为充分挖掘水质数据在短时震荡中的变化特征,提升预测模型的精度,提出一种基于VMD(变分模态分解)、TCN(卷积时间神经网络)及GRU(门控循环单元)组成的混合水质预测模型,采用VMD-TCN-GRU模型对汾河水库出水口高锰酸盐指数进行预测,并与此类研究中常见的SVR(支持向量回归)、LSTM(长短期记忆神经网络)、TCN和CNN-LSTM(卷积神经网络-长短期记忆神经网络)这 4 种模型预测结果对比表明:VMD-TCN-GRU模型能更好挖掘水质数据在短时震荡过程中的特征信息,提升水质预测精度;VMD-TCN-GRU模型的MAE(平均绝对误差)、RMSE(均方根误差)下降,R2(确定系数)提高,其MAE、RMSE、R2 分别为0.055 3、0.071 7、0.935 1;其预测性能优越,预测精度更高且拥有更强的泛化能力,可以应用于汾河水质预测.
In order to fully excavate the variation characteristics of water quality data in short-term shocks and improve the accuracy of pre-diction model,we proposed a VMD-TCN-GRU water quality prediction model based on Variational Mode Decomposition(VMD),Temporal Convolutional Network(TCN),and Gated Recurrent Unit(GRU).The VMD-TCN-GRU model was applied to predict the permanganate in-dex at the outlet of Fenhe River Reservoir.Comparative analysis with commonly used models in water quality prediction,including Support Vector Regression(SVR),Long Short-Term Memory(LSTM),TCN,and Convolutional Neural Network and Long Short-Term Memary(CNN-LSTM),demonstrates that the VMD-TCN-GRU model excels in extracting features during short-term oscillations in water quality data,identifying actual patterns of variation,facilitating comprehensive model learning,and thereby improving prediction accuracy.The VMD-TCN-GRU model achieves a Mean Absolute Error(MAE)of 0.055 3,Root Mean Square Error(RMSE)of 0.071 7,and a determination co-efficient R2 of 0.935 1,indicating high predictive accuracy and strong generalization capabilities.This model can be effectively applied to wa-ter quality prediction tasks in the Fenhe River.