Development of a Revised Multi‐Layer Perceptron Model for Dam Inflow Prediction

Hyeon Seok Choi, Joong Hoon Kim, Eui Hoon Lee, Sun Kwon Yoon

Research output: Contribution to journalArticlepeer-review

1 Citation (Scopus)

Abstract

It is necessary to predict dam inflow in advance for flood prevention and stable dam op-erations. Although predictive models using deep learning are increasingly studied, these existing studies have merely applied the models or adapted the model structure. In this study, data prepro-cessing and machine learning algorithms were improved to increase the accuracy of the predictive model. Data preprocessing was divided into two types: The learning method, which distinguishes between peak and off seasons, and the data normalization method. To search for a global solution, the model algorithm was improved by adding a random search algorithm to the gradient descent of the Multi‐Layer Perceptron (MLP) method. This revised model was applied to the Soyang Dam Basin in South Korea, and deep learning‐based discharge prediction was performed using historical data from 2004 to 2021. Data preprocessing improved the accuracy by up to 61.5%, and the revised model improved the accuracy by up to 40.3%. With the improved algorithm, the accuracy of dam inflow predictions increased to 89.4%. Based on these results, stable dam operation is possible through more accurate inflow predictions.

Original languageEnglish
Article number1878
JournalWater (Switzerland)
Volume14
Issue number12
DOIs
Publication statusPublished - 2022 Jun 1

Keywords

  • dam inflow prediction
  • data normalization
  • machine learning
  • multi‐layer perceptron
  • seasonal division
  • weights update algorithm

ASJC Scopus subject areas

  • Geography, Planning and Development
  • Biochemistry
  • Aquatic Science
  • Water Science and Technology

Fingerprint

Dive into the research topics of 'Development of a Revised Multi‐Layer Perceptron Model for Dam Inflow Prediction'. Together they form a unique fingerprint.

Cite this