Forecasting of ozone level in time series using MLP model with a novel hybrid training algorithm

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

39 Scopus Citations
View graph of relations



Original languageEnglish
Pages (from-to)913-924
Journal / PublicationAtmospheric Environment
Issue number5
Publication statusPublished - Feb 2006


As far as the impact of tropospheric ozone (O3) on human heath and plant life are concerned, forecasting its daily maximum level is of great importance in Hong Kong as well as other metropolises in the world. This paper proposed a multi-layer perceptron (MLP) model with a novel hybrid training method to perform the forecasting task. The training method synergistically couples a stochastic particle swarm optimization (PSO) algorithm and a deterministic Levenberg-Marquardt (LM) algorithm, which aims at exploiting the advantage of both. The performance of such a hybrid model is further compared with ones obtained by the MLP model trained individually by these two training methods mentioned above. Based on original data collected from two typical monitoring sites with different O3 formation and transportation mechanism, the simulation results show that the hybrid model is more robust and efficient than the other two models by not only producing good results during non-episodes but also providing better consistency with the original data during episodes. © 2005 Elsevier Ltd. All rights reserved.

Research Area(s)

  • Daily maximum ozone level, Levenberg-Marquardt algorithm, Multi-layer perceptron model, Particle swarm optimization algorithm, Performance comparison