基于PCA-PSO-GRU模型的重力拱坝诱导缝开合度预测

Prediction of Inducing Joint Opening and Closing Degree of Gravity Arch Dam Based on PCA-PSO-GRU Model

针对以往裂缝开合度时间序列数据预测中未有效利用众多实测温度数据,且各自变量之间存在多重相关性的问题,考虑主成分分析法(PCA)在处理多维数据上的优势以及门限循环单元(GRU)神经网络在处理复杂时间序列数据问题上的优势,构建了PCA-PSO-GRU组合预测模型。以某混凝土重力拱坝坝左诱导缝的开合度监测数据为样本,提取输入变量的主成分来降低输入数据的维度,然后进行模型训练和多步预测,采用平均绝对误差和均方根误差来评价模型的预测精度,并将预测结果与PSO-GRU、PCA-PSO-BP及传统的统计回归模型进行对比分析。结果表明,PCA-PSO-GRU组合预测模型在诱导缝时间序列数据预测方面具有更高的准确性,可为大坝诱导缝开合度评价提供一定的指导。

Aiming at the problem that many measured thermometer data are not effectively used in the previous prediction of crack opening and closing time series data, and there are multiple correlations between their variables, considering the advantages of principal component analysis(PCA) in dealing with multidimensional data and gate recurrent unit(GRU) neural network in dealing with complex time series data, this paper constructed the PCA-PSO-GRU combined prediction model. Taking the monitoring data of the opening and closing of the left inducing joint of a concrete gravity arch dam as a sample, the principal components of the input variables were extracted to reduce the dimension of the input data. And then the model training and multi-step prediction were carried out. The mean absolute error and root mean square error were used to evaluate the prediction accuracy of the model. The prediction results were compared with PSO-GRU, PCA-PSO-BP and the traditional statistical regression models. The results show that the PCA-PSO-GRU combined prediction model has higher accuracy in the prediction of inducing joint time series data, which can provide guidance for the evaluation of opening and closing degree of dam inducing joints.