Abstract
Due to signal loss and instrument failure, the arch dam monitoring system fails to obtain a complete deformation monitoring sequence. Therefore, the framework to fill in missing sequence of arch dam deformation is of significant importance. Affected by environmental features like water level, temperature, and aging, the deformation patterns exhibit distinctive regional characteristics. Existing methods for filling missing deformation sequence only consider neighborhood deformation, neglecting environmental features. Considering environmental features and neighborhood deformation comprehensively, an Enhanced Deep Belief Network-Light Gradient Boosting Machine (EN-DBN-LightGBM) is proposed to fill the missing sequence of arch dam deformation. Specifically, based on a feature set comprising environmental features and neighborhood deformation, a Multi-objective Grey Wolf Optimization (MOGWO) is employed to conduct optimal information selection. This process utilizes criteria of maximum correlation and minimum redundancy, resulting in a Pareto-optimal subset. Relief-F is utilized to assess the contribution of each feature subset and select optimal one. Deep Belief Network-Light Gradient Boosting Machine (DBN-LightGBM) is implemented to extract the deformation features and influence factors, subsequently filling in the missing values of the deformation sequence. The engineering case study demonstrates that EN-DBN-LightGBM can effectively fill missing deformation sequence with high accuracy and generalization capability.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call