The security of pipeline systems draws attention increasingly; therefore, a novel method based on neural network and graph theory is proposed for the detection and localization of pipeline networks in this paper. First, the detection algorithm based on the broad learning system (BLS) is used to distinguish abnormities under large-scale pipeline network environments. During the process, the varied BLS models result in indeterminate performance and fast ergodic structure search is executed via adaptive mutation particle swarm algorithm (APSO) to generate an appropriate structure, succinct parameters, speedability, and accuracy. And manual features are implanted into the BLS feature layer to targetedly improve performance for complex pipeline network signals. Second, based on the detection results, a universal Dijkstra-based applicable localization method is proposed for diverse topological pipeline structures, including mesh-form networks, which have fewer sensors than anchors. The synchronous approximation is adopted to shun local minimum, and the shrinkage of search domain economizes time. Revised BLS was contrasted with several networks trained by real pipeline data and the system was integrated into SCADA and applied on an operational large-scale pipeline network successfully.