Comparison of Different Artificial Neural Networks Techniques and Autoregressive Models for Forecasting of PM 10

Abstract

Atmospheric particulate matter (PM₁₀) is one of the pollutant affecting human health significantly and has become a global issue. Data collected during three years in an urban area of Allahabad, Uttar Pradesh, India, are analysed and compared for 1-month ahead forecasting of PM₁₀ using four models: Levenberg algorithm (LM) based artificial neural network (ANN), radial basis function neural network (RBFNN), generalized regression neural network (GRNN) and autoregressive (AR) model. Measured PM₁₀ concentration are used as input to forecast the monthly averaged concentration of PM₁₀ for one month ahead. The mean absolute percentage error (MAPE) for AR models varies from 10.20% to 32.78% whereas MAPE for ANN, RBFNN and GRNN are found 4.75%, 13.40% and 11.43% respectively, showing ANN model with LM algorithm forecast PM₁₀ at one month ahead better than RBFNN, GRNN and AR models. In addition GRNN forecast is better than RBFNN with good accuracy. The average values of PM₁₀ for Bharat Yantra Nigam Allahabad and Ardali Bazar Varanasi are found to be 182.16 and 262 respectively, showing Varanasi has high value of PM₁₀. This study is useful for researcher working in forecasting of PM₁₀.