<p>After a space robot captures an out-of-control spacecraft, it is necessary to identify the inertia matrix of the combined spacecraft formed by the two to effectively control its overall attitude. This paper proposes a novel physics-informed parameter identification neural network (PIPINN) to accurately identify the inertia matrix of combined spacecraft in the presence of complex angular rate measurement noise. Unlike parameter identification methods based on physics-informed neural network (PINN), the PIPINN utilizes the strong fitting capabilities of the deep neural network to directly learn the mapping relationship between the state-control sequences of the combined spacecraft and the inertia matrix, which enables the PIPINN to maintain high identification accuracy in environments with complex measurement noise. Moreover, after a single training session, the PIPINN can identify any inertia matrix within a predefined parameter space without the need for repeated training, thereby enhancing the operational efficiency of space robots in orbit. More importantly, the PIPINN integrates physics constraints based on the principle of PINN, which allows it to achieve high identification accuracy and strong generalization capabilities even when training data are scarce. Numerical simulations validate the effectiveness of the proposed PIPINN under various types of measurement noise. The results demonstrate that the proposed PIPINN offers superior identification performance compared to other inertia matrix identification methods, and can effectively enhance the identification accuracy of the inertia matrix for the combined spacecraft even with the small training set.</p>