Slope deformation prediction is a core aspect of safety monitoring in geotechnical engineering. To address the limitations of traditional models in handling the nonlinear, non-stationary, and noise-prone characteristics of slope settlement data, this study proposes a novel SSA-FPA-SVR-GRU (SFSG) model integrating Singular Spectrum Analysis (SSA), Flower Pollination Algorithm (FPA), Support Vector Regression (SVR), and Gated Recurrent Unit (GRU). The methodology involves: firstly, adaptively decomposing the original settlement series into trend and fluctuation components using SSA combined with Wavelet Transform (WT) for denoising; secondly, constructing a hybrid prediction framework where SVR and GRU respectively model the trend and fluctuation components before integrating their outputs; finally, employing FPA for automated hyperparameter optimization. Validated against field settlement data from the 1216 platform slope at Bulian Gou Coal Mine in Inner Mongolia, the SFSG model demonstrated superior performance compared to four other intelligent algorithm-enhanced SSA-SVR-GRU variants, achieving MAPE of 1.3313%, RMSE of 0.0750mm, MAE of 0.0626mm, and R2 of 97.45%. Consistent superiority was verified through monitoring points P2 and P3 (both R2>97%). The developed SFSG model effectively handles complex characteristics of slope settlement time series, providing reliable technical support for precise prediction and safety early-warning in slope engineering.