In the regeneration mode, precise control of the Diesel Oxidation Catalyst (DOC) outlet temperature is crucial for the complete combustion of carbon Particulate Matter (PM) in the subsequent Diesel Particulate Filter (DPF) and the effective conversion of Nitrogen Oxides (NOx) in the Selective Catalytic Reduction (SCR). The temperature elevation process of the DOC involves a series of intricate physicochemical reactions characterized by high nonlinearity, substantial time delays, and uncertainties. These factors render effective and stable control of the DOC outlet temperature challenging. To address these issues, this study proposes an approach based on Long Short-Term Memory (LSTM) neural networks for Model Predictive Control (MPC), emphasizing precise control of the Diesel Oxidation Catalyst’s outlet temperature during the regeneration mode. To tackle the system’s nonlinear characteristics, LSTM is employed to construct a predictive model for the outlet temperature of the Diesel Oxidation Catalyst, thereby enhancing prediction accuracy. Simultaneously, model predictive control is applied to mitigate the significant time delays inherent in the system. The gradient descent algorithm is utilized within a rolling optimization cycle to optimize the objective function, enabling the rapid determination of the control law. To validate the performance of the proposed control strategy, tracking performance and disturbance rejection tests are conducted. Simulation results demonstrate that, compared to the traditional Proportional Integral Derivative (PID) controller, this control strategy exhibits superior tracking performance and disturbance rejection capabilities. In the regeneration mode, the adoption of this control strategy enables more effective and precise control of the Diesel Oxidation Catalyst’s outlet temperature.