Concrete dam structures respond to various influencing factors with complex nonlinear characteristics and notable time lags. Deformation serves as a crucial monitoring metric, providing a direct indication of the structural response of these dams. An effective deformation analysis and prediction model is essential for accurately assessing the health of concrete dam structures. Current deformation prediction models have limitations in simulating time-delay effects. This study introduces time-shifted correlation coefficients and time-delayed transfer entropy to analyze the direction of information transmission and the time delays among environmental temperature, dam body temperature, and deformation monitoring variables. A methodology is proposed to determine the dimensions of temperature factors and their respective time delays. Utilizing a long short-term memory (LSTM) neural network integrated with Dropout regularization, a concrete dam deformation prediction model that accounts for the time delay effect of environmental temperature is developed. The results demonstrate that the proposed deformation prediction model offers superior fitting accuracy and predictive capability, effectively elucidating how environmental and dam body temperatures influence dam deformation.