Techno-economic and deep learning-based assessment of wind-driven green hydrogen fuel production in Scandinavia

Abstract

Green hydrogen production via wind energy is critical for decarbonization, yet its viability is challenged by the spatiotemporal intermittency of wind resources. This study presents a novel techno-economic assessment framework that integrates a physics-based conversion model with advanced spatiotemporal forecasting architecture to analyze green hydrogen potential across six strategic locations in Scandinavia. An efficient dual-path deep learning architecture integrating patch-based transformers and convolutional neural networks is proposed to capture both long and short temporal dependencies and spatially localized weather conditions from 19-year hourly datasets (2005-2023). The proposed model achieved a mean squared error of 5.02 & times; 10(-)(4) across all locations, a 21.31% improvement against conventional CNN-LSTM hybrid architectures, with R-2 values ranging from 0.9276 to 0.9544. Coastal sites show the best prediction stability, with RMSE < 0.012 while inland locations exhibit higher uncertainty. Feature importance analysis identified wind speed as the most important predictor, with the coastal positioning significantly enhancing the prediction accuracy by 2.8-fold compared to inland locations. Monte Carlo simulation with region-specific sensitivity coefficients was conducted to assess LCOH, which showed Norway as the most economically competitive region (& euro;6.47 +/- 0.56 kg(-)(1)), and Denmark offers superior operational stability. Sensitivity analysis confirms that capacity factor (elasticity = -0.59) outweighs CAPEX as the primary driver of economic viability. Seasonal analysis showed winter production peaks in all locations and summer minimum, a critical factor for reducing downstream hydrogen storage requirements. These findings provide a quantified basis for a differentiated regional hydrogen infrastructure, prioritizing Norway for cost-effectiveness and Denmark for grid reliability.