在跨断层场地形变分析中，形变异常并非全是地壳应力活动的地表反应，形变资料异常变化会受到多种因素的干扰。因此，在进行场地形变分析前必须识别并排除这些干扰因素的影响。单个跨断层场地观测值出现的异常与地震的关系常常具有较大的不确定性，这就需要通过一定的方法技术把多场地的数据资料进行综合处理，使这些群体观测数据来反映地壳运动。综上，排除干扰因素影响，建立异常识别模型，将单一场地分析延伸到多场地分析就显得尤为重要。 本文通过对场地水准监测数据和干扰因素（气温、气压、降雨量等）观测数据的分析，建立了以时间观测序列相似性为基础的灰色关联分析模型，并尝试探讨了一个灰色关联分析评价模型。文中以窝子滩和大泉口两个场地监测数据为实例进行了灰色关联分析，其结果与SPSS相关性分析的结果相一致。在形变异常识别方面，本文通过灰色预测GM(1,1)模型模拟值来识别形变异常。根据新陈代谢GM(1,1)模型和残差GM(1,1)模型建立了一个动态的异常值识别模型。运用GM(1,1)模型模拟水准观测值建立水准观测基值线，进行单一场地分析。根据观测值曲线的趋势性异常和突跳异常来辅助判定地震前兆反应。同时，通过灰色关联分析法，计算多个场地之间的两两点关联度，并由点关联度序列生成等值线图，根据等值线图来辅助判定地震前兆。 通过实例验证，利用本文提出灰色关联分析模型与评价模型能够有效的识别干扰因素；GM(1,1)模型对监测值序列的模拟精度和预测精度都很高，联合残差GM(1,1)模型计算出的异常值判定阈值能有效判定水准监测值的突跳异常；同一场地对地震前兆表现出的形变异常具有重复性，单一场地可利用重复性异常表现提高地震前兆判定的准确性；多场地联合分析很大程度上避免单一场地出现无震异常的情况，但参与联合分析的场地选取对多场地联合分析的结果影响比较大。

Deformation data anomaly of cross-fault site will be influenced by many factors. Therefore, before the deformation analysis we must exclude the influence of these extraneous factors. The single across-fault site abnormal deformation relationship with earthquake appears often has larger uncertainties, which require a certain amount of technology integration observation data make these groups to reflect crustal movements. In conclusion, eliminate the interference factors establish anomaly model, extend the single site analysis to Multi-site analysis is particularly important. Through the site leveling data and confounding factors (temperature, pressure, rainfall, etc.) analyzing establishes the observation model of grey relational analysis based on sequence similarity, and attempt to establish an evaluation model of grey relational analysis. For WO Zi tan site and Da Quan kou site establishing grey relation analysis, the results are consistent with the results of correlation analysis on SPSS. In terms of deformation anomaly recognition, through establish grey GM (1, 1) model to identify deformation anomaly. According to metabolized GM (1, 1) model and residual error GM (1, 1) model establishes a dynamic outlier’s identification model, using a dynamic threshold to determine deformation anomaly values. Using the GM (1, 1) model to simulate the observation values, we can establish base value line, which trend anomalies and the sudden jump abnormal can assist determining earthquake precursor reaction. At the same time, by grey relational analysis calculating relational degrees of multiple sites each other generate contour maps and according contour maps to assist determine earthquake precursors. Verified by examples, gray relational analysis model and evaluation model can effectively identify confounders. The simulate precision and forecast precision of GM (1, 1) model are high. Combined with residual error GM (1, 1) model be calculated threshold values can effectively determine the jump abnormal of level monitoring. In the same venue deformation anomalies are repetitive for earthquake precursor; a single site can be use of deformation anomaly repeatability to improve the accuracy of the determination of earthquake precursors. Although multi-site joint analysis largely avoid aseismic anomaly, but Sites selected influence on the results of the site combined analysis is relatively large.