Improving the accuracy of a flood forecasting model by means of machine learning and chaos theory

Abstract

Monitoring natural environments is a challenging task on account of their hostile features. The use of wireless sensor networks (WSNs) for data collection is a feasible method since these domains lack any infrastructure. However, further studies are required to handle the data collected for a better modeling of behavior and thus make it possible to forecast impending disasters. In light of this, in this paper an analysis is conducted on the use of data gathered from urban rivers to forecast flooding with a view to reducing the damage it causes. The data were collected by means of a WSN in São Carlos, São Paulo State, Brazil, which gathered and processed data about the river level and rainfall by means of machine learning techniques and employing chaos theory to model the time series; this meant that the inputs of the machine learning technique were the time series gathered by the WSN modeled on the basis of the immersion theorem. The WSNs were deployed by our group in the city of São Carlos where there have been serious problems caused by floods. After the data interdependence had been established by the immersion theorem, the artificial neural networks were investigated to determine their degree of accuracy in the forecasting models.

How to Improve Fault Tolerance in Disaster Predictions: A Case Study about Flash Floods Using IoT, ML and Real Data

“…To determine the input and output parameters of the forecast models, created by the MLP, we used the procedure described in [ 25 ]. In this approach, the river level time series ( Figure 7 ) is analysed using Chaos Theory, more specifically the Takens’ Immersion Theorem [ 27 ], where the time series is unfolded to better extract its behavior and facilitating its analysis.…”

“…The output of the other model is the forecast of the average of the river level from to (5 minutos in the future—used for generate red alerts). These parameters were defined on the basis of previous research, as in [ 19 , 25 ].…”

Avaliação de Diferentes Arquiteturas de Redes Neurais Recorrentes na Previsão do Nível de Atividade de Abelhas

“…As RNAs com apenas uma camada são suficientes para resolver problemas linearmente separáveis, já as redes com mais camadas permitem resolver problemas não lineares. Sendo assim, as redes neurais têm sido empregadas em uma série de problemas de predição e interpolação de variáveis, como pode ser visto em [Furquim et al 2015], [Faiçal et al 2016] e [Carvalho et al 2017].…”

Development of Heavy Rain Damage Prediction Model Using Machine Learning Based on Big Data

“…Although the inclusion of socioeconomic factors besides weather factors led to some improvement in the prediction performance of these linear regression models, the nonlinear character of disasters and their damage scale present problems that cannot be solved by them. More recently, rapid advances in computing technology and data processing speed have led to the emergence of studies that apply big data and machine learning to disaster management [17,18]. e predominant approach in all these studies is to use just a handful of explanatory variables in a regression model to estimate the damage scale of disasters.…”