In this paper we discuss the tunnel safety monitoring principle, the monitoring system layout and data acquisition method using embedded FBG strain sensors in a certain practical tunnel construction project. Based on above discussion, the synchronous multidimensional strain data streams collected from the same cross section are analyzed, an abnormal data realtime diagnosis algorithm is proposed based on double slide window model technology. With the first sliding window, the current stress data gathered by the FBG sensors are combined with the historical data to form the current observation window data; then, principal component analysis is conducted on the data in every observation window, corresponding eigenvectors are extracted to obtain the dynamic characteristics of the present observation window. Next, the second sliding window is used to merge current dynamical observation vector into historical ones to get the dynamical feature matrix, and correlation analysis is performed. Finally, the variance of the correlation coefficient is computed, which is used as the criterion to judge the stability of the supervised data streams. Field observation report and comparison experiment result show that the proposed method achieves better real time monitoring effects. This method also provides a strong support for the application of FBG strain sensor in realtime engineering safety monitoring.