Wireless sensor nodes for flood forecasting using artificial neural network

Abstract

The Pampanga River is considered as the fourth largest river basin in the Philippines. The lower basin of the river is one of the most frequently affected by flooding such as Masantol, Pampanga. At present, the Disaster Risk Reduction Management Office (DRRMO) uses a conventional way of water level measurement. The study aims to develop a real-time flood water level for medium and high risk areas and use these data for short forecasting. A standalone sensor station was developed with ultrasonic sensor, microcontroller, GSM module, and solar panel. Nonlinear autoregressive and Nonlinear autoregressive network with external input were used for modeling and prediction carried into 5 cases. Backpropagation technique, feed forward architecture, and optimized training algorithm known as Levenberg-Marquardt were used to develop the model in Matlab. The result with model prediction accuracy ranging 1.2e-3 to 3.12e-2 in terms of root mean square error (rmse), 9.97e-4 to 1.35e-2 mean absolute error (mae), 7.5e-1 to 1 correlation coefficient (r-value) for cases 1–3; and for cases 4–5, the result range from 1.3e-3 to 2.39e-2, 1.1e-3 to 2.11e-2, 7.618e-1 to 1 in terms of rmse, mae and r-value, respectively. This study may be a useful tool to DRRMO to provide early warning to the community.