The magnitude and distribution of residual stress field inside thin-walled parts is a critical factor influencing machining deformation. Traditional methods relying on offline data struggle to rapidly and accurately predict evolutionary state of residual stress fields, due to the time-varying and nonlinear characteristics in machining. This paper presents an online prediction method for residual stress in machining thin-walled parts based on deep learning. The multi-channel vector model is proposed to incorporate geometric, physical, and process information as input for deep learning models. A deep learning framework utilizing the IncepU-net network is developed and trained using both experimental and finite element simulation data. Results indicate a mean error of 6.2 MPa compared to experimental values, with an overall prediction time of 0.17 s. The proposed method can online predict the magnitude and distribution of residual stress field in machining, which offers cost-effectiveness and strong generalization capabilities.