Abstract Accurate forecasting of solar radiation is critical for optimizing renewable energy production, enhancing grid stability and supporting planning and management in environmental and water resource applications. However, the non-stationary nature of solar radiation and variability in atmospheric conditions often result in unreliable estimates, particulary in the case of traditional models. To address this challenge, the current study proposed a number of hybrid models by integrating the Adaptive Neuro-Fuzzy Inference System (ANFIS) with Fuzzy C-Means(FCM) clustering and Subtractive Clustering(SUB). This resulted in the ANFIS-FCM and ANFIS-SUB models, while the classical ANFIS model, was used as a benchmark model. The models were evaluated against deep learning techniques, such as Long Short-Term Memory(LSTM), machine learning models like Extreme Learning Machine(ELM), and ensemble methods including Random Forest(RF). Furthermore, the study integrated FCB and SUB-clustering with LSTM, RF, and ELM for one-day-ahead solar radiation forecasting at the Fargo station in North Dakota. The study used a dataset of daily solar radiation measurements recorded from January 1, 2000, to December 31, 2019. Two input selection scenarios were considered: the first used the partial autocorrelation function to generate ten input combinations, while the second incorporated additional weather parameters, including humidity and air temperature. The results demonstrated that the ANFIS-FCM model outperformed all other models, achieving higher correlation coefficient(0.831), Willmott Index(0.921), and low relative error. Sensitivity analysis indicated that air temperature had a greater overall influence on forecasting than humidity. Overall, the proposed ANFIS model backed by FCM clustering demonstrated strong forecasting ability for daily solar radiation, making it an efficient tool for managing solar-powered systems in real-world power grids.
AlOmar et al. (Thu,) studied this question.