With the rapid development of distributed energy resources, accurate prediction of their output has become a critical component in the reliability assessment of distribution networks. To enhance the accuracy of such assessments, this paper proposes a reliability evaluation method for active distribution networks that integrates VMD-QRCNN-BiLSTM-based forecasting with DFT-MP-DBN modeling. First, the original time series data of wind power, solar power, and load are decomposed into intrinsic mode components using Variational Mode Decomposition (VMD). Then, a Quantile Regression Convolutional Neural Network (QRCNN) is employed to extract the temporal features, and a Bidirectional Long Short-Term Memory (BiLSTM) network is used to model each variable and generate accurate forecasts. These predicted values are then input into a Dynamic Fault Tree (DFT), where a Continuous-Time Markov Process (MP) is used to compute the state transition rate matrix. Finally, a Dynamic Bayesian Network (DBN) is applied to capture the temporal dependencies among system states and incorporate observed or control variables. Case studies based on the IEEE RBTS Bus 2 system show that the proposed method achieves superior reliability performance, with SAIFI, SAIDI, AENS, and ASAI values of 0.231 times/customer/year, 3.496 hours/customer/year, 17.465 kWh/year, and 99.943%, respectively—significantly outperforming traditional approaches. These results validate the effectiveness and advantages of the proposed method in improving the precision and efficiency of distribution network reliability assessments.