基于BOA-ELM的混凝土抗压强度预测研究
    点此下载全文
引用本文:吴小平1,李元栋2?覮,张英杰2,阮映辉3,刘志文4.基于BOA-ELM的混凝土抗压强度预测研究[J].计算技术与自动化,2020,(1):140-144
摘要点击次数: 765
全文下载次数: 0
作者单位
吴小平1,李元栋2?覮,张英杰2,阮映辉3,刘志文4 (1. 浙江省公路管理局浙江 杭州 3100092. 湖南大学 信息科学与工程学院湖南 长沙 410082 3. 浙江台州市沿海高速公路有限公司浙江 台州 3180004. 湖南大学 土木工程学院湖南 长沙 410082) 
中文摘要:为控制控制混凝土生产成本,在混凝土拌和期限制抗压强度不足的缺陷构建产出,可以有效降低原料的浪费,是节能降耗的关键方法之一。针对混凝土抗压强度的传统测量方法严重滞后的问题,提出了基于贝叶斯优化极限学习机(BOA-ELM)的混凝土抗压强度预测方法。首先,分析了混凝土拌和过程中对抗压强度预测值实时获得的需求。以各物料的用量为分析基础,28天标准养护后混凝土抗压强度值为预测目标,设计了基于极限学习机的强度预测模型。其次,为进一步提高模型的稳定性以及准确行,提出基于贝叶斯优化的极限学习机模型,根据模型超参数的分布特征,以高斯过程作为超参的先验分布,预测误差最小化作为目标,寻找最优的模型超参。最后,在实际施工产生的C50标号混凝土数据集上测试文中模型,并对比分析了其他预测模型和寻优算法。结果表明,结合了贝叶斯优化的极限学习机预测模型相较于经典算法具有更高的预测准确性和模型训练的高效性。
中文关键词:混凝土  抗压强度预测模型  极限学习机  贝叶斯优化  软测量
 
Prediction of Concrete Compressive Strength Based on BOA-ELM
Abstract:In order to control the cost of concrete,it is one of the key methods to save energy and avoid the waste of materials by limiting the output of defects whoes compressive strength is not up to standard,during the concrete mixing period. Because the traditional measurement method for concrete compressive strength is seriously lagging,the prediction method of concrete compressive strength based on extreme learning machine Bayesian optimized(BOA-ELM) is proposed. Firstly,the real-time demand of the value of the compressive strength predicted during the concrete mixing process is analyzed. Based on the analysis of the amount of each material and the compressive strength of the concrete after 28 days of standard curing ,the strength based on the extreme learning machine is designed. Secondly,in order to further improve the stability and accuracy of the predictive model,a extreme learning machine model Bayesian optimized(BOA-ELM) is proposed. According to the distribution of the model hyperparameters,the Gaussian process is used as the prior distribution ,and take prediction error minimized as the target,to find the best model super parameters. Finally,the model is tested on the C50 concrete dataset generated by the actual construction,and other prediction models and optimization algorithms are compared and analyzed. The results show that the extreme learning machine prediction model combined with Bayesian optimization has higher prediction accuracy and more efficiency of model training than classical algorithms.
keywords:concrete  prediction model of compressive strength  extreme learning machine  Bayesian optimization algorithm  soft-sensor
查看全文   查看/发表评论   下载pdf阅读器