Chi Dianwei, Huang Qi, Liu Lizhen
School of Artificial Intelligence, Yantai Institute of Technology, Yantai 264003, China.
Key Laboratory of Poyang Lake Wetland and Watershed Research, Ministry of Education, School of Geography and Environment, Jiangxi Normal University, Nanchang 330022, China.
Entropy (Basel). 2022 Mar 25;24(4):457. doi: 10.3390/e24040457.
Dissolved oxygen concentration has the characteristics of nonlinearity, time series and instability, which increase the difficulty of accurate prediction. In order to accurately predict the dissolved oxygen concentration in the dish-shaped lakes in Poyang Lake of Jiangxi Province, China, a dissolved oxygen concentration prediction model, based on wavelet transform (WT)-based denoising, maximal information coefficient (MIC)-based feature selection, and the gated recurrent unit (GRU), was proposed for this study. In experiments, the proposed model showed good prediction performance, achieving a root-mean-square error (RMSE) of 0.087 mg/L, a mean absolute percentage error (MAPE) of 0.723%, and a coefficient of determination (R2) as high as 0.998. It shows that the prediction model based on the combination of the wavelet transform and the GRU has a relatively high prediction accuracy and a better fitting effect. The model proposed in this study can provide a reference for protecting this type of lake-water body and the restoration of missing values in lake water quality monitoring data.
溶解氧浓度具有非线性、时间序列性和不稳定性等特点，这增加了准确预测的难度。为了准确预测中国江西省鄱阳湖碟形湖中的溶解氧浓度，本研究提出了一种基于小波变换（WT）去噪、基于最大信息系数（MIC）的特征选择以及门控循环单元（GRU）的溶解氧浓度预测模型。在实验中，所提出的模型表现出良好的预测性能，均方根误差（RMSE）为0.087mg/L，平均绝对百分比误差（MAPE）为0.723%，决定系数（R2）高达0.998。结果表明，基于小波变换和GRU相结合的预测模型具有较高的预测精度和较好的拟合效果。本研究提出的模型可为保护此类湖泊水体以及湖泊水质监测数据缺失值的恢复提供参考。
