COMPARISON OF LONG-SHORT TERM MEMORY NETWORKS WITH DIFFERENT OPTIMIZATIONS IN RIVER FLOW PREDICTION AND THE EFFECT OF SINGULAR SPECTRUM ANALYSIS

Home » Creative » COMPARISON OF LONG-SHORT TERM MEMORY NETWORKS WITH DIFFERENT OPTIMIZATIONS IN RIVER FLOW PREDICTION AND THE EFFECT OF SINGULAR SPECTRUM ANALYSIS

...to create

THE BEST SLIDER

with no compromises!

COMPARISON OF LONG-SHORT TERM MEMORY NETWORKS WITH DIFFERENT OPTIMIZATIONS IN RIVER FLOW PREDICTION AND THE EFFECT OF SINGULAR SPECTRUM ANALYSIS

Submitted By:

Asst. Prof. Dr. Huseyin Cagan Kilinc 

Istanbul Esenyurt University, Engineering Faculty, Civil Engineering Department, Istanbul, Türkey

drabidnib@gmail.com

Article

Periodic river flow measurements are required to ensure sustainable water resources. For this, different estimation methods are required.  In this study, Deep Learning (DL) and Aksu River flows were estimated by LSTM (Long-Short Term Memory) neural network, which is one of the Artificial Intelligence methods. In the study, the data belonging to Başpınar Flow Measurement Station (FMS) (D20A002) on Aksu River between 2000-2019 were used as input for analysis. In addition, the performance effect of Single Spectrum Analysis (TSA) on LSTM was examined. Adam, Adamax and AdaGrad algorithms were applied to the TSA-LSTM model. The most accurate estimation model has been determined by comparing the estimate and actual values. It has been observed that the Adamax optimizer provides the best performance in flow estimation. TSA-LSTM model coefficient (R2) determination was found to be 0.9851 during the test phase. When the obtained results were examined, it was seen that the TSA-LSTM model gave better results in estimating flow studies. 

Keywords: LSTM, Singular Spectrum, Stream Flow​ 

___________________________________________________________________

Disclaimer: Any opinions and views expressed in this submission are the opinions and views of the person who has submitted the article, and are not the views of or endorsed by the Scientific and Technical Research Association (STRA). The accuracy of the content should not be relied upon and should be independently verified with primary sources of information. The person submitting the article does not necessarily be the author of the article. The Scientific and Technical Research Association (STRA) shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, about, or arising out of the use of the content. For any issues or any reporting, write an email to convener@eurasiaresearch.info

___________________________________________________________________

2021-11-16T05:24:53+00:00
Translate »