Abstract: The establishment of mathematical model based on slope monitoring data is an important method for slope deformation and stability analysis. Grey prediction, support vector machine(SVM) and related improvement models are the hot spots in slope deformation prediction. However, the form and application scope of single prediction model limit the utilization of monitoring data, resulting in low model accuracy and poor applicability. On the one hand, the inherent defects of the grey prediction model are difficult to be eliminated, and the dependence on historical data is strong, which leads to this model not suitable for many slope engineering. Although the SVM model shows great advantages in nonlinear fitting, its prediction accuracy for future trends is low. Aiming at the large error of a single prediction model, we propose an optimally combined PSO-SVR-NGM model based on the entropy weight method. The new model combines the high-precision variable-weight buffer NGM(1, 1, k, c)model and the PSO-SVR model. It can make up for the shortcomings of the single prediction model and improve prediction accuracy. We first improve the unbiased NGM(1, 1, k, c)model, through introducing the variable weight buffer operator λ and the background value weight coefficients η, κ to construct a new three-parameter variable weight buffer NGM(1, 1, k, c)model. And we use the improved particle swarm algorithm(PSO)to search and determine the best parameter combination, so that it can meet the requirements of fitting and prediction accuracy at the same time. The improved PSO integrates the maximum grey correlation degree and the minimum average relative error to reconstruct its fitness function. The support vector regression(SVR) is a kind of SVM. We also use PSO to search the best parameters of SVR model to achieve better modeling results. After obtaining these two high-precision models, we use the entropy method to weight the improved variable weight buffer NGM(1, 1, k, c)model and PSO-SVR model to establish a combined model. We use three slope projects with different deformation characteristics to verify the effectiveness of the combined model, by calculating and comparing average relative error(MRE), root mean square error(RMSE) and coefficient of determination(R²) of single model and combined model. The results show that, in contrast with single prediction model, the proposed combined model has lower error, higher fitting and prediction accuracy, better correlation with the original displacement data, can more truly reflect the slope deformation law, which make it has stronger engineering applicability. It is also found that the combined model combines the advantages of these single models, and make full use of the known monitoring information. Through a lot of research, we find that the combined model has strong applicability to displacement data with certain trend, which may be related to the single model selected. With the improvement of data processing technology and the optimization of models, the combined prediction model that considering various influencing factors of slope deformation will become the research direction in the future. At the same time, we find that the proposal and development of the combined model also promote the optimization and improvement of the single model, they complement each other and provide good ideas for solving practical landslide engineering problems.