Towards Low Cost Nowcasting of Flash Floods in Mauritius: Assessing the Effectiveness of a Combined Approach Involving Wireless Sensor Network and Machine Learning
MRC-RUN-1708 Final Report.pdf
Version: 2,
Uploaded by: MRIC Publisher 01,
Size: 6.9 MB,
Date Uploaded:
2 June 2020
No associated document available
No related project available
No graphical abstract available
Reference (Identifier)
MRR-20-00026
Title
Towards Low Cost Nowcasting of Flash Floods in Mauritius: Assessing the Effectiveness of a Combined Approach Involving Wireless Sensor Network and Machine Learning
This study aims at assessing the effectiveness of a low-cost prototype to nowcast flash floods in the Mauritian context. A system based on simulation, using Wireless Sensor Network and modern Machine Learning techniques has been designed and implemented for the prediction of flash floods. The Wireless Sensor Network component reads and collects different features from river flow and rainfall monitoring. It has been tested through simulation and is estimated to have a relatively low unit cost. The Machine Learning component of the system is based on a deep learning approach with the implementation of the Recurrent Neural Network (RNN), and has been trained and tested using simulated datasets. The efficiency of the model has been further optimised with the application of Genetic Algorithm, and experiments demonstrate a relatively low error in predictions. The results achieved in this study cannot be generalized for the Mauritian context at large, but serve as an approach for the development of an automated flash flood nowcasting system based on rainfall and water flow monitoring in rivers/canals.