Optimization of ANFIS Network using Wavelet Transform for simulation of Long term Rainfall of Rasht City

Estimation and forecasting of precipitation pattern in different parts of the world, especially in arid and semi-arid regions such as Iran is very important. In addition, various numerical methods such as artificial intelligence methods due to high accuracy and speed have the ability to simulate the phenomenon of precipitation and similar subjects. The use of these techniques plays an important role in saving time and costs in field and laboratory studies. Therefore, the application and popularity of various artificial intelligence techniques to estimate and simulate different issues such as rainfall is increasing day by day. The purpose of this study is to estimate the long-term rainfall in Rasht by a hybrid ANFIS and wavelet transform model.

In this study, long-term rainfall of Rasht city was simulated using an optimum artificial intelligence model for a 62 years period from 1956 to 2017. In other words, the wavelet transform was utilized to enhance the performance of the ANFIS model and hybrid WANFIS (Wavelet-ANFIS) model was defined. Firstly, the effective lags of time series data were detected through the autocorrelation function (ACF). Then, using the lags, eight models were developed for each ANFIS and WANFIS model. It should be noted that 42 years data was applied for training and 20 years data to test the artificial intelligence models. Next, the number of optimal membership functions of ANFIS model was selected equal to two.

results of ANFIS 1 to ANFIS 8 were evaluated. Additionally, different mother wavelets were examined to optimize the ANFIS model. This means that the demy was introduced as the best mother wavelet for increasing the performance of the ANFIS model. The comparison between ANFIS and WANFIS models signified that the wavelet transform enhanced the performance of the ANFIS model. Also, results of the hybrid WANFIS models were analyzed, indicating that WANFIS 8 was the superior model. The model estimated the rainfall with an acceptable accuracy. For instance, the R, MARE and RMSE for the superior model were computed 0.961, 0.855 and 24.510, respectively. Additionally, the values of VAF and NSC for this model were respectively estimated as 92.273 and 0.913.

Results showed that (t-1), (t-2), (t-3) and (t-12) were identified as the most influenced lags for estimation of long-term rainfall of Rasht city using the hybrid WANFIS model.