Improving Weather Forecasting Using De-Noising with Maximal Overlap Discrete Wavelet Transform and GA Based Neuro-Fuzzy Controller

Abstract

Adaptive Neuro-Fuzzy Inference Systems (ANFIS) is one of the most important neuro-fuzzy systems. ANFIS can be successfully applied to different real-world problems. However, it is difficult to create the ANFIS structure whose parameters directly influence the solutions. Therefore, hybrid ANFIS methods are generally used to increase efficiency and adaptability. This paper used an integrated neuro-fuzzy controller that is also known as PATSOS. The main purpose of this study is to improve the performance of the PATSOS method for weather forecasting. Our proposed PATSOS method is different from the previous ones since it embeds Genetic Algorithm (GA) into the PATSOS and also de-noising with Maximal Overlap Discrete Wavelet Transform (MODWT) is used to improve the data quality. GA is employed to optimize the moving average type, moving average degree, and de-noising degree. Furthermore, epoch number, membership function type, and membership function number for the PATSOS are optimized by GA. The results obtained by the hybrid PATSOS method are presented and compared with different cities and different models. It is concluded that proposed hybrid method forecasts daily mean temperature accurately. Proposed GA based PATSOS method can also provide remarkable advantages for determining parameter values in other complex, dynamic and non-linear forecasting problems.