Modern large-aperture alt-azimuth solar telescopes suffer from severe dynamic Instrumental Polarization （IP） caused by the relative rotation of optical elements during tracking. Resolving the inherent contradiction between the lon

g observation time occupied by dense hardware calibration and the high-precision requirements of polarization measurements remains a critical challenge. Under sparse calibration conditions， conventional numerical interpolation methods fail to accurately reconstruct the highly nonlinear evolution of IP in non-calibrated blind zones， while pure physical models struggle to account for unknown microscopic asymmetric errors. Therefore， the objective of this study is to propose and validate a Model-Guided Residual Interpolation （MGRI） hybrid calibration framework. By integrating a multi-parameter decoupled physical model with sparse calibration data， this paper aims to provide a high-fidelity， full-pointing-angle polarization correction solution that minimizes hardware calibration time while maximizing IP suppression accuracy for ground-based solar observation facilities.To achieve this objective， a comprehensive theoretical model and an experimental validation methodology are formulated. Based on Fresnel reflection theory and coordinate transformation matrices， a physical Mueller matrix model characterizing the dynamic IP of the telescope is established. Departing from the traditional ideal symmetric assumption， a fully decoupled 10-parameter physical model is proposed. This model assigns independent static polarization parameters （reflectance and initial phase） and dynamic phase evolution coefficients to each configuration mirror， capturing the inherent physical manufacturing asymmetries between the optical components. Building on this physical baseline， the MGRI algorithm is designed. The hybrid framework initially utilizes the 10-parameter model to absorb the high-frequency nonlinear polarization trends driven by geometric rotations. Subsequently， utilizing a limited number of sparse hardware calibration points， linear interpolation is performed on the smoothed residual matrix to compensate for low-frequency systematic deviations， finally yielding the reconstructed full-domain correction matrix. To rigorously valid

ate the proposed theoretical framework， an equivalent indoor double-reflector dynamic polarization platform is constructed using a 532 nm laser. A synchronous dual-camera detection scheme， combined with a six-state polarization demodulation algorithm， is employed to execute high-precision empirical Mueller matrix measurements， thereby establishing a strict operational baseline for evaluating the sparse calibration algorithms.Experimental results systematically demonstrate the performance and efficiency of the proposed MGRI framework. Initial baseline stability tests confirm that the temporal measurement repeatability of the experimental system achieves a Root Mean Square （RMS） precision superior to 1.46 × 10

-4

. Parameter inversion indicates that the 10-parameter fully decoupled model effectively absorbs the physical differences between the mirrors， reducing the global fitting root-mean-square error （RMS

fit

） to 2.71 × 10

-2

. In an applied operational context within the 20°–80° pointing range， traditional dense calibration mandates 31 pointing measurements and consumes approximately 2.5 to 5 hours. Conversely， the MGRI method requires only 4 discrete calibration points， demanding 20 to 40 minutes and effectively saving approximately 90% of calibration time. Evaluated under these sparse conditions， the MGRI framework successfully mitigates the truncation errors associated with standard discrete interpolation. Quantitatively， the global Mueller matrix RMSE is reduced from 0.0193 （pure linear interpolation） to 0.0070， corresponding to a 64.0% suppression in residual error. Alongside this， the mean magnitude of instrumental polarization is suppressed by 84.0%， decreasing from an uncorrected level of 0.0453 to 0.0073， while interrelated polarization crosstalk is also inhibited. Furthermore， comprehensive parameter-scanning strategy evaluation indicates that altering the specific selection of calibration points from a standard uniform distribution to an inward-contracted spatial topology （

e.g.， ［24°， 40°， 60°， 76°］） prevents boundary interpolation divergence. This structural optimization further lowers the global RMSE to 0.0059， yielding a 15.3% accuracy enhancement relative to standard fixed-endpoint selection schemes.The proposed MGRI framework provides a robust and highly efficient engineering paradigm that resolves the tension between dynamic IP correction accuracy and limited scientific observation windows in large-aperture solar telescopes. While the current methodology is validated on the core geometric rotation mechanism， future research will transition to on-sky operations utilizing the 1.8-meter Chinese Large Solar Telescope （CLST）. Real-condition optimizations will extend the physical model constraints to incorporate full-chain transmissive and reflective elements， as well as environmental thermal drift factors. Additionally， to accommodate broadband observational requirements， subsequent efforts will extend the wavelength applicability of the MGRI method by integrating empirical complex refractive index dispersion formulas or by executing independent calibration procedures across distinct targeted solar spectral lines.