Optimizing solar-electrolysis for green hydrogen production: A novel spatiotemporal attention framework (STAF) for solar-electrolysis prediction and economic viability analysis

The increasing economic activity and population across the globe has pushed the electricity demand, as it is crucial for industrialization [1]. Traditionally, this electricity demand has primarily been met through fossil fuel-based power plants [2]. However, these conventional energy sources cause a negative environmental effect by emitting pollutants in the environment as byproduct of electricity generation. Therefore, global efforts are being made to transition towards sustainable and clean energy production systems to meet the world’s growing energy demand. To address this global environmental threat, renewable energy sources provide a cleaner alternative to generate the electricity [3]. However, the intermittent generation characteristics, high installation cost and low efficiency poses a significant challenge for renewable energy power plants. Different energy carriers can be employed to address the limitation of renewable energy power plants such as batteries, hydrogen and synthetic fuels [4]. Each energy carrier has its own advantages and disadvantages. Lithium ion batteries offers a low-cost way to store the energy for short-term storage but have low energy density [5]. Synthetic fuels are new and innovative approach to carbon-neutral fuels, albeit with current high costs and energy-intensive production processes makes them uneconomical in current prospect.

Hydrogen energy presents a promising alternative to address issues associated with renewable energy by offering a versatile and clean fuel source that can be integrated into various sectors, including transportation, industry, and residential heating, thus reducing dependency on fossil fuels and decreasing greenhouse gas emissions [6]. Hydrogen offers several compelling benefits such as high energy density, versatility and can be stored during periods of low demand or high renewable production, and then used when energy demand is high or renewable energy supply is low, helping to balance the grid [7]. One of the biggest advantages of hydrogen is that it only releases water as byproduct in combustion. Currently, the world is focusing on using hydrogen as a fuel for automobiles, which could transform the transportation sector by reducing carbon emissions [8]. The global hydrogen fuel cell vehicle market has shown significant growth, with projections indicating an increase from 20,000 units in 2020 to over 1.5 million units by 2030, representing a compound annual growth rate of 54.7 % [9,10]. Moreover, the studies also have demonstrated that hydrogen fuel cell vehicles can reduce well-to-wheel greenhouse gas emissions by 20–85 % compared to conventional vehicles, depending on the hydrogen production method [11]. Furthermore, government initiatives worldwide, particularly in Japan, South Korea, and the European Union, have established ambitious targets for hydrogen vehicle adoption, supported by expanding refueling infrastructure networks [12]. Alongside these efforts, recent advancements in artificial intelligence are playing a pivotal role in enhancing hydrogen-powered transport; for instance, emerging studies have introduced an intelligent cross-type transferable energy management framework based on deep transfer reinforcement learning, which paves the way toward a fossil-free urban transport system [13]. Furthermore, innovative approaches such as a longevity-aware energy management framework for fuel cell hybrid electric buses, based on proximal policy optimization, and a battery health-aware, naturalistic data-driven energy management system for hybrid electric buses utilizing the deep reinforcement learning algorithm, have demonstrated significant improvements in energy optimization and vehicle longevity [14,15].

Hydrogen is the most abundant element on the earth, but it is not found in its natural form and must be extracted from other compounds. Hydrogen can be generated through several methods such as steam methane reforming (SMR), water electrolysis, coal and biomass gasification, and biological methods [[16], [17], [18]]. Each of these techniques has its own advantages and disadvantages. For instance, SMR and coal gasification are cost-effective and well-established methods for hydrogen generation but emit carbon dioxide during the process [16]. These generation methods are called grey hydrogen production methods, and currently SMR alone fulfills the major hydrogen demand, as shown in Fig. 1. Biological methods use microorganisms such as algae or bacteria to produce hydrogen through biological process. However, these methods have low production rates and separation of hydrogen is a challenging process from the output. The energy sector is striving to reduce carbon dependency, increasing interest in green hydrogen, which is produced using renewable energy sources to split water into hydrogen and oxygen. These renewable energy-based hydrogen energy systems are being considered as a potential solution to enhance energy independence, could stimulate domestic economies, and mitigate greenhouse gas emissions [19]. Moreover, the growth in solar power plants across the world is providing the base for solar power-based hydrogen production systems.

