Prediction of hourly solar radiation with multi-model framework

• A novel approach to predict solar radiation through the use of clustering paradigms.
• Development of prediction models based on the intrinsic pattern observed in each cluster.
• Prediction based on proper clustering and selection of model on current time provides better results than other methods.
• Experiments were conducted on actual solar radiation data obtained from a weather station in Singapore.

In this paper, a novel multi-model prediction framework for prediction of solar radiation is proposed. The framework started with the assumption that there are several patterns embedded in the solar radiation series. To extract the underlying pattern, the solar radiation series is first segmented into smaller subsequences, and the subsequences are further grouped into different clusters. For each cluster, an appropriate prediction model is trained. Hence a procedure for pattern identification is developed to identify the proper pattern that fits the current period. Based on this pattern, the corresponding prediction model is applied to obtain the prediction value. The prediction result of the proposed framework is then compared to other techniques. It is shown that the proposed framework provides superior performance as compared to others.