Abstract: Factors such as rock type, structural characteristics of the rock body, hydrogeology, natural environment, and mining activities easily affect the deformation of open pit slopes, resulting in a high degree of temporal correlation, time-varying, high-dimensional, and non-linear characteristics of the slope deformation monitoring data. Aiming at the problem of traditional slope deformation prediction models being unable to exploit the back-and-forth dependence of monitoring data series, a multi-step high steep slope deformation prediction model with a fusion network of bi-directional long and short-term memory network (Bi-LSTM) and self-attention mechanism (SA) was proposed, which takes the advantages of Bi-LSTM network mining the pre and post dependence of monitoring data and SA network analyzing the correlation between monitoring data. The effective prediction of multi-step high and steep slope deformation was realized. The results show that, under the same input conditions, compared with the prediction results of the BP neural network, LSTM model and Bi-LSTM model, the overall prediction error of the Bi-LSTM-SA fusion model for the deformation prediction results of multi-step high slope in three monitoring directions is smaller. The prediction results of Bi-LSTM-SA fusion model are closer to the measured results. The Bi-LSTM-SA fusion model has better prediction performance, stability and robustness.