پيش بيني ابعاد آبشستگي در حوضچه ي استغراق سرريزهاي سرويس با ‌روش‌هاي هوش مصنوعي

Prediction of scour dimension in the Plunge Pools below Outlet Bucket with Artificial intelligence method

پيش بيني دقيق ابعاد حفره آبشستگي در پايين دست سازه هاي هيدروليكي از جمله سرريزهاي جامي شكل، به دليل پيچيدگي ‌هاي ناشي از بررسي همه جانبه و همزمان جريان حاوي آب و رسوب و اعمال كليه متغيرهاي مؤثر در پديده آبشستگي به سادگي ميسر نمي ‌باشد. ابعاد حفره آبشستگي اغلب با استفاده از معادلات تجربي تعيين ميگردد كه اين روابط در محدوده خاصي از داده ‌ها و شرايط آزمايش پاسخگو مي باشد. از آنجايي كه ساخت مدل فيزيكي مشكلات و محدوديت هايي به همراه دارد و معمولا در تعيين نگاشت ميان پارامتر هاي مؤثر بر آبشستگي نمي توان اثر دقيق همه پارامترها را در نظر گرفت، لذا در مقاله حاضر بهينه يابي ابعاد حفره آبشستگي براي مجموعه ‌اي از مشاهده‌ ها آزمايشگاهي محققان قبلي طراحي شده است. در اين تحقيق ازشبكه عصبي مصنوعي و سيستم تطبيقي عصبي- فازي بهره گيري شده و نتايج آن با معادله حاصل از روش رگرسيون غيرخطي بين داده ‌هاي مشابه و همچنين فرمول هاي تجربي پيش بيني حداكثر عمق آبشستگي مقايسه شده است. نتايج اين تحقيق حاكي از دقت و برتري قابل ملاحظه سيستم تطبيقي عصبي - فازي با حداكثر خطاي 5/2 درصد نسبت به نتايج حاصل از مدل شبكه عصبي و معادله رگرسيون غيرخطي و فرمول تجربي با حداكثر خطا به ترتيب 10/38، 12/42 و 14/05 درصد ميباشد.

Accurate prediction of sediment scour hole dimensions downstream of hydraulic structures, e.g. the outlet bucket, is a complex and not straight forward engineering problem encountered worldwide. Because of the complexities of the study, its comprehensive, simultaneous including water flow, sediment and applying all of the effective variables involved in scouring it is not easy possible. Dimensions of a scour hole are usually determined by empirical equations which their validation is limited by experimental conditions. As constructing physical models has its own difficulty, determining of scour hole parameters has been applied in this paper for a collection of previous experimental studies. Two artificial intelligence techniques (ANN & ANFIS) are used and the results are compared with empirical equation for maximum scour holes using nonlinear regression method. Artificial Neural Network (ANN) simply represents interconnection of neurons, each of which carries out the task of combining the input, determining its strength by comparing the combination and finding out the result. On the other hand, ANFIS is a hybrid scheme which uses the learning capability of the ANN to derive the fuzzy rules with membership functions. The results showed that maximum error caused by applying ANFIS techniques in estimating scour hole dimensions was 5.2 percent while the error in neural network model was 10.38 percent. The significance of different parameters was discussed and a simple, innovative formula was proposed. This formula is an interesting tool for the engineering community due to its preferences for estimating the parameters of complex phenomena like erosion procedures. It has been established that scour estimations could be improved if soft computation is used in place of the traditional formulae.