Abstract: In order to improve the pointing and tracking accuracy of the Coude focal plane of the altazimuth solar telescope. Firstly, this paper simulates the tracking error introduced on the Coude focal plane when the main optical axis, azimuth axis, height axis and racemic axis of the folded optical axis of the horizon solar telescope are not concentric. The complexity of the pointing and tracking error of the Coude focal plane is analyzed, which can't be solved by the usual night telescope focal pointing model. Then, the paper constructs pointing model based on support vector regression, and carries out an experiment on NVST. The experimental results show that before the model correction, the maximum pointing error of NVST on the Coude focal plane is 650.55 arcsec, RMS is 115.88 arcsec, and the 30-minute tracking error is 6.46 arcsec; after model correction, the maximum pointing error is 25.02 arcsec, the RMS value is 3.98 arcsec, and the 30-minute tracking error is 1.10 arcsec. It can be seen that the support vector regression modeling method based on machine learning can effectively improve the pointing and tracking accuracy of the Coude focal plane of altazimuth solar telescope.