چکیده
کلید واژه ها
مقدمه
روش ها
شبکه های عصبی مصنوعی (ANN)
تبدیل موجک
تبدیل موجک افزایشی
تبدیل موجک گسسته (DWT)
حوزه مطالعه و داده ها
معیارهای بهره وری
توسعه مدل
مدل های مبتنی بر موجک
نتایج
بحث
نتیجه گیری
منابع
Abstract
Keywords
Introduction
Methods
Artifcal neural networks (ANN)
Wavelet transform
Additive wavelet transform
Discrete wavelet transform (DWT)
Study area and data
EfFiciency criteria
Model development
Wavelet‑based models
Results
Discussion
Conclusions
Declarations
References
چکیده
بهبود روشهای پیشبینی برای سریهای جریان یک وظیفه مهم برای برنامهریزی، مدیریت و فرآیند کشاورزی منابع آب است. این مطالعه توسعه و اثربخشی یک مدل ترکیبی جدید برای پیشبینی جریان جریان را نشان میدهد. در مطالعه حاضر، شبکههای عصبی مصنوعی (ANN) همراه با تبدیل موجک، یعنی تبدیل موجک افزایشی (AWT)، پیشنهاد شدهاند. تجزیه و تحلیل مقایسه ای تبدیل موجک گسسته (DWT) مبتنی بر ANN و تکنیک های ANN معمولی با روش پیشنهادی ارائه شد. تجزیه و تحلیل این مدلها با سری جریان ماهانه برای چهار ایستگاه در حوضه Çoruh، که در شمال شرقی ترکیه قرار دارد، انجام شد. الگوریتم آموزش پس انتشار منظم سازی بیزی برای بهینه سازی شبکه ANN استفاده شد. نتایج پیشبینیشده مدلها با ریشه میانگین مربعات خطا (RMSE)، معیار اطلاعات آکایک (AIC) و ضریب تعیین (R2) تجزیه و تحلیل شدند. نتایج بهدستآمده نشان داد که مدل ترکیبی پیشنهادی دقت قابلتوجهی در مقایسه با مدلهای دیگر نشان میدهد و بنابراین میتواند یک رویکرد جایگزین مفید برای پیشبینی مطالعات باشد.
توجه! این متن ترجمه ماشینی بوده و توسط مترجمین ای ترجمه، ترجمه نشده است.
Abstract
Improving predicting methods for streamfow series is an important task for the water resource planning, management, and agriculture process. This study demonstrates the development and efectiveness of a new hybrid model for streamfow predicting. In the present study, artifcial neural networks (ANNs) coupled with wavelet transform, namely Additive Wavelet Transform (AWT), are proposed. Comparative analyses of Discrete wavelet transform (DWT) based ANN and conventional ANN techniques with the proposed method were presented. The analysis of these models was performed with monthly streamfow series for four stations on the Çoruh Basin, which is located in northeastern Turkey. The Bayesian regularization backpropagation training algorithm was employed for the optimization of the ANN network. The predicted results of the models were analyzed by the root mean square error (RMSE), Akaike information criterion (AIC), and coefcient of determination (R2 ). The obtained revealed that the proposed hybrid model represents signifcant accuracy compared to other models, and thus it can be a useful alternative approach for predicting studies.
Introduction
Short- and long-term reliable prediction of river fows are vital for the management, planning and design of water resources. It is also important for various issues related to water resources such as food control, hydropower generation in drought periods, land use, agriculture and transport planning in rivers. A number of streamfow prediction methods have been proposed and employed in previous studies. They generally fall under statistical/stochastical based and conceptual/physically based methods. Statistical- or stochastic-based techniques include simple and multiple linear and nonlinear regression, autoregressive moving average (ARMA) models, ARMA with exogenous variables (ARMAX), and transfer function techniques (Salas et al. 2000). These techniques are generally called as black-box type of models. On the other hand, conceptual or physically based techniques that are usually employed for streamfow prediction depend on mathematical descriptions of the physical processes that take place in a watershed.
Conclusion
A new hybrid method based on coupling Additive Wavelet Transform (AWT) and artifcial neural networks (ANN) was presented for monthly streamfow predicting of four stations in the Çoruh basin which is situated in the northeast of Turkey. To totally shown the efectiveness of the suggested model, the AWT–ANN models were compared to single ANN models and DWT–ANN models. Using the Additive Wavelet Transform and discrete wavelet transform, original series were decomposed into subcomponents containing important information about the original data at diferent resolution levels, which were then used for predicting in ANN modelling.