Akademik Veri Yönetim Sistemi
International Journal of Hydrogen Energy, cilt.219, 2026 (SCI-Expanded, Scopus)
This study forecasts volatile stock prices of fuel cells (FCs), a critical component of the renewable energy industry, using three artificial intelligence (AI) models: a multilayer perceptron (MLP), a long short-term memory (LSTM), and a deep LSTM. It analyzes five years of highly dynamic stock data from a leading FC company. A multi-dimensional grid search is conducted to identify the optimal configuration for each architecture by minimizing validation root mean squared error (RMSE), and the selected models are evaluated through comparative performance analysis and error-distribution diagnostics. Multi-step forecasting is performed for horizons up to 360 days, and predictive uncertainty is assessed through an optimized Monte Carlo (MC)–based analysis implemented via MC Dropout. By keeping dropout layers active during inference and generating stochastic forward passes, it derives 95% prediction intervals (PIs) that widen as the forecast horizon increases, reaching their widest values at 180–360 days. Overall, sequence models (LSTM and deep LSTM) consistently outperform the feed-forward MLP baseline as the horizon extends. Quantitatively, the best-performing sequential model (deep LSTM) achieves an RMSE of 0.97 and an R2 of 0.98 on the 30-day horizon, while maintaining stronger long-horizon stability than alternative recurrent configurations. These results provide an uncertainty-aware forecasting perspective for FC equity markets and offer data-driven insights relevant to clean-energy investment and policy discussions in the hydrogen economy.