Multimodal deep learning for anomaly detection in urban infrastructure networks: improving the resilience of public management systems

Urban infrastructure network is the core supporting part of the public management system, which includes key subsystems such as transportation, energy, water supply and drainage, and communication. The stability of its operation is directly related to the normal operation of urban functions and public safety. At present, the traditional anomaly detection (AD) methods have some problems, such as insufficient processing ability of multi-source heterogeneous data, poor adaptability to complex scenarios, and lagging real-time response, which make it difficult to meet the higher requirements of new smart cities for infrastructure network security management and control. In view of this situation, this paper gives a framework for AD of urban infrastructure networks based on multi-modal in-depth learning, namely MM-ADF, which integrates multi-modal information such as sensor time series data, device appearance images, running audio and network logs, with the help of cross-modal feature fusion, spatio-temporal correlation modeling and network logs. Accurate identification and real-time early warning of equipment failure, data leakage, abnormal performance and other issues. The experimental results show that the detection accuracy of the framework reaches 94. 7%, the false detection rate is reduced to 3. 2%, and the average response time is shortened to 280 ms on the real urban infrastructure data set, which is much better than traditional single-mode detection method, and effectively enhances the robustness of the urban public management system against interference and strengthens its emergency response resilience. It provides a practical technical scheme for the safety management and control of new smart city infrastructure.