中文摘要：

以统计学的自助法和灰色系统理论的灰关联分析为基础，提出自助融合与灰检验（BFGT）方法，以解决多传感器滑坡时间序列的最佳融合及其假设检验问题。借助再抽样方法和最大熵原理模仿出不同时刻和不同位置的多传感器滑坡时间序列的自助分布，刻画滑坡过程的某些瞬时特征；用自助分布进行加权均值计算，提取这些瞬间特征，构成自助融合序列。通过定义灰差、属性权重映射和灰置信水平等概念，提出假设检验多传感器时间序列相容性的灰否定域。根据来自滑坡体不同部位的分布式多传感器系统的时间序列信息，BFGT可以模拟出整个滑坡体的瞬时状态和演化过程，并用最大熵概率分布来描述。此外，采用最大熵概率加权均值估计技术，BFGT可以识别同时刻多传感器信源噪声的概率分布，减少噪声，进而削弱噪声对估计结果的影响。滑坡工程应用研究结果表明，BFGT对概率分布无任何要求，允许趋势项的变化规律未知，也允许数据个数很少，可以得到90％的最低灰置信水平和95％的平均灰置信水平。

英文摘要：

Time series of multi-sensor belong to non-stationary random process. Based on bootstrap of statistics and grey＇ relational analysis of grey system theory, a method called BFGT（bootstrap fusion and grey testing） is proposed to resolve the problem about optimum fusion and its hypothesis testing for landslide time series of multi-sensor. By means of bootstrap resampling method and maximum entropy theory to imitate bootstrap distribution of time series of multi-sensor at different hours and in different positions, some of instantaneous features of the landslide process are depicted. Extracting these instantaneous features via calculation of the weighting mean using bootstrap distribution, bootstrap fusion series are formed. And grey rejection region of hypothesis testing for compatibility of time series of multi-sensor through definition of concepts, grey difference,attribute weight mapping and grey confidence level. By using time series information from distributed system of multi-sensor placed in different positions, the BFGT can simulate both transient state and evolvement process of overall landslide, which are described with maximum entropy probability distribution. In addition, by the aid of weighting mean value estimating technique of maximum entropy probability, the BFGT can identify probability distribution of the noise from information sources of multi-sensor at the same hour, reducing the noise, and then weakening influence of the noise on estimated results. Engineering application in landslide shows that the BFGT permits trend of non-stationary random process unknown and the number of data small without any requirements for probability distribution, having the lowest grey confidence level of 90% and the mean grey confidence level of 95%.