Improvement of the multilayer perceptron for air quality modelling through an adaptive learning scheme

Multilayer perceptron (MLP), normally trained by the offline backpropagation algorithm, could not adapt to the changing air quality system and subsequently underperforms. To improve this, the extended Kalman filter is adopted into the learning algorithm to build a time-varying multilayer perceptron (TVMLP) in this study. Application of the TVMLP to model the daily averaged concentration of the respirable suspended particulates with aerodynamic diameter of not more than [email protected]?m (PM"1"0) in Macau shows statistically significant improvement on the performance indicators over the MLP counterpart. In addition, the adaptive learning algorithm could also address explicitly the uncertainty of the prediction so that confidence intervals can be provided. More importantly, the adaptiveness of the TVMLP gives prediction improvement on the region of higher particulate concentrations that the public concerns.

[1]  Hsin-Chung Lu,et al.  Meteorologically adjusted trends of daily maximum ozone concentrations in Taipei, Taiwan , 2005 .

[2]  Andrew Harvey,et al.  Forecasting, Structural Time Series Models and the Kalman Filter , 1990 .

[3]  Harri Niska,et al.  Forecasting daily pollen concentrations using data-driven modeling methods in Thessaloniki, Greece , 2010 .

[4]  Fabrício P. Härter,et al.  Multilayer Perceptron Neural Network in a Data Assimilation Scenario , 2010 .

[5]  K. Mok,et al.  Effects of meteorological conditions on PM10 concentrations - A study in Macau , 2005, Environmental monitoring and assessment.

[6]  Hassiba Nemmour,et al.  Comparison between object- and pixel-level approaches for change detection in multispectral images by using neural networks , 2004, SPIE Remote Sensing.

[7]  OPTIMIZING THE PERFORMANCE OF KALMAN FILTER BASED STATISTICAL TIME-VARYING AIR QUALITY MODELS , 2010 .

[8]  Matthias Plöchl.  Neural network approach for modelling ammonia emission after manure application on the field , 2001 .

[9]  D Student.  Selection of noise parameters for Kalman filter , 2007 .

[10]  A. Comrie.  Comparing Neural Networks and Regression Models for Ozone Forecasting , 1997 .

[11]  Morton J. Canty,et al.  Boosting a fast neural network for supervised land cover classification , 2009, Comput. Geosci..

[12]  Ka-Veng Yuen,et al.  Bayesian Methods for Updating Dynamic Models , 2011 .

[13]  M. C. Hubbard,et al.  A Comparison of Nonlinear Regression and Neural Network Models for Ground-Level Ozone Forecasting , 2000, Journal of the Air & Waste Management Association.

[14]  Hassiba Nemmour,et al.  Kalman filtering as a multilayer perceptron training algorithm for detecting changes in remotely sensed imagery , 2003, IGARSS 2003. 2003 IEEE International Geoscience and Remote Sensing Symposium. Proceedings (IEEE Cat. No.03CH37477).

[15]  Sharad Singhal,et al.  Training feed-forward networks with the extended Kalman algorithm , 1989, International Conference on Acoustics, Speech, and Signal Processing,.

[16]  Ka-Veng Yuen,et al.  Prediction of daily averaged PM10 concentrations by statistical time-varying model , 2009 .

[17]  Andrew Harvey,et al.  Forecasting, Structural Time Series Models and the Kalman Filter. , 1991 .

[18]  Ka-Veng Yuen,et al.  On the complexity of artificial neural networks for smart structures monitoring , 2006 .

[19]  James L. Beck,et al.  Updating Properties of Nonlinear Dynamical Systems with Uncertain Input , 2003 .

[20]  C. Willmott.  Some Comments on the Evaluation of Model Performance , 1982 .

[21]  Haroldo F. de Campos Velho,et al.  New approach to applying neural network in nonlinear dynamic model , 2008 .

[22]  J. Fung,et al.  Effects of urbanization on the land sea breeze circulation over the Pearl River Delta region in winter , 2010 .

[23]  Sik Chung Tam,et al.  Short-term prediction of SO2 concentration in Macau with artificial neural networks , 1998 .

[24]  阮家荣,et al.  Selection of noise parameters for Kalman filter , 2007 .

[25]  Goutami Chattopadhyay,et al.  Autoregressive forecast of monthly total ozone concentration: A neurocomputing approach , 2009, Comput. Geosci..

[26]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[27]  K. Yuen.  Bayesian Methods for Structural Dynamics and Civil Engineering , 2010 .