PCA-MLP神经网络模型在黄河宁夏段径流预测中的应用

Application of PCA - MLP neural network model in runoffff prediction in Ningxia Section of Yellow River

摘要:

为提高短时间尺度的月径流量预测效果并简化神经网络模型结构，将主成分分析（PCA）和多层感知器神经网络（MLP）相结合，构建 PCA-MLP 神经网络模型预测汛期月径流量。模型首先采用主成分分析法确定影响径流量的主要影响因子，再将主要影响因子数据输入 MLP 神经网络模型，预测月径流量数据。将宁夏青铜峡水文站 2010—2019 年汛期的月径流量和影响因子数据作为训练样本训练神经网络模型，以 2020—2022 年汛期月径流量和影响因子数据作为检验样本进行对比分析。预测结果表明：目前影响汛期径流量的因素主要是历史径流和气候特征，检验集预测结果确定性系数为 0.851，模型可为宁夏汛期月径流量预测提供相应指导。

Abstract:

To improve the prediction accuracy of monthly runoff volume on a short-time scale and to simplify thestructure of the neural network model，PCA-MLP neural network model combined with Principal ComponentAnalysis（ PCA） and Multi-Layer Perceptron（ MLP） neural network is proposed to forecast the monthly runoff volumeduring the fl ood season. The model fi rst employs PCA to determine the main infl uencing factors on runoff volume andthen inputs the data of the main infl uencing factors into MLP neural network model to predict monthly runoff volume.Monthly runoff volume and infl uencing factors data from Qingtongxia Hydrological Station in Ningxia during thefl ood season from 2010 to 2019 were used as training samples to train the neural network model，with data from 2020to 2022 used as testing samples for comparative analysis. The forecast results indicate that the factors infl uencing thefl ood season’s runoff volume are mainly historical runoff and climatic characteristics. The prediction results of thetesting set achieved a coeffi cient of determination of 0.851，demonstrating that the model can provide guidance for theprediction of monthly runoff volume in Ningxia during the fl ood season.