating swarm intelligence and reinforcement-based neural learning can result in a scalable and robust control framework for autonomous energy systems (AESs) found in PV-dominate smart cities. Also, hardware-in-the-loop validation and multi-agent interaction involving inter-microgrid collaboration with peer-to-peer trading will be investigated in the future.

Keywords: self-evolving neural network, reinforcement learning, multi-objective particle swarm optimization, photovoltaic smart grid, dynamic energy management, urban sustainability.

Read the full text on Global Journal of Engineering and Technology Research →The full peer-reviewed article and PDF are hosted on the journal's site (the version of record).

Elgammal, & Adel (2025). Intelligent Self-Evolving Neural Control Using Multi-Objective Swarm Optimization and Reinforcement Learning for Dynamic Energy Management in PV-Driven Smart Cities. Global Journal of Engineering and Technology Research, 1(3), 96-108. https://doi.org/10.65150/EP-gjetr/V1E3/2025-03

@article{Elgammal2025,
  title   = {Intelligent Self-Evolving Neural Control Using Multi-Objective Swarm Optimization and Reinforcement Learning for Dynamic Energy Management in PV-Driven Smart Cities},
  author  = {Elgammal and Adel},
  journal = {Global Journal of Engineering and Technology Research},
  year    = {2025},
  volume  = {1},
  number  = {3},
  pages   = {96-108},
  doi     = {10.65150/EP-gjetr/V1E3/2025-03},
  url     = {https://doi.org/10.65150/EP-gjetr/V1E3/2025-03}
}

← Back to all articles