Infinite hidden Markov model for short-term solar irradiance forecasting

Hidden state models are among the most widely used and efficient schemes for solar irradiance modeling in general forecasting particular. However, complexity of such – terms number states is usually needed to be specified a priori. For data this assumption very difficult justify. In paper, an infinite hidden Markov model (InfHMM) introduced short-term probabilistic irradiance, where fixed priori relaxed determined during training. InfHMM non-parametric Bayesian (NPB) indexed with dimensional parameter space which allows automatic adaptation “correct” complexity. This facilitates all weather conditions locations. Posterior inference performed using chain Monte Carlo algorithm, namely beam sampler. Data from 13 different sources validate proposed subsequently it compared two well-established literature: Markov-chain mixture distribution (MCM) complete-history persistence ensemble (CH-PeEn) models. Important results found, that cannot derived existing finite models, as variation within across sites. The comparison shows more consistent term horizon. reproducibility methodology presented we have provided R script supplementary material.