Neural modeling of relative air humidity

The objective of the present study was to use artificial neural networks for the estimation and prediction of relative air humidity. Neural modeling was carried out using MATLAB and STATISTICA software. Relative air humidity was predicted with a feedforward multilayer perceptron artificial neural network with time delay. The backpropagation algorithm was used for ANN training in MATLAB. The forecasting horizon was one time interval (3 h). The forecast was extended to 48 h (16 measurements) by re-introducing a newly estimated value as an input. The mean relative prediction error for the horizon adopted was 2.1%, and the Pearson r correlation coefficient −0.972. Estimation was performed using a Generalized Regression Neural Network (GRNN) model. This model estimated relative air humidity at the highest value of the Pearson r correlation coefficient −1.000. The GRNN developed with MATLAB tools did not show overfitting, although 100% of the empirical data were used to generate its topology.