[Purposes] In order to solve the problem that the measured sample data of earth pressure balance of shield machines are unevenly distributed, resulting in low accuracy of the earth pressure prediction model, mechanism model data were introduced to construct an earth pressure prediction model of shield machines based on the integration of data and mechanism. [Methods] The mechanism model of the earth pressure of shield machines was established based on geotechnical mechanics. The earth pressure prediction model was established by integrating the mechanism model data and the measured data using several multi-fidelity surrogate models. Then, the complementary advantages of the measured data and the mechanism model data were fused by comparing the optimal earth pressure prediction model of shield machines. Finally, a multi-objective collaborative optimization strategy for the earth pressure of shield machines was proposed, which realized the collaborative optimization of the earth pressure of shield machines in different directions. [Findings] The prediction results of the earth pressure of shield machines show that compared with the prediction model based only on measured data, the introduction of the mechanism model data greatly improves the prediction accuracy, and the optimal coefficient of determination of prediction results is increased from 0.941 to 0.977. The optimization test results of the earth pressure of shield machines in different directions show that the overall root mean square error of the optimized earth pressure is significantly reduced, with a decrease of about 14.47%. [Conclusions] Introducing mechanism model data can improve the accuracy of the earth pressure prediction model of shield machines. The variation range of earth pressure in different directions has been greatly reduced after collaborative optimization, which provides new methods for earth pressure prediction and collaborative optimization of shield machines.