MHA-ConvLSTM Dam Deformation Prediction Model Considering Environmental Volume Lag Effect

The construction of a reasonable and reliable deformation prediction model is great practical significance for dam safety assessment risk decision-making. Traditional models are susceptible to interference from redundant features, weak generalization ability, lack interpretation. Based on this, that considers the lag effect environmental quantities proposed. first constructs new influence factor based plain HST through quantization algorithm. Secondly, attention memory capacity improved by introducing multi-head mechanism features long-time domain factor, finally, extracted dynamic transferred ConvLSTM learning, training, prediction. results simulation tests measured data an active show introduction not only improves interpretation but also makes more accurate, it can improve evaluation indexes such as RMSE 50%, nMAPE 40%, R2 10% compared with traditional model. combined capable mining hidden has deeper picture overall peak local extremes data, which provides way thinking