A real time multivariate robust regression based flood prediction model using polynomial approximation for wireless sensor network based flood forecasting systems

Abstract

The paper introduces a statistical model to be used in wireless sensor network (WSN) for forecasting floods in rivers using simple and uncomplicated calculations and provide a reliable and timely warning to the people who may be affected. The statistical process used for this real time prediction uses linear robust multiple variable regression method to provide simplicity in cost and feature, and yet efficiency is speed, power consumption and prediction accuracy which is the prime goal of any design algorithm. This model is theoretically independent of the number of parameters, which may be varied according to practical needs. When increasing, the water level trend is approximated using a polynomial and its nature is used to predict when the water level may cross the flood line in future. We have simulated the comparison of predicted water level with the actual level in a time interval, around and below the flood line. The accuracy of prediction above flood line is of no value in real life and but a data above flood line is shown in our simulation results for the sake of continuity and logical justification of the algorithm. © 2012 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.