توسعه مدل‌هاي هيبريدي شبكه عصبي مصنوعي- الگوريتم تكاملي جهت پيش‌بيني عدد فرود جريان در كانال‌هاي باز در مدل‌سازي انتقال رسوب

Development of a Hybrid ANN-evolutionary Algorithms Models to Predict the Froude Number in Open Channel Flows in Modeling of Sediment Transport

ﺗﺨﻤﯿﻦ ﻣﻘﺪار ﻋﺪد ﻓﺮود در ﻣﻬﻨﺪﺳﯽ آب ﺑﺴﯿﺎر ﻣﻬﻢ اﺳﺖ. از ﻃﺮﻓﯽ اﺳﺘﻔﺎده از ﻣﺪلﻫﺎي ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ ﺑﻪ دﻟﯿﻞ اﯾﻨﮑﻪ ﻧﯿﺎز ﺑﻪ اﻃﻼﻋﺎت ﺳﺎﺧﺘﻤﺎﻧﯽ ﺳﺎﻣﺎﻧﻪﻫﺎ ﻧﺪارﻧﺪ، ﻣﻮردﺗﻮﺟﻪ ﻗﺮار ﮔﺮﻓﺘﻪ اﻧﺪ. در ﭘﮋوﻫﺶ ﺣﺎﺿﺮ از ﻣﺪل ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ (ANN) اﺳﺘﻔﺎدهﺷﺪه و از اﻟﮕﻮرﯾﺘﻢﻫﺎي ﺗﮑﺎﻣﻠﯽ ازدﺣﺎم ذرات(PSO)، اﻟﮕﻮرﯾﺘﻢ ﮐﺮم ﺷﺐﺗﺎب (FA) و اﻟﮕﻮرﯾﺘﻢ ﺗﮑﺎﻣﻠﯽ ﺗﻔﺎﺿﻠﯽ (DE) ﺟﻬﺖ ﺑﻬﺒﻮد ﻧﺘﺎﯾﺞ ANNدر ﭘﯿﺶﺑﯿﻨﯽ ﻣﻘﺪار ﻋﺪد ﻓﺮود ﺑﻬﺮه ﮔﺮﻓﺘﻪ ﺷﺪ. ﭘﺎراﻣﺘﺮﻫﺎي ﻣﺆﺛﺮ ﺑﺮ ﭘﯿﺶﺑﯿﻨﯽ ﻋﺪد ﻓﺮود ﭘﺎراﻣﺘﺮﻫﺎي رﺳﻮب ﺣﺠﻤﯽ، اﻧﺪازه داﻧﻪﻫﺎ ، ﻋﺎﻣﻞ اﺻﻄﮑﺎك و ﻧﺴﺒﺖ اﻧﺪازه داﻧﻪﻫﺎ ﺑﻪ ﺷﻌﺎع ﻫﯿ ﺪروﻟﯿﮑﯽ ﺑﻮدﻧﺪ ﮐﻪ ﺑﺎﻫﻢ ﺗﺮﮐﯿﺐ ﺷﺪﻧﺪ و اﻟﮕﻮﻫﺎي ورودي ﻣﺨﺘﻠﻔﯽ ﺑﻪﻋﻨﻮان ورودي ﺑﻪ ﻣﺪلﻫﺎ ﺗﺪو ﯾﻦ ﺷﺪ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد اﻟﮕﻮﯾﯽ ﮐﻪ ﺑﺎ ﻣﺸﺎرﮐﺖ ﻫﻤﻪ ﭘﺎراﻣﺘﺮﻫﺎي ورودي ﺑﺎﺷﺪ ﺑﯿﺶﺗﺮﯾﻦ دﻗﺖ را در ﭘﯿﺶﺑﯿﻨﯽ ﻋﺪد ﻓﺮود دارد. اﺳﺘﻔﺎده از اﻟﮕﻮرﯾﺘﻢﻫﺎي ﺗﮑﺎﻣﻠﯽ ﭘﯿﺸﻨﻬﺎدي ﺗﻮاﻧﺴﺖ دﻗﺖ ﻣﺪل ANN را ﺑﻪﺻﻮرت ﻗﺎﺑﻞﺗﻮﺟﻬﯽ ﺑﻬﺒﻮد ﺑﺨﺸﺪ. در ﺑﯿﻦ اﻟﮕﻮرﯾﺘﻢﻫﺎي ﻣﻮرد ﺑﺮرﺳﯽ، اﻟﮕﻮرﯾﺘﻢﻫﺎي FA و PSO در ﺑﯿﺶﺗﺮ ﻣﻌﯿﺎرﻫﺎي ارزﯾﺎﺑﯽ دﻗﺖ ﻣﻨﺎﺳﺒﯽ داﺷﺘﻨﺪ. در ﺣﺎﻟﺖ ﮐﻠﯽ اﺳﺘﻔﺎده از اﻟﮕﻮرﯾﺘﻢ ﻫﺎي ﺗﮑﺎﻣﻠﯽ، دﻗﺖ ﻣﺪل ANN را ﺑﻪ ﻣﯿﺰان 0/07 m ،0/89 mو 0/1 ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺮاي ﻣﻌﯿﺎر ارزﯾﺎﺑﯽ MAPE ،RMSE و R2ﺑﻬﺒﻮد ﺑﺨﺸﯿﺪ. ﻧﺘﺎﯾﺞ ﻣﺪلﻫﺎي ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﻣﻌﺎدﻻت ﺗﺠﺮﺑﯽ ﻧﯿﺰ اﺧﺘﻼف ﭼﺸﻢﮔﯿﺮي داﺷﺘﻨﺪ و ﺣﺎﮐﯽ از دﻗﺖ ﺑﺴﯿﺎر ﻣﻨﺎﺳﺐ ﻣﺪلﻫﺎي ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ ﺑﻮدﻧﺪ. ﺑﻪ ﻫﻤﯿﻦ ﺳﺒﺐ اﺳﺘﻔﺎده از ﻣﺪلﻫﺎي ﯾﺎدﮔﯿﺮي ﻣﺎﺷﯿﻦ ﺑﻪﺟﺎي اﺳﺘﻔﺎده از ﻣﻌﺎدﻻت ﺗﺠﺮﺑﯽ در ﺑﺮآورد ﻣﻘﺪار ﻋﺪد ﻓﺮود ﭘﯿﺸﻨﻬﺎد ﻣﯽﮔﺮدد.

In water engineering, estimating Froude number’s value is very important. The applying of machine learning models has received much attention because it does not require system building information. For this purpose, in the present study, the artificial neural network (ANN) model was used. To improve the ANN results, evolutionary algorithms including Particle Swarm Algorithm (PSO), Firefly Algorithm (FA) and Differential Evolution Algorithm (DE) have been used to predict the amount of Froude number. The parameters affecting the prediction of Froude number were λ, d / R, Dgr, Cv that combined, and different input patterns have been developed as input to the models. The results showed that the model involving all the affecting input parameters has the highest accuracy in predicting the Froude number. The proposed evolutionary algorithms significantly improved the accuracy of the ANN model. Among the algorithms, FA and PSO algorithms had good accuracy in most evaluation criteria. In general, the applying of evolutionary algorithms improved the accuracy of the ANN model by 0.89 m, 0.07 m and 0.1. The results of machine learning models were also significantly different from the experimental equations and indicated the high accuracy of machine learning models. Therefore, it is recommended to use them instead of using experimental equations in estimating the value of the Froude number.