Abstract: In order to improve the installation accuracy of steel members of bridges, an improved BP neural network based on the optimized fruit fly algorithm was established to accurately predict the spatial coordinates of steel members. In the fruit fly algorithm, Logistic mapping was introduced to optimize the initial position of fruit fly population, chaotic dynamic weight and multi-population cooperative search were added to enhance the global search and local exploration ability of the algorithm, which effectively solved the problems of long training time and low accuracy of BP neural network in dealing with complex engineering problems. The improved BP neural network could identify many variables that affected the positioning of steel components and screened out the key variables. Combined with an engineering example of a corrugated steel web-prestressed concrete composite beam low tower cable-stayed bridge, the optimized neural network was used to predict the spatial coordinates of steel web and steel anchor box in the section to be constructed. The research shows that the hole shaft deviation and cable casing deviation of the steel anchor box and corrugated steel web opening connector of the bridge have been greatly improved through the verification of the construction process, which reflects the good application prospect of the improved BP network in accurately predicting the spatial coordinates of steel members.

Key words: bridge engineering; steel and concrete composite bridge; positioning accuracy of member; optimized fruit fly algorithm; BP neural network; chaotic dynamic weight

摘要： 为提高桥梁钢构件的安装精度，建立了基于改进果蝇优化算法的BP神经网络，实现了对钢构件空间坐标的精确预测。在果蝇算法中，引入Logistic映射以优化果蝇种群初始位置，增加混沌动态权重和多种群协同搜索用于加强算法的全局搜索和局部探索能力，有效改进了BP神经网络在处理复杂工程问题时训练时间长和精度低等问题。改进后的BP神经网络可对影响钢构件定位的多个变量进行识别，并筛选出其中的关键变量。结合某波形钢腹板-预应力混凝土组合梁矮塔斜拉桥的工程实例，利用优化后的神经网络对即将施工节段的钢腹板和钢锚箱的空间坐标进行预测。研究表明：经过施工过程验证，该桥钢锚箱与波形钢腹板开孔连接件的孔轴误差、钢锚箱拉索套管误差等问题得到了大幅改善，体现了改进的BP网络在准确预测钢构件的空间坐标。

关键词: 桥梁工程 ；钢混组合桥梁；构件定位精度；优化果蝇算法；BP神经网络；混沌动态权重