Achieving a low-carbon and economically viable energy future is central to sustainable development. Green hydrogen-based integrated energy systems (GHIES) offer a promising solution, especially when coupled with renewable energy. However, challenges remain due to the intermittent nature of renewables and the complexity of managing multi-energy systems.

To address these issues, researchers have explored mechanisms such as carbon trading, which encourages emission reductions by placing a cost on carbon. Studies show this can reduce emissions significantly—by as much as 28% in some energy systems. However, this often comes at the expense of economic performance. Conversely, electricity trading has been found to improve financial outcomes, such as reducing compensation costs for distributed network operators. Yet, electricity trading alone offers limited incentives for decarbonisation.

To balance both carbon reduction and cost-effectiveness, the integration of carbon and electricity trading mechanisms within GHIES is gaining attention. Still, few studies have explored how these dual mechanisms function within hydrogen and energy storage systems—especially under conditions of uncertainty, such as fluctuating wind energy inputs.

This is where deep reinforcement learning (DRL) emerges as a game-changer. Unlike traditional model-based optimisation methods—such as genetic algorithms or stochastic programming—DRL does not require explicit mathematical models or predefined distributions for uncertainty. Instead, it learns optimal scheduling strategies through trial and error, adapting to complex, uncertain environments.

Among DRL algorithms, Proximal Policy Optimisation (PPO) has proven particularly effective due to its stability and reduced sensitivity to hyperparameters. Its application in energy systems is still novel, especially for GHIES, but early evidence suggests it can simultaneously lower operational costs and emissions.

This study proposes a new framework that integrates PPO-based DRL with carbon and electricity trading mechanisms to optimise GHIES performance. Using a 3E (energy, economic, environmental) evaluation model, the research demonstrates the dual benefit of reduced carbon emissions and operational costs. The smart scheduling system adapts dynamically to uncertainties in energy supply, setting a precedent for future energy infrastructure.

By combining advanced AI with policy-driven mechanisms, this approach establishes a scalable and sustainable pathway for green hydrogen systems and the broader decarbonisation of multi-energy networks.

Source: 

https://www.sciencedirect.com/science/article/abs/pii/S0960148125018403