Abstract: To address the aggravated operational risks in weak distribution networks caused by the high penetration of distributed resources, this paper proposes a risk early warning scheme integrating multi-source data and a hybrid-driven model. By analyzing the fluctuation patterns of distributed resources under dual-peak load and extreme weather conditions, as well as the multi-dimensional risk coupling mechanism, a multi-indicator dynamic risk perception model based on the unscented Kalman filter (UKF) is established to realize real-time monitoring of system operating states and noise suppression. Meanwhile, an outage early warning model integrating G-ReliefF feature selection and an AdaBoost-enhanced decision tree (AdaBoost-DT) is constructed to optimize the input feature combination and improve classification prediction accuracy. By integrating multi-source data such as meteorological, measurement, and equipment data, a complete technical framework covering dynamic perception, feature optimization, and hierarchical early warning is developed. Case study results show that the proposed method can effectively improve the accuracy of state estimation, and the Ada-DT early warning model achieves an overall accuracy of 89.5%, with excellent performance in identifying medium- and high-risk events. This method provides reliable technical support for risk management and power supply resilience enhancement in weak distribution networks.