Abstract: For ultra-rapid orbits provided by the Global Navigation Satellite System (GNSS), the key parameters, accuracy and timeliness, must be taken into consideration for real-time and near real-time navigation and positioning applications. With reference to the low accuracy in observed orbits (last 3 hours), an improved model based on the orbit dilution of precision (DOP) values is proposed by building the function models between DOP values and the orbit accuracy. Firstly, the prediction model of the DOP values is set based on observations and Akaike information criterion. Secondly, the function model between orbit accuracy and DOP values is built.At last, the predicted DOP values with highly precision are taken into orbit correction model to improve the observed ultra-rapid orbit accuracy. With the comparison experiments, it is found that there is a correlation between the DOP values of orbit parameter and its precision, and the variation trend of DOP values can be fitted based on polynomial model. Furthermore, to verify the availability of improved model, two experiment schemes were designed from the aspect of the length of observations and prediction models. In addition, there are no significant differences of orbit correction for different function models. With 10-day orbit determination experiments, the results show that the observed ultra-rapid orbit errors, generated by insufficient observations, can be corrected by 12.4%-22.0% for the last 3 hours of the observed orbits. Thus, the observed ultra-orbit correction model of our proposed is meaningful and can improve the ultra-rapid orbit accuracy of GNSS analysis center.