Forecasting the Class of Daily Clearness Index for PV Applications

This paper deals with the problem of forecasting the class of the daily clearness index which can be relevant for PV applications. A large number of solar stations, publicly available, was processed by using five different approaches, namely, the feed-forward neural networks, the Hidden Markov models, the Naive-Bayes models, the Surrogate models and the Persistent models. Experimental results show that one-day ahead forecasting of the class of daily clearness can be reliable performed in a 2-class framework and with less accuracy in a 3-class framework. Furthermore, for this purpose, the HMM approach is recommended among the considered ones. The global performance of the class prediction models, evaluated by calculating the average confusion rate (CR), showed that using HMM models provide CR 0.3 for 2-class clustering classes, while, for the 3-class framework it rises to 0.35.