Trajectory prediction of non‑cooperative spacecraft constitutes an important component of space defense. Existing studies have primarily focused on configuration classification and parameter identification under the assumption that the motion conforms to a fixed configuration or that the control laws possess a known form, thereby enabling prediction of the future motion of non‑cooperative spacecraft. In this paper, a diffusion‑based time‑series prediction model is introduced, which is not constrained by the aforementioned assumptions. A pursuit–evasion trajectory dataset is constructed in a representative close‑range pursuit–evasion game scenario using reinforcement learning. The applicability of the Denoising Diffusion Probabilistic Model to the trajectory prediction task is then demonstrated. Finally, the effectiveness of the proposed model is validated through numerical simulations, in which comparisons with other time‑series prediction models are presented, and model predictive control is subsequently employed to intercept the non‑cooperative spacecraft based on the predicted trajectories.