Hybrid fuzzy time series methods applied to solar radiation forecasting

Solar radiation incident on a surface varies randomly due to the dynamic characteristics of Earth's atmosphere. In order to plan, manage solar energy installations efficiently and to guide system designers, accurate solar radiation forecasting is essential. For this purpose, our study analyzes various time series and fuzzy integrated forecasting methods. Experimental tests have been carried out with both reference enrollment and solar radiation data. Statistical forecasting errors have been selected as performance measures. Fuzzy time series (FTS) are effective forecasting tools with uncertain data and they are widely used in economics, education, etc. This study is the first successful attempt at implementing FTS on radiation which possesses an irregular and random nature. Additionally, existing fuzzy models have been improved using 8 different hybrid models which combine and develop aspects of the original FTSs. As the radiation contains seasonal pattern, a deseasonalization procedure has been performed in order to reduce errors. The results have proved that the proposed Hybrid Model-8 shows higher performance compared to other fuzzy models and traditional time series methods.