Sea water level forecasting using soft computing techniques: comparative performance of neural networks and genetic programming
Daiane G. Faller1; Luana Borato2; Adrieni F. Andrade1; Eduardo Marone2; Nelson F. F Ebecken1
1 COPPE/UFRJ; 2 UFPR
doi:10.20906/CPS/CILAMCE2015-0632
Resumo
Artificial neural network (ANN)s and genetic programming (GP) have been applied to solve a range of different problems in coastal and ocean areas in order to develop evolve an estimation or forecasting of environmental and weather parameters, structural loads and responses. The purpose of this study is to experiment differents ANN and genetic programming structures to predict sea level variations due to a combination of astronomical and meteorological forces. Conventional tidal forecasting has been based on harmonic analysis using the least squares method to determine harmonic parameters. Although, a large number of parameters is required for the prediction of a long-term tidal level with harmonic analysis. The tidal water level forecast improvement may be conducted using harmonic analysis and meteorological variables. The use of tools capable to train and translate the combined influence of meteorological and astronomical forcing to predict sea level variations and reduce the margin of error can be a powerful alternative. In this work, the sea level forecasts improvement in the Estuarine Complex of Paranaguá (Paraná State, Brazil) is expected giving the knowledge of meteorological and calculated astronomical tide (by the harmonic analysis) behavior to the ANN and GP, at the same time period, obtaining results as close to real observed values as possible. To perform this, the models will be trained by using hourly time series of atmospheric pressure, wind, and harmonically derived tides for a specific year (to be defined) as input data and hourly time series of measured tides as output data. The data will be obtained from Paranaguá meteorological and tide gauge station. If missing data were found in time series, interpolation based on FFT analysis will be used. For water level forecast, the meteorological time series will be used as input data, and the resulting water level outputs will be compared with the water level measurements in the same period. In both methods, different configurations will be tested to id
Palavras-chave: Artificial networks; Genetic programming; Sea level variations; harmonic variables