The Steam Isolation Valve (SIV) plays an important role in ensuring the safe operation of the secondary loop in a high-temperature gas-cooled reactor (HTR). The pneumatic-hydraulic actuator with a complex structure is integrated into the SIV to ensure the SIV can be closed within three seconds under emergency conditions. The intricate structure and function of the SIV make it challenging to detect anomalies effectively. This paper proposes a novel anomaly detection method for multivariate time series data associated with SIV. Given that univariate could not reflect the operating status of SIV comprehensively, various types of sensors are employed to monitor the SIV and its actuator. These sensors capture pressure, temperature, displacement, and thrust signals respectively. The interdependent relationships among the parameters of the SIV are analyzed. The coupling mechanism derived from multivariate time series is clarified in this research. This method defines several thresholds and characteristic indexes based on the coupled model to facilitate online anomaly detection. By integrating sliding windows with adaptive thresholds, this method can efficiently process data streams, improving detection accuracy. Compared to existing anomaly detection techniques for NPPs, the proposed approach focuses on SIV, demonstrating enhanced interpretability, superior anomaly detection performance, and more efficient management of abnormal conditions.