The volatility and uncertainty brought about by the large-scale integration of distributed photovoltaic (PV) systems in the distribution network can lead to operational risks such as voltage violations, power flow congestion, and PV curtailment. To address this situation, this paper proposes a risk early warning and scheduling method based on imitation learning. Firstly, a risk early warning model for the distribution network is constructed based on imitation learning, and the optimization model is trained using historical operational data from the distribution network. Secondly, a real-time optimization model is built based on the same framework, with asynchronous inputs changed to synchronous inputs. Finally, considering the severity of risks in real-time, the real-time optimization model and the risk early warning model are superimposed to reuse historical experience. Simulation experiments demonstrate that this method can reduce the risks in the distribution network even when real-time measurement data is incomplete. Compared to methods like reinforcement learning, this approach more efficiently utilizes a small amount of risk scenario data, thereby more effectively reducing the probability of risk occurrence and improving robustness.