adaptive neural-fuzzy inference system

Density current is one of the factors influencing the transfer of sediments to reservoirs of dams. One of the practical methods to control sediments is to build an obstacle in the path of these currents.

In this laboratory research, the behavior of the Density current under the effect of cylindrical obstacles made of wood with a diameter of 1.5 cm and a height of 30 cm (more than the height of the body of the Density current) was evaluated. Therefore, by considering variables such as floor slope, concentration and discharge, the values of the density current head were determined. Machine learning algorithms such as adaptive neural fuzzy inference system and artificial neural network were used to model the results.

Based on the results, the density salt flow head was modeled using machine learning algorithms such as adaptive fuzzy neural inference system and artificial neural network and the performance of these two methods were compared. The results showed that machine learning algorithms are useful in modeling the density salt flow head. And the regression of the adaptive neural fuzzy inference system for the training and test data was 0.99 and the regression of the artificial neural network was 0.94 and 0.91, respectively.

By comparing the two methods, it was found that the adaptive neural-fuzzy inference system is more effective in modeling the percent reduction of the head of Density current than the feed-forward artificial neural network method.

Different regions have different potentials in dust release, and the increase in dust storms indicates the dominance of the desert ecosystem in each region. Prediction of the occurrence of dust storms in critical regions allow desion-makers to efficiently manage and to mitigate its probable damages to landscape. This study aims to predict the frequency of dust storm days (FDSD) over two critical regions (west and southeast) in Iran on a seasonal  scale. To this end, the hourly dust data and World Meteorological Organization codes were gathered in six synoptic stations of Zabol and Zahedan (southeast Iran), Abadan, Ahvaz, Bostan, and Masjed Soleiman (west Iran) covering the statistical period of 25 years (1990-2014). After calculating the frequency of dust storm days, using four artificial intelligence methods including multi-layer perceptron (MLP), adaptive neuro-fuzzy inference system (ANFIS), radial basis function (RBF), and general regression neural network (GRNN), the frequency of dust storm days for the next season were predicted. The results showed an increase in the accuracy of the predictions with increasing the number of dust storm days in such a way that based on the results obtained from the MLP model, the correlation coefficient between the observed and predicted values ​​of the frequency of dust storm days for Masjed Soleiman and Zabol stations were 0.8 and 0.97, respectively; explaining that Zabol have the highest frequency among these stations. Also, according to the results of ANFIS, RBF, and GRNN models, the correlation coefficient calculated for prediction in Masjed Soleiman and Zabol stations varied from 0.41 to 0.95, 0.35 to 0.92 and 0.22 to 0.98, respectively. Overall, by comparing the results of the proposed models, ANFIS had the best performance which was followed by GRNN. The results of this study can be useful in managing the issues caused by dust storms and in the combating plans to desertification in the study regions.