基于GA-BP神经网络长服役期内结构混凝土的强度演变预测

Strength evolution prediction of concrete structures during longterm service life based on GA-BP neural network

为研究长服役期内既有混凝土结构的强度演变规律及其预测模型,以唐包铁路、西户铁路等实际工程为研究背景,以服役时间为2、16、25、30、40、52、66、88、95和100 a的在役桥涵为研究对象,基于混凝土回弹法,开展役桥涵混凝土强度试验,分析长服役期内既有桥涵混凝土强度动态发展过程。同时,基于试验实测混凝土强度数据与收集的230组同类条件下在役桥涵(服役时间2~88 a)混凝土强度数据,构建GA-BP神经网络混凝土强度预测模型。此外,为提高模型可应用性,基于高精度GA-BP神经网络强度预测模型,建立一般矩阵公式和简化公式。基于本文构建的混凝土强度预测模型,分析该类地区(试验中已调研区域)长服役期内混凝土结构的强度演变规律。研究结果表明:相较于既有混凝土强度预测模型,本文构建的GA-BP神经网络混凝土强度预测模型可有效预测不同服役时间下的混凝土强度,预测数据的平均绝对百分比误差为8.76%,决定系数为0.83。本文简化公式(C25)精度较高,平均绝对百分比误差为6.6%,为便于简化计算,推荐简化公式(C25)作为长服役期内混凝土强度预测公式。百年服役期内混凝土强度经历2个时间阶段,即混凝土强度缓慢上升期(1~49 a)、混凝土强度快速下降期(49~100 a)。随混凝土结构服役时间增加,混凝土结构劣化速率增加,导致混凝土结构在长期服役过程中,混凝土强度不能满足混凝土结构服役要求。

In order to study the strength evolution law and prediction model of existing concrete structures during long-term service,the actual projects such as Tangbao Railway and Xihu Railway were taken as the research background,and the in-service bridges and culverts with service time of 2,16,25,30,40,52,66,88,95 and 100 a were taken as the research objects.Based on the concrete rebound method,the concrete strength test of the in-service bridge and culvert was carried out,and the dynamic development process of the strength of the existing bridge and culvert concrete during the ultra-long service period was analyzed.At the same time,based on the concrete strength data measured by the test and the concrete strength data of 230 sets of in-service bridges and culverts(service time 2-88 a)collected under similar conditions,a GA-BP neural network concrete strength prediction model was constructed.In addition,in order to improve the applicability of the model,a general matrix formula and a simplified formula were established based on the high-precision GA-BP neural network strength prediction model.Based on the concrete strength prediction model constructed in this paper,the strength evolution law of concrete structures during the long service period in such areas(the areas that have been investigated in the experiment)was analyzed.The results show that compared with the existing concrete strength prediction model,the GA-BP neural network concrete strength prediction model constructed in this paper can effectively predict the concrete strength under different service times,and the average absolute percentage error of the prediction data is 8.76%,and the coefficient of determination is 0.83.The simplified formula(C25)in this paper has high accuracy,the average absolute percentage error is 6.6%.Therefore,in order to facilitate the simplified calculation,the simplified formula(C25)is recommended as the concrete strength prediction formula during the ultra-long service period.The concrete strength during the 100 a service period goes through two time stages,namely,the period of slow rise of concrete strength(1-49 a)and the period of rapid decline of concrete strength(49-100 a).With the increase of the service time of concrete structures,the deterioration rate of concrete structures increases,resulting in the concrete strength of concrete structures cannot meet the service requirements of concrete structures in the long-term service process.