刘海峰, 刘浩天, 李罗胤,,等.基于改进BP模型的沙漠砂混凝土高温后抗压强度预测[J].河南理工大学学报（自然科学版）,2026,45(1):179-188.

LIU H F, LIU H T, LI L Y,et al.Prediction of compressive strength of desert sand concrete after high temperature based on an improved BP model[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(1):179-188.

基于改进BP模型的沙漠砂混凝土高温后抗压强度预测

刘海峰, 刘浩天, 李罗胤, 陈小龙, 车佳玲, 杨维武

宁夏大学 土木与水利工程学院，宁夏 银川 750021

摘要: 目的 为探究高温历程对沙漠砂混凝土（desert sand concrete， DSC）抗压强度的影响，考虑沙漠砂替代率、温度、升温速率和静置时间对高温后DSC进行抗压强度试验。 方法 借助X射线衍射和扫描电子显微镜分析高温后DSC微观形貌和物相组成变化规律，以反向传播算法为基准，融合粒子群算法和遗传算法训练人工神经网络，建立高温后DSC抗压强度预测模型，并采用十折交叉验证的方法对该模型进行验证。 结果 结果表明：随着温度升高，DSC抗压强度呈下降趋势，材料内部水化产物大量分解，微观裂缝逐渐扩展并连接贯通；静置时间越长，抗压强度越高；升温速率越快，DSC破坏速率随之增大；沙漠砂替代率为20%时，DSC抗压强度达到最大值。3种预测模型预测值与实测值的平均绝对百分比误差均控制在8%以内。模型优化程度越高，误差范围越小。采用粒子群优化遗传混合算法神经网络模型预测结果更为精准，该模型预测值均方差RMSE为1.127 2，平均绝对百分比误差MAPE为3.98%，28 d抗压强度预测决定系数R²为0.987 8。 结论 本文方法显著提高了DSC高温后力学性能预测的准确性。 

关键词:沙漠砂混凝土;抗压强度;高温;神经网络模型

doi:10.16186/j.cnki.1673-9787.2024030006

基金项目:国家自然科学基金资助项目（52168034）；宁夏回族自治区自然科学基金资助项目（2023AAC03039）；2023年自治区级大学生创新项目；宁夏高等学校一流学科建设（水利工程学科）资助项目（NXYLXK2021A03）；2024年宁夏大学研究生创新项目（CXXM202454）

收稿日期:2024/03/03

修回日期:2024/05/21

出版日期:2025-12-03

Prediction of compressive strength of desert sand concrete after high temperature based on an improved BP model

Liu Haifeng, Liu Haotian, Li Luoying, Chen Xiaolong, Che Jialing, Yang Weiwu

School of Civil and Hydraulic Engineering， Ningxia University， Yinchuan 750021， Ningxia， China

Abstract: Objectives To investigate the effect of high-temperature history on the compressive strength of desert sand concrete （DSC）， compressive strength tests of DSC after exposure to high temperatures were conducted by considering the effects of desert sand replacement rate （DSRR）， temperature， heating rate， and resting time. Methods X-ray diffraction （XRD） and scanning electron microscopy （SEM） were employed to analyze the changes in the microstructure and phase composition of DSC after high-temperature exposure. Based on the back-propagation （BP） algorithm， an artificial neural network （ANN） for predicting the compressive strength of DSC after high temperature was developed by integrating particle swarm optimization （PSO） and genetic algorithm （GA）. The model was validated using ten-fold cross-validation. Results The results indicated that the compressive strength of DSC decreased with increasing temperature， accompanied by significant decomposition of hydration products and progressive propagation and interconnection of microcracks. Longer resting time led to higher compressive strength， while higher heating rates accelerated the damage of DSC. The compressive strength reached its maximum when the DSRR was 20%. The mean absolute percentage errors （MAPE） of the three predictive models were all within 8%. A higher degree of model optimization resulted in smaller prediction errors. The hybrid PSO-GA neural network model achieved the highest prediction accuracy， with a root mean square error （RMSE） of 1.127 2， MAPE of 3.98%， and a determination coefficient （R²） of 0.9878 for the 28-day compressive strength. Conclusions The proposed method significantly improves the prediction accuracy of the mechanical properties of DSC after exposure to high temperature.

Key words:desert sand concrete;compressive strength;high temperature;neural network model