استفاده از سيستم استنتاجي فازي عصبي در تخمين بار رسوبي و مقايسه آن با مدل‎هاي MLR وSRC در حوضه رودخانه قرانقو

انتقال رسوب‎ها در رودخانه‎ها با توجه به نقش آنها در مباحث هيدرولوژيكي، از اهميت ويژه‎اي برخوردار است. اين رسوب‎ها به روش‎هاي گوناگون اندازه‎گيري مي‎شوند. اندازه‎گيري مستقيم بار معلق رسوبي در رودخانه، هزينه‎بر بوده و امكان احداث ايستگاه‎هاي اندازه‎گيري در تمام طول رودخانه وجود ندارد. همچنين معادله‎هاي مورد استفاده در تخمين بار رسوبي، براي تمام مناطق قابل استفاده نبوده و علاوه‎بر آن، نيازمند ديده‎باني‎هاي بلندمدت است. با اين حال، برخي از روش‎ها در تخمين بار معلق رسوبي به نتايج مطلوبي دست يافته‎اند. در اين مطالعه، سيستم استنتاجي فازي عصبي (ANFIS) با بهره‎گيري از تركيب‎هاي ورودي مختلف براي تخمين بار معلق رسوبي روزانه به‎كار گرفته شد. به اين منظور در اولين بخش از پژوهش، مدل ANFIS با استفاده از داده‎هاي دِبي روزانه و بار معلق رسوبي روزهاي پيشين، تعليم داده شده و براي تخمين بار معلق رسوبي رودخانه قرانقو مورد استفاده قرار گرفت. در دومين بخش از پژوهش، مدل ANFIS با استفاده از شاخص‎هاي ضريب تبيين (R2) و خطاي مجذور ميانگين مربعات (RMSE) با مدل‎هاي منحني سنجه رسوبي (SRC) و رگرسيون چندمتغيره (MLR) مقايسه شد. نتايج نشان داد كه مدل ANFIS با برخورداري از مقادير ضريب تبيين (R2) برابر 9668/0، RMSE برابر 190، در مقايسه با ساير روش‎ها از قابليت بهتري در تخمين بار معلق رسوبي برخوردار است. در اين بين، مدل SRC با برخورداري از مقادير R2 و RMSE كه به‎ترتيب معادل 8384/0 و 454 تخمين‎زده شده است، به ضعيف‎ترين تحليل در پيش‎بيني بار معلق رسوبي دست يافته است.

Introduction Prediction of sediment load is used in a wide range of topics to estimate volume of dams, sediment transport in rivers and etc. In recent years, artificial neural network was used in rainfall-runoff modeling, prediction of discharge intensity and estimation of sediment load. Sediments are sources of pollutions such as chemical compounds. The results of the many researches indicated the effectiveness of modeling in hydrological predictions. Jin (2001) used Artificial Neural Network (ANN) method to assess the relationship between discharge and sediment load and stated that the ANN model can achieve better results than the sediment rating curves. Tayfor (2002) used the neural network model in sediment transport and concluded that this model was more predictive than the physical models. In this paper, Neural Fuzzy Inference System (ANFIS) is used as a non-linear model to estimate the suspended sediment load. The comparisons showed that the ANFIS method has achieved better results in predicting the daily suspended sediment load than MLR models and SRC models. Dogan et al (2005) also used Artificial Neural Network model (ANN) and fuzzy logic (FL) to predict monthly suspended sediment load in the Sakarya River Basin in Turkey. Methodology In this study, to determine the amount of suspended sediment load, average daily discharge, rainfall and Gharnghu river basin sediment data (1387 to 1388) have been used as the material. Thus, the above data first have been entered in fuzzy neural models (ANFIS), multivariable regression (MLR) and the sediment rating curve (SRC). Then a comparison between them has been made to determine the ability of each model. Observed data and predicted data replaced with R2 and RMSE and according to these values the best model has been determined. Results and Discussion The purpose of the suspended sediment modeling studies is establishing significant relationships between discharge and sediment data. For this purpose several methods have been used. In this paper, daily discharge, current and the previous day rainfalls and suspended sediment load data have been used as the inputs for the model. The amount of sediment has been predicted by the neural fuzzy inference system, multiple regression equations and sediment rating curves. Then, a comparison was made between the results and the ability of each model. Table1. Performance of ANFIS, MLR and SRC models R2 RMSE Models 0.9668 190 ANFIS 0.8946 381 MLR 0.8384 454 SRC The comparisons have showed that the ANFIS model with R2 value about 0.9668 and RMSE about 190 has achieved the best result. Table 2 shows that the ANFIS model performs better than the MLR and SRC models. The ANFIS and MLR models have given better estimates of the maximum sediment load than the SRC model. The ANFIS, MLR and SRC models have predicted the maximum amount of the sediment load up to 6549, 5982 and 5329, respectively. These values have been estimated 11, 19, and 28% lower than the observed value. ANFIS models in comparison with the MLR and SRC models have high potential in establishing relationship between discharge and suspended sediment load. Sediment rating curve models establish the linear regression relations between the logarithm of the sediment and discharge values. Thus, these models require a normal distribution of the data and this is one of the main weaknesses of the models. The main characteristic of the ANFIS model is its flexibility and ability in making nonlinear relationships. Conclusion sediment load. The inputs of these models are rainfall, discharge and sediment data. In the first part of this research, regression equations have been set between discharge and rainfall data. In the second stage, discharge, rainfall and sediment variables set as the ANFIS model inputs and have been used in estimating suspended sediment load. Then in the third phase, the ANFIS model is compared with SRC and MLR models. The value about 0.9668 has been obtained for ANFIS model by using R2 factor and it shows that the ANFIS model has better performance than the other models. Besides, the MLR model has achieved better results than the SRC model. To estimate suspended sediment load in SRC model, the discharge factor has been applied. Conducted researches indicate that rainfall and sediment data must also be used beside discharge data. The main advantage of the ANFIS model relative to other models is their capabilities in modeling nonlinear relationships. Overall, the ANFIS model achieves better results than other models.