Ozone Level Forecasting: Time Series Analysis Using Multi-Layer Perceptron (MLP) Artificial Neural Networks Trained with Bayesian Regulation Back-propagation

Concerns of tropospheric ozone effects on human life motivate decision makers to forecast ozone concentration specially in metropolitan and tropic regions. Due to nonlinearity of ozone variations, using neural networks methods is considered a proper tool for forecasting air quality. This paper proposes a multi-layer perceptron (MLP) model trained with Bayesian Regulation Back-Propagation (BRP) algorithm for forecasting maximum daily ozone concentrations in Tucson city. The performance of proposed model trained with BRP algorithm shows better results than other training algorithms in which R=0.9627 (Pearson correlation coefficient) for measured and predicted data and the coordination between their errors and the standard normal distribution curve are reasons of our claim. Achieved results approve that proposed model has a fair forecasting of ozone concentrations time series.