Currently most of the hydrogen is generated through the non-renewable sources which in turn pollute the environment in the process. So, to fulfill the hydrogen demand world is thriving towards the utilization of solar power to produce green hydrogen [7]. Solar power can be converted into hydrogen with photovoltaic electrolysis, direct solar water splitting, and solar thermochemical cycles. In photovoltaic electrolysis solar panels converts the solar radiation into direct current (DC) electricity. This DC is fed to the electrolyzer which splits the water molecule into hydrogen and oxygen molecules [20]. Direct solar water splitting directly utilizes the solar radiation to generate the hydrogen without any intermediate step with the help of semiconductors or biological microorganisms. However, solar thermochemical cycles utilize solar concentrators to drive thermochemical cycles. Thermochemical cycles include different chemical reactions to split the water into hydrogen [21]. In the current scenario the electrolysis method is gaining momentum and has highest market share for generating hydrogen [22]. Remaining two green hydrogen generation methods have significant potential due to higher efficiency but requires cost competitiveness and research progress.

The electrolysis of water with renewable energy sources provides a clean method to generate large quantity of hydrogen with no pollutant emission [23]. In the electrolysis water is divided into hydrogen and water molecules. Electrolyzers play a crucial role, in this process consisting of two electrodes submerged in water to facilitate the movement of hydrogen and oxygen molecules. The efficiency and cost of production largely depend on the selected electrolysis method for generating hydrogen. The three principal electrolysis techniques—alkaline electrolysis (AE), polymer electrolyte membrane (PEM), and solid oxide electrolysis cell (SOEC), each utilize divergent mechanisms for decomposing water into hydrogen and oxygen gases [24,25]. While each approach carries both benefits and limitations, a full assessment requires acknowledging their differing impacts. AEL method is one of the oldest methods to generate electricity and has lower installation cost and is well suited for large renewable energy installations [26]. However, it has slow response time and lower hydrogen conversion efficiency. PEM address the issues of AEL by providing higher efficiency and lower response time. SOEC has highest efficiency among all electrolysis methods but its higher temperature requirement makes it unsuitable for smaller installation.

Among all renewable energy utilization for hydrogen production, solar energy proves to be more viable due to its abundant availability [27]. Moreover, electrolysis powered by solar plants significantly reduces investment costs. Many research studies focus on identifying optimal locations for solar power plants to maximize hydrogen production. In some search research works hybrid renewable energy systems are also analyzed for finding optimal configuration of power plants [28]. Hybrid renewable energy power plant will be more consistent in providing the hydrogen compared to other isolated renewable energy power plants [29]. However, the cost of the overall system will increase. So solar energy-based system is an economical way to produce the hydrogen.

Accurately assessing hydrogen production from SPP is crucial for planning, utilization, and profitability. Effective planning requires precise forecasting of hydrogen generation. So, to make accurate prediction of solar radiation and power different research works have focused on different strategies [[30], [31], [32]]. In conventional prediction methods autoregressive moving average and autoregressive moving average methods are utilized for the prediction. These models are simple and less computationally extensive, however with the increase in number of variables and non-linearity these models fail to give good predictions. As the solar radiation and power is highly non-linear in characteristics to address this non-linearity the researchers have also utilized random forest, support vector machine, and ensemble methods for providing better accuracy of the prediction [33,34]. However, machine learning models are shallow in nature and not perform well on large dataset with high nonlinearity. The advancement in computational speed paved the way of deep learning (DL) models. In contemporary study of research, DL models are getting lot of attention. The neural networks are the basic building block of DL models. The weights and biases of the models are trained in multi-layer architecture. The multi-layer architecture of DL provides the strength to these models for better representation capacity [35]. And the integration of DL models for the prediction of solar powered hydrogen production will improve the operational efficiency of the hydrogen generation power plants. The DL models provide better green hydrogen prediction accuracy and ensemble decomposition of the dataset also improves the accuracy by targeting the relevant feature variable in the functional space of learning [36,37]. The efficient functioning of hydrogen production systems depends on the forecasting of solar energy output, which has led to the development of machine learning models, including hybrid dynamic optimization methods.

Additionally, the economic assessment of the hydrogen production is a crucial parameter for the effective planning of the project. Different research studies have compared the hydrogen production cost with different generation methods and concluded that hydrogen production with renewables are comparatively costlier with their counterparts.

The remining work is sub-divided into five sections. Section 2 describes the dataset utilized in this work, and its characteristics. Section 3 provides the detailed discussion of methodology adopted in this work. Section 4 discusses about the system design which is used as a base for simulation and next step training of the models. Subsequently in section 5 the outcome of the simulation and analysis is discussed, and finally the conclusion section concludes the outcome of the present work.