Abstract:

Evolution features research and multi-step prediction of water quality time series have important significance for making reasonable and applicable environmental protection measures. But the structural complexity and non-stationary is the bottleneck of auto-regressive model application to water quality series prediction. Focusing on these problems, Markov switching theory was introduced to combined with autoregressive model for water quality series simulating and predicting. Markov switching auto-regressive model(MS-AR) is suitable for the analysis of non-stationary time series with variable structure, and has a good adaptability to heteroscedastic time series. In the modeling the water quality time series of Guoheqiao section, the Box-Cox transformation was  used to transform the sample data to make it approximately Gaussian, and then MS-AR model was applied to the analysis and prediction of the water quality sequence. The result showed that MS-AR model identified the two evolution patterns of the series, and the prediction of MS-AR was better than the classical AR model, which indicates that MS-AR model has a good prospect in water quality sequence dynamic structure analysis and its prediction. This model can automatically identify the regime switching of a time series’ evolution process.

Key words: water quality time series, Markov theory, regime switching, Box-Cox transformation, autoregressive model, water pollution