Amounts of sensor stream data are collected in industrial area, diverse modes and dynamic working condition, and puts forward higher requirements for efficient and effective anomaly detection. The interplay and mutual influence among sensor streams suggest that underlying correlation can be used to identify and explain abnormal problems. This paper introduces an innovative service-based anomaly detection method that leverages lag-correlation analysis of stream data. It first constructs correlation graph model based on lag-correlation analysis of history data, and divides sensor stream data groups based on correlation degree. Then sensor stream groups are encapsulated as corresponding stream data services, and realize the real-time anomaly detection within and out of stream groups in a discrete way based on service collaboration. Experimental results on a real industrial sensor dataset demonstrate the effectiveness of the proposed method in detecting anomalies across multiple sensor streams.