With the ongoing advancement of smart grid deployment and the growing diversification of user service requirements, electric power call platforms are shouldering increasingly critical operational roles, and their system stability and communication security face formidable challenges. Traditional anomaly detection approaches, which typically rely on simplistic rule matching and static threshold monitoring, prove inadequate for addressing complex, evolving communication irregularities and latent security threats. In this paper, we address the problem of communication topology modeling and anomaly detection for electric power call platforms by proposing a multi-stage algorithmic workflow. Our approach integrates graph-based modeling, unsupervised clustering analysis, and time-series dynamic threshold detection to construct the communication topology and identify anomalous patterns. Core contributions include: (1) a cross-view consistency metric that integrates data from the physical, logical, and security views to enhance detection of anomalous links and nodes; (2) an event-driven dynamic window mechanism which, compared to traditional fixed windows, improves sensitivity to topology evolution; (3) a multi-source information fusion mechanism—incorporating communication latency, status codes, and session frequency—into topology edge-weight computation, enabling dynamic adjustment of weights according to emerging anomaly trends; (4) targeted preprocessing methods to address algorithmic time/space complexity and to cleanse input-log noise such as formatting irregularities, duplicates, and missing entries.