Abstract: In order to solve the problem of low prediction accuracy, the bad robustness of the traditional forecasting methods and the standard least squares support vector regression (LSSVR) in water quality prediction, the adaptive particle swarm optimization weighted least squares support vector regression (APSO-WLSSVR) model for water quality prediction was proposed. Different weights were set for various samples according to its different importance for the model. A weighted least squares support vector regression model (WLSSVR) was established, which could avoid low prediction accuracy of the standard LSSVR, and ignore the importance of the samples differences. The particle swarm optimization algorithm was adopted to optimize and choose the model parameters to reduce the blindness and the impact of human factors of the standard LSSVR trial method when obtaining the parameters. In order to verify the performance of the model, the water quality of intensive farming river crab in Yixing, Jiangsu Province, was predicted, which was also compared with other forecast methods. The results showed that the model prediction accuracy was obviously improved, and also had good robustness and generalization ability, which could met the practical needs of the intensive aquaculture water quality management.

Key words: least squares support vector regression, selfadaptive particle swarm optimization algorithm, water quality prediction, parameters optimization, intensive aquaculture