Optimized multi-output LSSVR displacement monitoring model for super high arch dams based on dimensionality reduction of measured dam temperature field

Establishing a reasonable displacement health monitoring model is essential for determining the safety of super high arch dams. However, previous studies only focus on a single measurement point, which leads to low efficiency and accuracy problems in evaluating the overall status of the dam. To solve this issue, a multi-output least square support vector regression (MLSSVR) model that can evaluate and forecast multiple monitoring points is developed in this manuscript. The optimal parameters of the model are determined by particle swarm optimization (PSO), in order to improve the precision and generalization ability of the model. On the other hand, the kernel principal component algorithm (KPCA) is introduced to extract the principal temperature components to construct the model, which brings advantages in revealing the actual temperature displacement accurately compared to the harmonic function and ambient temperature, as well as overcoming the multicollinearity caused by the superabundant of temperature factors. The feasibility and accuracy of the proposed model are tested with long-term measured displacements of a super high arch dam. The results show that the proposed model is superior to the multiple linear regression (MLR) and support vector regression (SVR), based on the hydrostatic -seasonal-time (HST) and hydrostatic-temperature-time (HTT) models. It also has outstanding medium and long-term predictive capacity, which provides a new approach for dam displacement safety monitoring.

Automated summary

Establishing a reasonable displacement health monitoring model is crucial for ensuring the safety of super high arch dams. This study proposes a multi-output least square support vector regression model capable of evaluating and predicting multiple monitoring points. The optimal parameters of the model are determined using particle swarm optimization. The introduction of kernel principal component algorithm helps extract the main temperature components accurately and overcome multicollinearity issues. Testing with real data shows that the proposed model outperforms other models and has outstanding medium and long-term predictive capacity.