基于优化变分模态分解的大坝变形组合预测模型:以丰满水电站为例

Combined Prediction Model of Dam Deformation Based on Optimal Variational Mode Decomposition:Taking Fengman Hydropower Station for Example

大坝的变形通常受到多种因素的影响,监测数据表现出一定的非平稳性和随机性,为提高大坝变形预测的精度,提出了基于优化变分模态分解的大坝变形组合预测模型。该模型首先采用粒子群优化算法(particle swarm optimization,PSO)寻找变分模态分解(variational mode decomposition,VMD)的最优超参数,然后将大坝变形分解为趋势项、周期项和随机项分量。针对分解后各分量的时序特点,采用时域卷积网络(temporal convolutional network,TCN)和长短时记忆网络(long short-term memory,LSTM)进行组合预测,对各分量预测值重构加成得到最终预测值。以实际工程数据为例,采用平均绝对误差(mean absolute error,MAE),均方误差(mean square error,MSE)和平均绝对百分比误差(mean absolute percentage error,MAPE)等指标对模型量化评估,并与单一的预测模型进行比较。结果表明:本文提出的基于优化变分模态分解的大坝变形组合预测模型精度更高,可以有效提取大坝变形数据中隐含的信息特征,降低大坝变形时序数据的非平稳性,具有较高推广应用价值,为精准预测大坝变形提供了借鉴和指导。

Dam deformation is usually affected by many factors,and monitoring data show certain non-stationarity and randomness.To improve the accuracy of dam deformation prediction,a combined prediction model of dam deformation based on optimized variational mode decomposition was proposed.The model first used PSO(particle swarm optimization) to find the optimal hyperparameters of VMD(variational mode decomposition).Then the dam deformation was decomposed into trend item,periodic item,and random item components.According to the timing characteristics of each component after decomposition,a combination of TCN(temporal convolutional network) and LSTM(long short-term memory network) was used for prediction.The final predicted value was obtained by reconstructing and adding the predicted values of each component.Taking actual engineering data as an example,the model was quantitatively evaluated using indicators such as MAE(mean absolute error),MSE(mean square error),and MAPE(mean absolute percentage error).It was also compared with a single forecasting model.The results show that the dam deformation combination prediction model based on optimized variational mode decomposition proposed in this paper has higher accuracy.It can effectively extract the hidden information features in the dam deformation data,and reduce the non-stationarity of the dam deformation time series data.It has a high value of popularization and application and provides reference and guidance for accurately predicting dam deformation.