Vietnam — MinhPhong Le of the Faculty of Electrical and Electronic, Thu Duc College of Technology, Vietnam, has developed an energy management framework that combines Digital Twin technology and artificial intelligence to improve the efficiency of electricity distribution networks powered by solar energy. Published in 2026, the study reports that integrating a Digital Twin with the Proximal Policy Optimization (PPO) algorithm can reduce operating costs by 22.2% compared with conventional rule-based control. The framework also improves voltage regulation, reduces excessive battery cycling, and delivers more consistent operating performance when solar generation and electricity demand fluctuate.
The growing adoption of renewable energy is creating new opportunities for more sustainable electricity systems. However, integrating solar photovoltaic generation into distribution networks presents operational challenges because solar output depends on sunlight availability and weather conditions. Changes in solar generation can affect the balance between electricity supply and demand, while variations in consumption throughout the day can further influence power flows and voltage stability.
Battery Energy Storage Systems (BESS) offer one way to address these challenges. Batteries can store excess electricity generated during periods of high solar production and release the stored energy when generation declines or demand increases. Nevertheless, battery operation must be carefully managed. Excessive charging and discharging can increase operating costs and contribute to battery degradation over time.
In the article titled “Digital Twin-Assisted PPO-Based Energy Management for PV–BESS Operation in Active Distribution Networks Under Uncertainty,” MinhPhong Le combines Digital Twin technology with reinforcement learning to determine when batteries should charge or discharge. A Digital Twin is a virtual representation of a physical system that is updated using operational information. In electricity networks, this technology can reconstruct network conditions and verify them through electrical simulations before control decisions are evaluated.
The proposed controller uses Proximal Policy Optimization, an algorithm within Deep Reinforcement Learning. This approach enables an artificial intelligence agent to learn control decisions through repeated interactions with a simulated environment. The agent receives information about electricity demand, solar power generation, battery state of charge, and network voltage conditions. It then determines battery charging and discharging actions while considering operating costs, voltage deviations, and battery degradation.
Le evaluated the framework using the IEEE 33-bus distribution network model implemented in MATLAB and MATPOWER. The simulated system had solar photovoltaic penetration equivalent to 60% of peak demand, a battery energy capacity of 1.5 megawatt-hours, and a maximum charging or discharging power of 0.5 megawatts. The scheduling horizon covered 24 hours, with control decisions updated every 15 minutes.
To represent changing operating conditions, the study used 100 Monte Carlo scenarios. These scenarios included solar generation uncertainty of up to ±20% and electricity demand variations of up to ±15%. The PPO controller was trained over 2,000 episodes and then evaluated against three alternative strategies: rule-based control, deterministic optimal power flow, and standalone PPO-based Deep Reinforcement Learning without Digital Twin integration.
The results showed that the combined PPO and Digital Twin framework achieved improvements across several important performance indicators. Operating costs decreased by 22.2% compared with rule-based control. Maximum voltage deviation declined by approximately 51.9%, while average voltage deviation decreased by 55.9%. These findings indicate that battery management informed by network operating conditions can improve economic efficiency while supporting more stable voltage profiles.
The comparison with standalone PPO also highlighted the value of integrating the Digital Twin. PPO without the Digital Twin reduced operating costs by 16.9%, whereas the integrated framework achieved a 22.2% reduction. This difference suggests that reconstructing network conditions and validating them through AC power-flow calculations can help the learning algorithm make decisions that better reflect the electrical behavior represented in the simulation.
Battery utilization also improved. The average number of battery energy storage cycles declined from 2.8 cycles per day under rule-based control to 1.6 cycles per day with the PPO and Digital Twin framework. Battery state-of-charge variation also decreased. Reducing unnecessary charging and discharging may help limit battery wear and long-term replacement costs, although the study did not directly measure battery lifetime through physical hardware testing.
The framework also produced more consistent operating costs under uncertain conditions. The coefficient of variation in operating costs fell from 6.10% under rule-based control to 2.17% with the integrated approach. This result indicates that operating costs were more consistent across the evaluated simulation scenarios. However, greater uncertainty in solar generation still increased costs. At 10% solar generation uncertainty, the reported operating cost was US$942 per day, rising to US$1,087 per day when uncertainty reached 40%.
The findings may be relevant to electricity distribution operators, renewable energy developers, and organizations planning battery storage infrastructure. By combining virtual system modeling with adaptive artificial intelligence, the framework can support decisions that account for economic performance, voltage stability, and battery utilization at the same time. This capability is particularly relevant to distribution networks with high levels of solar generation, where electricity supply and demand can change rapidly.
Nevertheless, the findings are based on computer simulations rather than direct experiments on a physical electricity network. Practical implementation will require further validation of network model accuracy, communication latency, measurement synchronization, hardware integration, and scalability to larger distribution systems. Additional research could also investigate the integration of electric vehicles, demand response, and other distributed energy resources.
Overall, MinhPhong Le’s study demonstrates the potential of combining Digital Twin technology and reinforcement learning to improve solar energy and battery management. Lower operating costs, reduced voltage deviations, and fewer battery cycles suggest that data-driven control supported by electrical simulation could contribute to more efficient and adaptable electricity distribution networks. Further real-world testing will be necessary to establish the framework’s reliability and practical benefits under actual operating conditions.
Author Profile
MinhPhong Le
Faculty of Electrical and Electronic, Thu Duc College of Technology, Vietnam.
Research Source
Article title: Digital Twin-Assisted PPO-Based Energy Management for PV–BESS Operation in Active Distribution Networks Under Uncertainty.
Journal: East Asian Journal of Multidisciplinary Research (EAJMR), Volume 5, Issue 10, 2026, pages 4091–4110.
DOI: https://doi.org/10.55927/eajmr.v5i10.17
Journal link: https://eajmrformosa.my.id/index.php/eajmr
.jpg)
0 Komentar