【摘要】 针对紫外-可见光谱法检测水质COD预测模型的精度低和收敛速度慢等问题,研究了一种基于粒子群算法联合最小二乘支持向量机(PSO_LSSVM)的水质检测COD预测模型优化方法,并引入主元分析(PCA)算法对模型输入光谱数据进行降维预处理,借以提高模型的收敛速度.结果表明,利用粒子群(PSO)算法收敛速度快和全局优化能力,优化了最小二乘支持向量机(LSSVM)模型的惩罚因子和核函数参数,避免了人为选择参数的盲目性,克服了传统LSSVM预测模型的精度较低、稳健性较差等缺点.通过以收敛时间、预测平均相对误差(MRE)和均方根误差(RMSE)为评价标准进行评估,输入样本经过PCA降维预处理的PSO_LSSVM模型的预测能力和输入样本未经过降维预处理的LSSVM模型与PSO_LSSVM模型进行了比较分析,输入样本经过PCA降维预处理的PSO_LSSVM模型预测效果最优,且此算法使用C语言实现,易于移植,这为紫外-可见光谱水质COD在线、实时性检测奠定了基础.更多还原

【Abstract】 There are some problems in the prediction model of the determination of water COD by using UV-visible spectroscopy,such as low precision and slow convergence speed. This paper studied an optimization method based on particle swarm optimization algorithm in combination with least squares support vector machine algorithm,and introduced the principal component analysis(PCA) algorithm to reduce the dimension of the input data in order to improve the convergence speed of the model. PSO had the ability of fast convergence speed and global optimization. The penalty factor and the kernel function parameter of the traditional LSSVM model had been optimized by PSO to overcome the blindness of selecting parameters manually and disadvantages of LSSVM prediction model of low precision,poor robustness. LSSVM model and PSO_LSSVM model had been established,which the dimensionality of input data had not been reduced. PSO_LSSVM prediction model had been established,which the dimensionality of input data had been reduced by PCA. Comparisons were conducted by computing the evaluation standard of the convergence time,average relative prediction error(MRE) and root mean square error(RMSE),and result were that the prediction ability of PSO_LSSVM model which using PCA superior than other two. The algorithm of the model were achieved by C language which more easy to transplant,and laid the foundation for real-time,online determination of Water COD by using UV –visible spectroscopy.更多还原