A multi-target prediction model for dam seepage field

Prediction of dam behavior based on monitoring data is important for safety and emergency management. It crucial to analyze predict the seepage field. Different from mechanism-based physical models, machine learning models directly with high accuracy. However, current prediction are generally environmental variables single measurement point time series. Sometimes point-by-point modeling used obtain multi-point values. In order improve accuracy efficiency field, a novel multi-target model (MPM) proposed in which two deep methods integrated into one frame. The MPM can capture causal temporal features between target values, as well latent correlation different points at each moment. these parts put fully connected layers establish mapping relationship comprehensive feature vector outputs. Finally, trained framework feed-forward neural network using standard back propagation. not only describe variation pattern values change load time, but also reflect spatial distribution effectiveness verified by cases. commonly applicable other types fields besides