To address the high concealment of pre-meter leakage locations in low-voltage distribution networks-as well as the limitations of traditional manual inspection methods that rely heavily on maintenance personnel experience and struggle with intermittent leakages-this paper proposes a branch-line leakage localization method targeting users on the customer side of the meter. First, leveraging electrical prior knowledge, a physical model of pre-meter leakage is established, and the underlying mechanism of changes in users′ shortest-path virtual impedance before and after leakage is analyzed. Multivariate linear regression equations are then constructed for users in a distribution area to derive temporal virtual impedance matrices, which are flattened column-wise as model input. A symmetric relative entropy model is proposed to capture both local (adjacent) and global dependencies among users. Its output accuracy is enhanced through a segment aggregation strategy, effectively transforming the leakage localization problem into a time-series anomaly detection task. To improve the model′s sensitivity to subtle feature deviations, a minimax adversarial optimization mechanism is introduced into the reconstruction loss function to amplify differences between normal and leaking users. This is further combined with a collaborative anomaly scoring method based on symmetric relative entropy, enabling robust identification of anomalous users exceeding a predefined threshold. Extensive simulations on the IEEE European low-voltage feeder system under various leakage scenarios are conducted to support hyperparameter tuning and ablation studies. Experimental results demonstrate that the proposed method outperforms existing algorithms in detection accuracy. Moreover, by addressing edge cases-such as outages, no-load users, measurement errors, and electromagnetic interference-the model exhibits strong anti-interference capability. Its effectiveness and generalization ability are further validated through deployment tests on real-world distribution networks.