Introduction: Forecasting the trend of storms provides the possibility of planning to control the dust crisis for managers and officials. Artificial neural network patterns are among the most widely used methods for prediction. In most studies, researchers form several models to find the most suitable artificial neural network model that is more accurate in order to choose the model that has the most accuracy. The aim of this research is to combine the multivariate regression model and ANFIS fuzzy neural network in order to achieve the appropriate model faster.
Material and methods: In this research, the data of dusty days, temperature, rainfall, relative humidity taken from Zabol synoptic station, statistical yearbooks and statistics related to Hirmand river flow have been used. At first, using rainfall data to check the drought situation, the value of SPI index was calculated in R-Studio programming software. Then, the multivariable regression model was estimated using temperature data, maximum wind speed, Hirmand river flow, SPI index and the number of dusty days with the help of Eviews software. Finally, two ANFIS neural-fuzzy network models with different inputs were developed in MATLAB software. In this way, the input variables to the first model were randomly selected and the output of the regression model was used to determine the input variables of the second model.
Results and discussion: The results showed that dusty days have the highest correlation with the Hirmand river with a coefficient of -0.70 and then with the drought index (SPI) with a coefficient of -0.65. The SPI index includes positive and negative values, in the years when the value of the SPI index has decreased and indicates drought, dusty days have increased. After that, the correlation coefficient of temperature with the number of dusty days is 0.56 and finally the correlation coefficient of wind speed with dusty days is 0.45. The results of multivariate regression estimation showed that the variables of Hirmand river flow, SPI index, temperature and wind speed have a significant effect on the number of dusty days and their coefficient sign is as expected. Finally, using MATLAB software and ANFIS model, two models with different inputs were checked. In the first model, the variables of temperature, relative humidity and rainfall were used as inputs. In the second model, the variables of Hirmand river flow, SPI index and temperature, which according to the results of correlation coefficients and regression modeling had the greatest impact on dusty days, were determined as inputs. The results of the evaluation of two models showed that the second model, which is based on the results of regression modeling and examining correlation coefficients, has less error.
Conclusion: Predicting incidents is very effective in informing managers and planners in order to manage risks. Artificial neural networks are among the most important methods for prediction. But it is very important to choose the right inputs in order to increase the prediction accuracy with artificial neural network. In some studies, researchers have to build several neural networks to choose the one with the least error, which requires spending a lot of time. In the present research, firstly, the correlation coefficients between all the considered variables were calculated and the influence of the selected variables on the dependent variable was evaluated by estimating the multivariate regression model. The results showed that if variables with the highest correlation and significant impact on the dependent variable are selected as inputs to the fuzzy neural network, the accuracy of the ANFIS neural-fuzzy network increases.