Home // ADAPTIVE 2019, The Eleventh International Conference on Adaptive and Self-Adaptive Systems and Applications // View article

Flood Prediction Through Artificial Neural Networks

Authors:
Pascal Goymann
Dirk Herrling
Andreas Rausch

Keywords: Machine learning; Neural networks; PCA; Flood prediction.

Abstract:
In this project, a system was developed, which allows a flood prediction based on given data sets for a level measuring station in the Goslar area for a period of four hours. First, existing neural networks, which were developed during a seminar at the TU Clausthal, were extended with the help of the framework Tensorflow and investigated whether larger water level values and further flood scenarios allow good qualitative prognoses. Furthermore, the influencing factors of possible floods were identified based on past scenarios. In addition to gauge and precipitation measuring stations in the immediate vicinity of Goslar, weather data from the Institute of Electrical Information Technology (IEI), which have been available every 15 minutes since 2003, were also taken into consideration. These data sets were processed and evaluated accordingly, so that a qualitative prediction can be made for exact water gauge heights. In addition, in order to reduce the training time, a dimension extraction was performed using a Principal Component Analysis (PCA), in which main components were identified and the data set examined for patterns in order to determine the possibility of a dimension reduction. In order to transfer the neural network to further scenarios, a prediction was made for the area of Bad Harzburg, where two measuring stations with additional weather data were used as inputs.

Pages: 56 to 62

Copyright: Copyright (c) IARIA, 2019

Publication date: May 5, 2019

Published in: conference

ISSN: 2308-4146

ISBN: 978-1-61208-706-1

Location: Venice, Italy

Dates: from May 5, 2019 to May 9, 2019