基于PSO-WSVR的短期水质预测模型研究-《郑州大学学报(工学版)》

文章信息/Info

Title:: Study of Short-term Water Quality Prediction Model Based on PSo-wSVR

作者:: 徐龙琴; 刘双印.; 广东海洋大学信息学院,广东湛江,524088, 广东海洋大学信息学院,广东湛江524088;中国农业大学现代精细农业系统集成研究教育部重点实验室,北京100083;中国农业大学北京市农业物联网工程技术研究中心,北京100083

Author(s):: XU Long-qin1; LIU Shuang-vin1; 2; 3; 1.School of Information ，Guangdong Ocean Universit , Zhanjiang 524025，China; 2.Key Laboratory of Modern Precision Agri-culture System Integration Research of Ministry of Education , China Agriculural University ,Beijing 100083 , China;3.Beijing En-gineering Research Center for Agricultural Internet of Things，China Agricultural University，Beijing 100083,China

Keywords:: water quality prediction; weighted support vector regression; particle swarm optimization; parameters optimization

摘要:: 针对传统方法很难建立精确的非线性水质预测模型的情况,提出了基于粒子群优化加权支持向量回归机(PSO-WSVR)的水质短期预测模型,在建模过程中,根据各样本重要性的差异,给各个样本的惩罚系数赋予不同权重,改进了标准支持向量回归机算法,克服了标准支持向量回归算法因不同样本均采用相同权重造成预测精度低的问题,并采用粒子群优化算法对加权支持向量回归机参数组合进行自适应优化,模型收敛速度明显加快.运用PSO-WSVR模型对江苏宜兴市集约化河蟹养殖池塘水质进行预测,与标准支持向量回归机和BP神经网络对比分析.结果表明,该模型性能可靠、泛化能力强,预测精度高,为集约化水产养殖水质短期预测提供了一种新思路.

Abstract:: In view of the difficulty in establishing precise nonlinear water quality forecast model using the tradition-al method,the paper proposed the short-term forecast model of the water quality using the weighted support vectorregression machine based on the particle swarm optimization ( PSO-WSVR). According to the importance of thesamples are significantly different，the authors proposed the penalty coefficient for the every sample to differentweighted values and improved the standard support vector regression algorithm to avoid different samples using thesame weight,which may cause the low prediction accuracy.The parameters combinations of the weighted supportvector regression machine were adaptively optimized using particle swarm optimization algorithm, which the conver-gence rate could be sped up significantly.The water quality of crab intensive aquaculture in Yixing,Jiangsu waspredicted using PSO-WSVR model. Compared with forecasting result of the standard support vector regression andBP neural network,the forecasting result of PSO-WSVR have reliable performance, generalization ability,and highforecast precision, so it provides a new way for short -term forecasting of the water quality in intensive aquaculture.