Abstract: Concerning the problem of low prediction accuracy because of multiple correlation factor in the traditional water quality prediction method, this paper introduces a partial least squares and support vector machine coupled method—the water quality prediction method（PLS-SVM）. Using partial least squares method extracts the variable component with strong influence, overcoming the information redundancy and reducing the dimension of support vectors. And using support vector machine modeling can be a better solution to the problem of high-dimensional nonlinear small samples. And using improved PSO algorithm to optimize SVM parameters reduces the parametric searching blindness. The results show that the coupled model fitting and forecasting accuracy is significantly better than the commonly used BP artificial neural networks and traditional SVM, can be better used in water quality prediction.

Key words: prediction of water quality, partial least squares regression, support vector machine, prediction model, Particle Swarm Optimization（PSO）

摘要： 针对传统水质预测方法中水质因子的多重相关性造成预测精度低的问题，提出了一种将偏最小二乘法和支持向量机相耦合的水质预测方法。利用偏最小二乘法提取对水质因子影响强的成分，从而克服了信息冗余问题，并降低了支持向量的维数。利用支持向量机建模可以较好地解决高维非线性小样本问题。同时利用改进的PSO算法优化SVM参数，减小参数搜索的盲目性。研究结果表明，本耦合模型的预测精度和运行效率明显优于常用的BP人工神经网络和传统的支持向量机，可以更好地应用于水质预测。

关键词: 水质预测, 偏最小二乘回归, 支持向量机, 预测模型, 粒子群优化算法