“三合一”自建房结构安全智能预警模型

Intelligent early warning model for structural safety of“three-in-one”self-built houses

“三合一”自建房虽便于业主生产、储存、经营活动,但因其功能复合,其结构安全性常存在较多严重的问题。为提前预警此类建筑的安全隐患,本研究基于机器学习算法构建了“三合一”自建房结构安全预警模型,并先后应用集成思想和智能优化算法,对预警性能进行了2次提升。首先,采用独热编码、过采样等方法对某地获取的数据进行预处理;其次,选择总体准确率、召回率和AUC值3项指标分别选出最佳的基分类器;之后,使用袋装法和提升法等集成思想优化预警模型的性能;然后,应用鲸鱼、金豺、粒子群等智能优化算法对预警模型性能进一步优化;最后,综合以上模型运算结果,挖掘预警指标中的关键指标。结果表明:(1)经鲸鱼算法优化的Bagging(KNN)模型可更高效地对“三合一”自建房的结构安全进行预警,其召回率为0.802;(2)经鲸鱼算法优化的Boosting(SVM)模型具有更稳定的预警鲁棒性,其AUC值为0.933;(3)默认参数的XGB模型整体预警效率更佳,其总体准确率为0.915;(4)建筑年份、砖混结构、地上层数、建筑面积等14个指标是“三合一”自建房结构安全预警的关键指标。

Although a“three-in-one”self-built house is convenient for the owner to produce,store and operate,there are often many serious problems in its structural safety due to its complex functions.In order to warn the potential safety hazard of such buildings in advance,this study constructed a“three-in-one”self-built housing structure safety intelligent early warning model based on machine learning algorithm,and applied the integration idea and intelligent algorithm to improve the early warning performance twice.Firstly,the data obtained in a certain place were preprocessed by means of one-hot coding and oversampling.Secondly,the overall accuracy,recall rate and AUC value were selected to select the best base classifier.Then,the integrated ideas such as bag method and lifting method were used to improve the early warning performance.The intelligent optimization algorithms such as WOA,GJO and PSO were applied to further improve the early warning performance.Finally,based on the above model calculation results,the key indicators in the early warning indicators were mined.The results are as follows:(1)The Bagging(KNN)model optimized by the whale algorithm can more efficiently warn the structural safety of the“three-in-one”self-built houses,with a recall rate of 0.802;(2)the Boosting(SVM)model optimized by the whale algorithm has more stable early warning robustness,and the AUC value is 0.933;(3)the XGB model with default parameters has better overall early warning efficiency,and the overall accuracy rate is 0.915;(4)14 indicators such as building year,brick-concrete structure,number of upper floors and building area are the key indicators for the early warning of“three-in-one”self-built houses.