基于极限学习机的钢桥面板腐蚀评估及预测

Corrosion Assessment and Prediction of Steel Bridge Deck Based on Extreme Learning Machine

浇注式导电沥青混凝土能够实现桥面及时、高效融雪化冰,但其通电工作时易加速钢桥面板腐蚀,对桥梁服役性能和交通运行安全造成影响。为解决上述问题,本工作设计制备了五种浇注式导电沥青混凝土,测试了浇注式导电沥青混凝土通电工作时流经钢板的电流密度,研究了浇注式导电沥青混凝土类型、工作条件和环境因素对钢板腐蚀的影响规律,并建立了基于极限学习机神经网络的钢板腐蚀程度预测模型,为浇注式导电沥青混凝土融雪化冰技术在钢桥面铺装领域的进一步推广应用奠定了坚实的基础。结果表明:不同因素对钢板腐蚀的影响程度排序为:通电次数>温度>湿度>通电时间,且通电次数和温度在0.05水平上对钢板腐蚀的影响较为显著;与传统神经网络预测模型相比,极限学习机预测模型具有更好的准确性和高效性,其平均绝对误差、平均绝对百分比误差和均方根误差分别比前者低了50.48%、45.89%和49.30%。

Conductive gussasphalt mixture can melt snow and ice timely and efficiently,but it is easy to accelerate the corrosion of steel deck when it is powered on.It will affect the service performance and traffic safety of bridges.In order to solve the above problems,five kinds of conductive gussasphalt mixtures were designed and prepared.When conductive gussasphalt mixtures were powered on,the current density flowing through the steel plate was tested.The effects of types,working conditions and environmental factors of conductive gussasphalt mixtures on steel plate corrosion were studied.And the prediction model of steel plate corrosion based on the extreme learning machine neural network was established.It lays a solid foundation for the further popularization and application of snow melting technology of conductive gussasphalt mixture in the field of steel bridge deck pavement.The results show that the influence degree order of different factors on steel plate corrosion was ranked as follows:number of time on power>temperature>humidity>electrified time.The influence of number of time on power and temperature on steel plate corrosion were more significant at the level of 0.05.Compared with the traditional neural network prediction model,the extreme learning machine prediction model had better accuracy and efficiency.And its mean absolute error,mean absolute percent error and root mean squared error were 50.48%,45.89%and 49.30%lower than the former,respectively.