عنوان مقاله [English]
نویسندگان [English]چکیده [English]
Rain is one of the most important climatic factors affecting human activities which has also an important role in the field of water resources management. This weather phenomenon is a complex atmospheric process, which is highly dependent on space and time and thus not easy to predict. The trends of change in rainfall with time is a non-stationary stochastic process with high uncertainty and it is subject to various random factors. There have been many attempts to find the most appropriate method for rainfall prediction using for example meteorological or satellite data with a numerical weather prediction model, or even applying several techniques such as the artificial neural network or fuzzy logic as a forecasting approach. Also some methods, such as the time sequence method, probability statistics method cannot fully reflect the characteristics of the rainfall phenomenon, and the prediction results cannot be satisfactory. In order to improve the accuracy of rainfall forecasts, it is necessary to use a new rainfall prediction model such as intelligent methods and meta-heuristic algorithms. In this study, the “imperialist competitive algorithm” (ICA for brevity) and the ICA combined with the fuzzy logic algorithm were used to evaluate and compare their performance and ability in forecasting the amount of daily rainfall in semi-arid climate of Kerman in the southeast of Iran. So, 30 years of daily data in Kerman’s synoptic station (1981– 2010) and 10 years of daily data in Zarand and Rafsanjan’s synoptic stations (2001–2010) were used in the rainy season (7 months of the year). Therefore, based on the previous studies, five parameters including precipitation, wet temperature, dew point, relative humidity and cloudiness were used to forecast rainfall in futures days. Having surveyed the data, first the applied computer codes were written in Matlab 14. In the ICA with fuzzy logic, the ICA was used for determining the membership functions’ ranges and values of the weights instead of the trial and error usually used in application of the fuzzy logic. Three higher accurate outputs were identified for each station separately. Among these outputs, for each station, the best output was chosen and used for the final phase of optimization. Four more effective variables in Kerman’s station (precipitation, wet temperature, dew point, and cloudiness), two more effective variables in Rafsanjan’s station (precipitation and cloudiness) and three more effective variables in Zarand’s station (precipitation, wet temperature, and relative humidity) were identified after optimizing with five input variables. Results showed that the rainfall height’s prediction was accompanied with a significant error based on the mentioned methods, so that the coefficients of determination (R2 values) were obtained 0.54, 0.44 and 0.40 in, respectively, Kerman, Rafsanjan and Zarand’s synoptic stations. On the other hand, the forecast of the occurrence and non-occurrence of the rainfall with the ICA indicated reasonable results and in the best results 61.4%, 51.9% and 51.2% of days were predicted correctly in, respectively, Kerman, Rafsanjan and Zarand’s synoptic stations. The accuracy of calculations was improved with the ICA combined with the fuzzy logic. Accordingly, 89.63%, 82.31% and 74.12% of days were predicted correctly in, respectively, Kerman, Rafsanjan and Zarand’s synoptic stations. The results of evaluating the performance showed that the ICA can produce a relatively appropriate simulation of the occurrence and non-occurrence of rainfall in future days, but falls short of ability to simulate the rainfall height properly. On the other hand, the combined ICA and fuzzy logic algorithm provides a better simulation of problems involving high uncertainty.