基于TimeGAN-CNN-LSTM模型的河流水质预测研究

Research on River Water Quality Prediction Based on TimeGAN-CNN-LSTM Model

为精确预测河流水质中的铵离子(NH_(4)^(+))浓度,针对某公开水质数据进行了研究,提出了一种基于时间序列对抗生成网络(TimeGAN)、卷积神经网络(CNN)和长短期记忆(LSTM)网络的混合模型。使用TimeGAN对河流水质历史数据进行数据增强,生成合成时间序列数据;采用CNN对输入的数据进行特征提取,并通过全连接层将数据输入到LSTM中得到预测值,从而建立TimeGANCNN-LSTM河流水质预测模型。试验结果表明,模型预测效果良好,其平均绝对误差(MAE)、均方根误差(RMSE)和决定系数(R^(2))分别为0.07、0.08和0.97,比CNN-LSTM模型分别提高了45.45%、47.06%和19.75%,比LSTM模型分别提高了50%、50%和21.25%。TimeGAN-CNN-LSTM既解决了训练模型时数据不充分的问题,又能够充分提取水质数据在时间和空间上的特征,具有较高的应用价值。

To accurately predict the concentration of ammonium ions(NH_(4)^(+))in river water quality,a hybrid model based on time series adversarial generative network(TimeGAN),convolutional neural network(CNN)and long short-term memory(LSTM)network is proposed for a public water quality data.TimeGAN is used to enhance the historical data of river water quality and generate synthetic time series data;CNN is used to extract features from the input data and input the data into LSTM through fully connected layer to get the prediction value,to establish TimeGAN-CNN-LSTM river water quality prediction model.The experimental results show that the model has good prediction effect,and its mean absolute error(MAE),root mean square error(RMSE)and coefficient of determination(R^(2))are 0.07,0.08 and 0.97,respectively,which are 45.45%,47.06%and 19.75%better than the CNN-LSTM model,and 50%,50%and 21.25%better than the LSTM model,respectively.TimeGAN-CNN-LSTM not only solves the problem of insufficient data when training the model,but also can fully extract the characteristics of water quality data in time and space,which has high application value.