因工程条件限制或其他原因, 部分闸首工程无法通过混凝土绝热温升试验来确定混凝土热学参数, 因此提出 基于BP 神经网络的船闸闸首热学参数反分析方法。首先使用均匀设计理论构造热学参数组合, 利用温度场三维 有限元正分析得到闸首的计算温度样本训练网络; 然后将实测温度值输入神经网络, 对闸首混凝土多个热学特性参 数进行同时反演; 最后利用反分析后的热学参数进行温度场正分析, 预测关键位置的温度时程曲线, 并与实测温度 时程曲线进行对比。实例分析结果表明, 利用均匀设计理论构造待反演参数组, 可以提高神经网络反分析收敛的速 度, 获得的参数满足工程要求。

Due to eng ineering constructio n limits, the thermal parameter s of concrete for some lo ck head pr ojects cannot be deter2 mined by the adiabatic temperatur e rising test. In this paper, an inver sion analy sis method w as pr oposed to determine the lock head thermal par ameter s based o n BP neural netw or k. First, the combinatio ns o f co ncr ete thermal par ameters w ere constructed based on the unifo rm desig n theor y and they w ere used to generate a series of training samples using the thermal FEM analysis. Then, the measured temper ature dat a w ere put into the neural netw ork to per form the inv ersio n ana lysis on the thermal parame2 ter s. Finally, t he therma l parameters o btained from the inv er sion ana lysis w ere used to ana lyze and pr edict the temper ature data at certain po ints and compared t hem w ith the actual temperature data. The results show ed the co nv erg ence r ate o f t he BP neural netw or k is improv ed by the t hermal par ameters developed by the unifo rm desig n, and the obtained parameter s meet the eng i2 neering requirements.