The integration of solar-powered electrolysis for green hydrogen production is constrained by the intermittent nature of photovoltaic (PV) generation, where forecasting inaccuracies can significantly increase operational risks and levelized costshhh. This study evaluates the techno-economic impact of machine learning (ML)-based solar forecasting on green hydrogen production costs by systematically comparing three distinct algorithms: Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM) networks. Using high-resolution solar and meteorological data (2005–2023) from the Tibetan Plateau, each model’s forecasting accuracy is assessed using standard metrics (RMSE, MAE, MAPE, R²). The predicted solar outputs are then integrated into comprehensive levelized cost of energy (LCOE) and levelized cost of hydrogen (LCOH) models for a 100 MW PV plant coupled with an alkaline electrolyzer. Results show that XGBoost achieves the most balanced performance with an RMSE of 0.52 MW, MAE of 0.39 MW, and R² of 0.994, yielding an LCOH of $3.53/kg H₂. Although SVR exhibits superior absolute error (RMSE = 0.29 MW), it results in a slightly higher LCOH ($3.54/kg H₂) due to higher relative errors during low-irradiance periods. Interestingly, LSTM, despite the weakest forecasting performance (RMSE = 6.52 MW), produces the lowest LCOH ($3.51/kg H₂), attributed to its smoothing of short-term variability, which stabilizes the electricity cost component. Sensitivity analysis identifies electricity cost as the dominant driver of LCOH, with a ±20 % variation causing an approximate change of ±$0.42/kg H₂, substantially outweighing the marginal economic impact of hourly forecast errors (< $0.02/kg H₂). The study concludes that while forecasting accuracy is important, the structure of forecast errors and broader economic assumptions particularly electricity price, electrolyzer efficiency, and utilization are more decisive for hydrogen cost competitiveness. These findings underscore the value of integrating ML-based solar forecasting into techno-economic models to enhance the bankability and deployment of cost-effective green hydrogen systems.

Keywords-  Solar power forecasting; Machine learning models; Levelized cost of hydrogen; Levelized cost of energy; Green hydrogen production; Artificial intelligence in energy.