پیش‌بینی بار معلق رودخانه با استفاده از سامانه‌های هوشمند

نویسندگان

چکیده

برآورد دقیق بار معلق رودخانه‌ها در طراحی و بهره‌برداری پروژه‌های آبی مهم است. تخمین میزان رسوبات با روش‌های مرسوم مانند منحنی سنجه نتایج دقیقی را دربر ندارد. در این پژوهش مدل برنامه‌ریزی بیان ژن که شکل توسعه یافته برنامه‌ریزی ژنتیک است،‏ برای تخمین میزان رسوبات معلق به کار گرفته شد. از نتایج حاصل از این مدل با نتایج مدل‌های فازی- عصبی،‏ شبکه‌های عصبی و منحنی سنجه مقایسه شد. در این راستا،‏ داده‌های جریان رودخانه و رسوبات معلق در ایستگاه ونیار واقع بر روی رودخانه آجی‌چای در استان آذربایجان شرقی استفاده شد. پارامترهای آماری ریشه میانگین مربعات خطا (RMSE)‎ و ضریب تعیین (R2) برای ارزیابی دقت مدل‌ها استفاده شدند. نتایج حاصله نشان از عملکرد بهتر مدل برنامه‌ریزی بیان ژن در مقایسه با سایر مدل‌های مورد استفاده بود. برای داده‌های دوره آزمون اختلاف نسبی بین RMSE مدل برنامه‌ریزی بیان ژن با مدل‌های فازی- عصبی از نوع افراز شبکه،‏ فازی- عصبی از نوع دسته‌بندی تفریقی،‏ شبکه‌های عصبی و منحنی سنجه به‌ترتیب برابر 8،‏ 10،‏ 13 و 21 درصد بود. همچنین به ازای بهترین الگوی مورد استفاده در مدل،‏ مقدار R2 برای مدل برنامه‌ریزی بیان ژن،‏ فازی- عصبی از نوع افراز شبکه،‏ فازی- عصبی از نوع دسته‌بندی تفریقی،‏ شبکه‌های عصبی و منحنی سنجه به‌ترتیب برابر 93‎/0،‏ 84‎/0،‏ 88‎/0،‏ 86‎/0 و 81‎/0 به دست آمد.

کلیدواژه‌ها


عنوان مقاله [English]

Estimation of suspended sediment in rivers using artificial intelligence techniques

نویسندگان [English]

  • h s
  • m n
  • d f
  • m m
چکیده [English]

Accurate estimation of suspended sediment in rivers is very important for designing and operation of water resources projects. Sediment estimation by conventional methods like rating curves don’t provide accurate results. In this paper, gene expression programming (GEP) model which is an extension of genetic programming (GP) technique, was used to estimate suspended sediment in the river. The GEP results were compared with those of the adaptive neuro-fuzzy, neural networks and rating curve models. In this regard, the streamflow and suspended sediment data from Vanyar station that located on Aji-chay river in East- Azarbaijan province are used. The root mean square errors (RMSE) and determination coefficient (R2) statistics were used to evaluate the accuracy of the models. The results showed that the GEP model had better performance than other considered models in estimating suspended sediment. The relative RMSE difference for the test period between GEP and ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 8, 10, 13 and 21%, respectively. The R2 values for GEP, ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 0.93, 0.84, 0.88, 0.86 and 0.81, respectively.
Accurate estimation of suspended sediment in rivers is very important for designing and operation of water resources projects. Sediment estimation by conventional methods like rating curves don’t provide accurate results. In this paper, gene expression programming (GEP) model which is an extension of genetic programming (GP) technique, was used to estimate suspended sediment in the river. The GEP results were compared with those of the adaptive neuro-fuzzy, neural networks and rating curve models. In this regard, the streamflow and suspended sediment data from Vanyar station that located on Aji-chay river in East- Azarbaijan province are used. The root mean square errors (RMSE) and determination coefficient (R2) statistics were used to evaluate the accuracy of the models. The results showed that the GEP model had better performance than other considered models in estimating suspended sediment. The relative RMSE difference for the test period between GEP and ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 8, 10, 13 and 21%, respectively. The R2 values for GEP, ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 0.93, 0.84, 0.88, 0.86 and 0.81, respectively.
Accurate estimation of suspended sediment in rivers is very important for designing and operation of water resources projects. Sediment estimation by conventional methods like rating curves don’t provide accurate results. In this paper, gene expression programming (GEP) model which is an extension of genetic programming (GP) technique, was used to estimate suspended sediment in the river. The GEP results were compared with those of the adaptive neuro-fuzzy, neural networks and rating curve models. In this regard, the streamflow and suspended sediment data from Vanyar station that located on Aji-chay river in East- Azarbaijan province are used. The root mean square errors (RMSE) and determination coefficient (R2) statistics were used to evaluate the accuracy of the models. The results showed that the GEP model had better performance than other considered models in estimating suspended sediment. The relative RMSE difference for the test period between GEP and ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 8, 10, 13 and 21%, respectively. The R2 values for GEP, ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 0.93, 0.84, 0.88, 0.86 and 0.81, respectively.
Accurate estimation of suspended sediment in rivers is very important for designing and operation of water resources projects. Sediment estimation by conventional methods like rating curves don’t provide accurate results. In this paper, gene expression programming (GEP) model which is an extension of genetic programming (GP) technique, was used to estimate suspended sediment in the river. The GEP results were compared with those of the adaptive neuro-fuzzy, neural networks and rating curve models. In this regard, the streamflow and suspended sediment data from Vanyar station that located on Aji-chay river in East- Azarbaijan province are used. The root mean square errors (RMSE) and determination coefficient (R2) statistics were used to evaluate the accuracy of the models. The results showed that the GEP model had better performance than other considered models in estimating suspended sediment. The relative RMSE difference for the test period between GEP and ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 8, 10, 13 and 21%, respectively. The R2 values for GEP, ANFIS-Grid Partitioning, ANFIS-Sub Clustering, ANN and rating curve methods were 0.93, 0.84, 0.88, 0.86 and 0.81, respectively.

کلیدواژه‌ها [English]

  • Gene expression programming-Rating curve-Neural networks-Neuro- Fuzzy-Suspended sediment.-