Abstract: Constructing a reasonable online monitoring model is an important guarantee for real-time control on dam safety.Aiming at the problems of conventional LSTM model such as easily affected by multi-parameters combination, weak generalization ability of optimal parameters and difficult manual selection of parameters, the influence of key parameters such as learning rate, block size, maximum number of iterations and number of hidden layer units on the accuracy of dam safety online monitoring model were deeply analyzed.An improved particle swarm optimization algorithm(IPSO) integrating nonlinear inertia weight, shrinkage factor and Cauchy disturbance term was proposed, and the IPSO-LSTM model for dam safety monitoring was constructed by coupling with LSTM model.The engineering verification showed that this model can automatically search for the optimal parameters, has high accuracy and strong robustness, and is suitable for dam safety monitoring data sequences of different types and lengths.The error can be reduced by at least30% compared with the conventional LSTM model with artificial parameters.Relevant experiences can provide technical support for online monitoring of dam operation safety.