中文摘要：

为明确大跨度悬索桥扁平钢箱梁温度和温度梯度特征,以南溪长江大桥正交异性钢箱梁为研究对象,基于健康监测系统中温度传感器的长期实测数据,采用分段函数描述环境温度和日照辐射共同作用下钢箱梁日温度变化曲线。在此基础上,采用高斯混合模型描述钢箱梁年温度多峰概率分布,并引入赤池信息判别准则（AIC）和贝叶斯信息判别准则（BIC）确定最优高斯分量数。统计钢箱梁一年日温差极值并进行参数评估,得到钢箱梁年温差极值分布模型。对年温差极值分布函数进行外推,得到设计基准期温差的极值分布函数并计算温差标准值。引入相关系数分析法对各温差组进行相关性分析,剔除实际不存在的温差模型。研究结果表明：相比正弦函数,分段函数能更准确地描述太阳辐射作用下箱梁截面日温度变化特征;当高斯分量数为3时,混合高斯模型拟合钢箱梁年温度概率分布最优;外推设计基准期模型能够较好地计算设计基准期温差标准值;通过相关性分析剔除了4组不存在的温差模式;得到顶板和腹板各8组温差模式;最后与《公路桥涵设计通用规范》（JTG D60—2015）中钢混结构竖向正温差设计值进行对比,一、二级温度梯度与设计规范分别相差4.2℃和2.3℃。

英文摘要：

In order to clarify the characteristics of temperature and temperature gradient of orthotropic steel box girder for long-span suspension bridge,taken the Nan-Xi Yangtze River bridge as the research object,the longtime measured data of the steel box girder was collected based on temperature sensors in health monitoring system（HMS）,and the daily temperature variation curves of steel box girder under interaction of ambient temperature and light radiation was described by piecewise functions.On this basis,the Gaussian Mixture Model（GMM）was selected to fit the annual temperature multi-peak probability distribution of steel box girder.Akaike information criterion（AIC）and Bayesian information criterion（BIC）was introduced to determine number of GMM component.The extremal values of daily temperature differences in a year were generalized and then parameters were got to evaluate.Extremal distribution models ofannual temperature differences of steel box girder were obtained.By extrapolating the annual extremal distribution functions of annual temperature differences,the extremal distribution functions of temperature differences in design reference period was obtained,and then the standard values of temperature differences was calculated.Correlation analysis was carried out in all groups of temperature differences by the correlation coefficient method,removing nonexistent models of temperature differences.The results show that piecewise function is a more accurate method to describe characteristics of daily temperature variation of box girder section than sine function,under the action of light radiation.The probability distribution of annual temperature reaches to optimum when the number of GMM component is 3.Standard values of temperature differences in design reference period can be better calculated by extrapolated design reference period model.Four groups of nonexistent temperature difference models are removed and 8groups of temperature difference models in roof and web respectively are got by correlati