نوع مقاله : مقاله پژوهشی
نویسندگان
1 کارشناس‌ارشد مهندسی منابع آب دانشگاه علوم کشاورزی و منابع طبیعی گرگان
2 استاد، گروه مهندسی آب دانشگاه تبریز
3 استاد دانشکدة منابع طبیعی و محیط زیست دانشگاه فردوسی مشهد
4 دانشیار گروه مهندسی آب دانشگاه علوم کشاورزی و منابع طبیعی گرگان
چکیده
برآورد و پیشبینی بارش اهمیت ویژهای دارد. به دلیل نبود قطعیت، هیچ یک از مدلهای آماری و مفهومی نتوانستهاند به منزلة یک مدل برتر در الگوسازی دقیق بارش شناخته شوند. اخیراً، به کاربردِ موجک در آنالیز سیگنالها و سریهای زمانی در هیدرولوژی توجه شده است. در این تحقیق، سیگنال بارندگی با استفاده از موجک مادر منتخب تجزیه شد و دادههای به‌دست‌آمده با دو روش برازش معادلات مستقیم و هیبرید عصبی- موجکی برای پیشبینی استفاده شد. روش مذکور در پیشبینی بارندگیِ ماهانة 33 سال ایستگاه زرین‌گل از سال آبی 1354 ـ 1355 تا 1386 ـ 1387 به‌کار گرفته شد و نتایج با یکدیگر مقایسه شد. نتایج نشان داد تجزیة سیگنال با موجک به طور قابل ملاحظهای موجب افزایش همبستگی میان دادههای مشاهداتی و محاسباتی می‌شود و سیگنالِ بارندگی با دقت بیشتری پیشبینی می‌شود؛ به طوری که در روش مستقیم میزان R2برابر با 74/0 و در روش هیبرید عصبی- موجکی در بهترین حالت برای چهار سطح تجزیه برابر 95/0 است. این نتیجه قدرت موجک در سادهسازی سیگنال و افزایش دقت پیشبینی دادههای کاملاً تصادفی بارندگی را در منطقة مورد نظر تأیید می‌کند. ضمن آنکه، معنیدار نبودن تست Fدر سطح 90 درصد و بالاتر تأیید دیگری بر این مطلب است.
عنوان مقاله [English]
Prediction of precipitation applying Wavelet and ANN Model
نویسندگان [English]
- Parivash Toufani 1
- Ahmad Fakheri fard 2
- Abolfazl Mosaedi 3
- AmirAhmad Dehghani 4
1 Former M.Sc. Student, Water Resources Engineering, Gorgan University of Agricultural Sciences and Natural Resources
2 Professor, Department of Water Engineering, Tabriz University
3 Professor, Faculty of Natural Resources and Environment, Ferdowsi University of Mashhad
4 Associate Professor, Department of Water Engineering, Gorgan University of Agricultural Sciences and Natural Resources
چکیده [English]
Prediction of Precipitation is very important. Regarding to the non- linear relationships and uncertainty of models, there is no superior and persuasive model among various proposed models to simulate precise precipitation and its amount. Wavelet is one of the novel and very effective methods in time series and signals analyzing, that has been considered in the field of hydrology in recent years. In this research, precipitation signal has been decomposed via selected mother wavelet, and then the resulted data are used by fitting direct equations and nero-wavelet hybrid in order to anticipate the precipitation. The mentioned method was applied in Zarringol station (Iran) to predict monthly precipitation since 1975-76 until 2007-2008 for the period of 33 years. The results showed that, by decomposing signal via wavelet, the correlation among observed and calculated data were significantly increased, and the precision of prediction was improved. So that in direct method the value of R2 is equal to 0.74 and in nero-wavelet hybrid in the best case and for 4 level decomposition the value is equal to 0.95. This shows the capability of wavelet in simplifying of signal and intensification of accuracy random data in prediction of precipitation. Moreover, the meaningless F test, verifies the mentioned object.
Keywords: Precipitation, prediction, Signal, Wavelet theory, Nero-wavelet hybrid, Zarringol.
Prediction of Precipitation is very important. Regarding to the non- linear relationships and uncertainty of models, there is no superior and persuasive model among various proposed models to simulate precise precipitation and its amount. Wavelet is one of the novel and very effective methods in time series and signals analyzing, that has been considered in the field of hydrology in recent years. In this research, precipitation signal has been decomposed via selected mother wavelet, and then the resulted data are used by fitting direct equations and nero-wavelet hybrid in order to anticipate the precipitation. The mentioned method was applied in Zarringol station (Iran) to predict monthly precipitation since 1975-76 until 2007-2008 for the period of 33 years. The results showed that, by decomposing signal via wavelet, the correlation among observed and calculated data were significantly increased, and the precision of prediction was improved. So that in direct method the value of R2 is equal to 0.74 and in nero-wavelet hybrid in the best case and for 4 level decomposition the value is equal to 0.95. This shows the capability of wavelet in simplifying of signal and intensification of accuracy random data in prediction of precipitation. Moreover, the meaningless F test, verifies the mentioned object.
)