Probabilistic wind speed forecasting via Bayesian DLMs and its application in green hydrogen production

The consolidation of green hydrogen as one of the mainstays of the energy transition depends on overcoming certain operational challenges, such as the predictability of primary energy. Building on this premise, some recent studies have incorporated deterministic wind forecasting techniques into the analysis of hydrogen production. However, few studies consider that, as accurate as these techniques are, they still involve inherent uncertainty. As a result, the hydrogen production process also inherits the uncertainty level present in deterministic forecasts. This research addresses this gap by evaluating the application of Bayesian Dynamic Linear Models as an alternative for probabilistic forecasting of wind speed in the context of hydrogen production. To this end, wind measurements from wind power plants located in the states of Bahia (BA), Ceará (CE) and Rio Grande do Sul (RS) in Brazil were used as input for a rolling forecast procedure. The 329 daily forecasts resulting from this procedure were used to estimate hydrogen production via three conversion methods, which differ from each other by calibrating the efficiency of the electrolytic process. Regarding point adjustment and probability of coverage, CE obtained the best results, while BA had relatively more accurate prediction intervals. The nRMSE medians for CE, RS, and BA are 0.1501, 0.2855, and 0.3272, respectively. The PICP medians for CE, RS, and BA are 97.92%, 70.83%, and 79.17%. The PINAW medians are 69.15%, 73.94%, and 65.66% for CE, RS, and BA. Among the electricity–hydrogen conversion methods, C2, which is theoretically calibrated, resulted in smaller differences between the observed and predicted values over the monthly horizon. From a daily perspective, the BA’s sample presented the highest number of critical periods in which fluctuations were greater than ±50% in relation to the immediately preceding period.