Forecasting of Solar Power Generation Using Hybrid Empirical Mode Decomposition and Adaptive Neuro-Fuzzy Inference System

Excellent learning capabilities of Artificial Neural Network (ANN) made it widely explored and practiced to predict future points of a time series. Similarly, Fuzzy Logic (FL) has proved its capability in dealing with non-linear and complex models. The combination of these is not widely experimented. This paper not only aims to construct hybrid forecasting models using integration of Adaptive Neuro-Fuzzy Inference System (ANFIS) and Empirical Mode Decomposition (EMD) to forecast solar power generation but also discovers the effects of two different types of fuzzy Membership Functions (MFs), i.e., G-bell and Trapezoidal membership functions on forecasting capability of models. All proposed models were evaluated using two performance measures. Based on the forecasting results obtained, it is observed that the hybrid EMDANFIS model is outperforming when compared with the original ANN and ANFIS models, individually. The empirical results indicate that the models with trapezoidal MF (level-3) and G-bell shaped MF (level-1) are better models, providing the highest level of accuracy.