<p>Doppler positioning based on Low Earth Orbit (LEO) communication satellite signals with abundant frequency resources and high power has been shown to be a reliable alternative for positioning and navigation in GNSS-denied environments. However, Doppler positioning encounters the technical bottleneck of low accuracy because of the influence of several error factors, including an unknown precise satellite ephemeris and clock information, and the absence of appropriate weighting optimization schemes for various satellite observations. To significantly diminish the performance degradation caused by multiple errors, we creatively propose a combined Doppler positioning method based on semi-parametric regression model with additional system parameters. The core lies in adding an equivalent clock error parameter to compensate for primary modeled errors, including along-track orbit and clock ambiguity errors, and introducing non-parametric components to mitigate unmodeled errors, including radial and cross-track orbit errors. Moreover, a new combined weighting scheme that takes Doppler position dilution of precision into account was designed to allocate the weights of Iridium NEXT/Orbcomm observations rationally and optimize the spatial geometry configuration effectively. Additionally, the extraction accuracy of the Doppler measurements was improved using the chip-z transform algorithm. Comparative experiments demonstrated the effectiveness of the proposed method, yielding a three-dimensional positioning root mean square error of 34.43&#xa0;m and a convergence time of 140&#xa0;s. The method successfully overcomes the performance loss caused by multiple errors, presenting a potentially viable positioning solution for elastic PNT systems.</p>