ARIMA-SVR组合方法在水质预测中的应用

针对复杂水域水质变化机理难以掌握、水质预测建模困难且预测精度低的问题,将时间序列分析方法与机器学习方法引入水质预测领域,提出了基于差分自回归移动平均(ARIMA)与支持向量回归(SVR)组合模型的水质预测方法。数据经过预处理后先由ARIMA模型对其进行线性拟合,然后通过SVR模型预测残差以补偿其中的非线性变化。选择巢湖水域2004—2015年间的pH和溶解氧监测数据作为试验样本,通过Hodrick-Prescott(HP)滤波方法分析,结果表明2组数据具有不同的趋势特性和波动特性。根据精度评价指标对比分析模型的预测效果,结果表明组合模型预测精度显著提高,pH和溶解氧预测值与观测值间的相关系数均达到了0.99,均方根误差分别为0.20和0.61,平均绝对百分比误差分别为2.2%和6.6%。本研究所建立的组合预测方法具有较高的预测精度和较强的泛化能力,适用于复杂水域的水质预测。

Water Quality Prediction Using an ARIMA-SVR Hybrid Model

The research aims to tackle the difficulty in obtaining the mechanism of water quality change in complex waters and modeling water quality prediction as well as the low accuracy of prediction. A water quality prediction method integrating Autoregressive Integrated Moving Average (ARIMA) model and Support Vector Machine Regression (SVR) model is established. After pretreatment, the data are linearly fitted by ARIMA model, and then the residual is predicted by SVR model to compensate for the non-linear change. The monitoring data of pH value and dissolved oxygen in Chaohu Lake basin from 2004 to 2015 are selected as experimental samples. Analysis by HP filtering method shows that the two sets of data have different trends and fluctuation characteristics. The prediction effect of the hybrid model is analyzed through comparative study on evaluation indicators. The results conclude that the present model has high accuracy with the correlation coefficients between predicted values and observed values of pH value and dissolved oxygen both reaching 0.99, the root mean square error 0.20 and 0.61, and the mean absolute percentage error 2.2% and 6.6%, respectively. The ARIMA-SVR hybrid model is of high prediction accuracy and strong generalization ability, suitable for water quality prediction in complex waters.