A time-series forecasting approach based on KPCA-LSSVM for lake water pollution

The time-series forecasting of lake water pollution is a very important and difficult issue of any lake water pollution control system. The time-series data of lake water pollution are huge, high-dimensional and nonlinear, so the information mining of it is difficult. To realize the data mining and forecasting for time-series data of lake water pollution efficiently, an improved prediction model based on the least squares support vector machine (LSSVM) is presented in this paper. To reduce the dimension of samples, the kernel principal component analysis (KPCA) method is used to extract the feature information, which contains the principal components of samples. Then the LSSVM method is used to set up the prediction model and the parameters in this model are optimized by the genetic algorithm. Finally, the proposed prediction model is applied in water pollution time-series data forecasting experiments of Taihu Lake. The experimental results show that the proposed approach has some better performances than the general LSSVM methods, such as the good predictive accuracy and stability in the time-series forecasting of lake water pollution.