A new fuzzy modeling method for the runoff simulation in the Karoon basin

نویسندگانTaheri Shahraiyni, H - Ghafouri, M - Bagheri, S - Saghafian, B
نشریهInternational Journal of Water Resources and Arid Environments
ارائه به نام دانشگاهSoil Conservation & Watershed Management Institute, Tehran, Iran
شماره صفحات440-449
شماره مجلد4
نوع مقالهFull Paper
تاریخ انتشار2011
رتبه نشریهISI
نوع نشریهچاپی
کشور محل چاپایالات متحدهٔ امریکا

چکیده مقاله

Fuzzy concepts and related inferences have been proposed a new approach to human modeling and calculation methods. Although, different powerful fuzzy modeling methods have been developed up to now, but some of these methods are different with real human modeling method, because of utilized mathematics and exact calculations in their constructions. Active Learning Method (ALM) is a fuzzy modeling method which uses a very basic level of mathematics. ALM has been innovated in 1997 and a new modified ALM developed in 2007. ALM has very simple algorithm that avoids of mathematical complexities. In this study, novel modified ALM has been utilized for the simulation of daily runoff in Karoon basin (one of the most important basins in Iran). Hence, the daily discharge data of Karoon River in Pol-e-Shaloo station from 1991 until 1999 were gathered for modeling. The first five years (1991-1995) were used for the training of ALM model and the residual data were used for the test of trained model. The impacts of changing the fuzzification points on the simulation results were investigated and the results showed that ALM for simulation of daily runoff is not sensitive to the position of fuzzy dividing points and best positions were determined. At the first step of modeling, the input data used for ALM modeling were daily precipitation, temperature, humidity and vapor pressure with different time lags. In this study, several statistical (The Nash-Sutcliffe, R , Bias, Root mean 2 square error (RMSE), peak weighted RMSE (PWRMSE) and Percent of total volume error (PTVE) values) and graphical (hydrograph, scatter plot and quantile-quantile) criteria were used for the evaluation of the ALM modeling results. NS, R , Bias, RMSE, PWRMSE and PTVE of the tested ALM model with 32 fuzzy 2 rules for daily runoff modeling were 0.29, 0.33, 65.8 (cms), 265.4 (cms), 418.9 (cms) and 22.3%, respectively. In general, the results of daily runoff simulation were not so good. Hence, in the next step of modeling, the daily discharges with different lags were added to the previous input dataset. The results showed that ALM with 32 rules is the best model for runoff simulation. NS, R , Bias, RMSE, PWRMSE and PTVE of the tested 2 ALM model with 32 fuzzy rules for daily runoff modeling were 0.75, 0.75, 1.3 (cms), 157 (cms), 357 (cms) and 0.5%, respectively. These excellent results demonstrated the effect of adding discharge data to input dataset and showed the ALM ability for daily runoff simulation. ALM could identify and rank the important variables for runoff simulation and determined that temperature and vapor pressure are unnecessary variables. In addition, training of ALM is very easy and straightforward in comparison with other artificial intelligence methods such as ANN (Artificial Neural Networks) and ANFIS (Adaptive Neuro-Fuzzy Inference Systems). Therefore, according to the ALM abilities, it has merit to be introduced as a novel and appropriate modeling method for the runoff simulation.

لینک ثابت مقاله