基于卷积神经网络预测模型的混凝土配合比优化  

Optimization of concrete mix proportion based on convolutional neural network prediction model

作　　者：李建华 刘雪平 郭家舜 LI Jianhua

机构地区：[1]四川港航建设工程有限公司,四川成都610000 [2]长安大学道路施工技术与装备教育部重点实验室,陕西西安710064

出　　处：《水泥》2025年第4期57-61,共5页Cement

摘　　要：基于岷江老木孔航电枢纽工程的二级配基础混凝土的性能试验数据,建立了一种基于卷积神经网络和双向长短期记忆网络的混合神经网络,并引入注意力机制,得到了通过配合比预测性能的模型。接着,采用非支配排序遗传算法Ⅱ(Non-dominated sorting genetic algorithmⅡ,NSGA-Ⅱ)对该预测模型的参数进行了优化,使该模型对28 d抗压强度的预测误差由优化前的38.54%降低到了0.26%,对坍落度的预测误差由优化前的0.9%降低到了0.23%。随后,基于优化后的预测模型,以混凝土的28 d抗压强度和坍落度最大为优化目标,采用NSGA-Ⅱ算法对混凝土的配合比进行了优化。优化后的28 d抗压强度比试验得到的最大值提高了13.6%,坍落度也满足工程需求。最后,对最优配合比进行了试验验证,结果表明,最优配合比的理论28 d抗压强度与实际值的误差为2.57%,坍落度与实际值的误差为0.79%,证明本文方法具有较好的优化效果。Based on the performance test datum of the second graded foundation concrete of the Minjiang Laomukong navigation and hydropower junction project,a hybrid neural network based on convolutional neural network and bidirectional long short-term memory network was established,and attention mechanism was introduced to obtain a model that predicts performance through mix proportion.Subsequently,non-dominated sorting genetic algorithm II(NSGA-II)was used to optimize the parameters of the prediction model,which reduced the prediction error of 28 day compressive strength from 38.54%before optimization to 0.26%,and reduced the prediction error of slump from 0.9%before optimization to 0.23%.Then,based on the optimized prediction model,the mix proportion of concrete was optimized using the NSGA-II algorithm with the maximum 28 day compressive strength and slump of concrete as the optimization objectives.The optimized 28 day compressive strength increased by 13.6%compared with the maximum value obtained from the experiment,and the slump also met the engineering requirements.Finally,experimental verification was conducted on the optimal mix proportion,and the results showed that the error between the theoretical 28 day compressive strength and the actual value of the optimal mix proportion was 2.57%,and the error of slump between the predicted value and the actual value was 0.79%,proving that the method proposed in this paper had good optimization effect.

关 键 词：混凝土配合比 卷积神经网络 预测模型 NSGA-Ⅱ 多目标优化 

分 类 号：TV431.2[水利工程—水工结构工程]