Abstract: In determining the displacement monitoring indicators of a concrete arch dam, the traditional method did not consider the uncertainty of dam material parameters. This paper develops a new method for the determination of these indicators based on the results of stochastic parameter back analysis by combining the Bayesian inference theory, particle swarm optimization, and a long short-term memory network. We use the particle swarm optimization to optimize the long short-term memory network as the surrogate model, and derive a joint likelihood function using the displacement values measured at different monitoring points and the ones calculated by the surrogate model. With the prior distribution of material parameters given, we can use the Bayesian inference to calculate their posterior distribution. This distribution is used to calculate the dam displacement component due to water pressure; then the displacement monitoring index of the dam can be determined using displacement measurements. Analysis of engineering cases shows the Bayesian inference method gives accurate posterior results of the material parameters. The monitoring indicators, determined using the mixed method, fully utilize the results of stochastic back analysis and have better warning effects.

Key words: arch dam, stochastic inversion, Bayesian inference, monitoring indicator