基于PCA-ELM模型的露采爆破振动对民房破坏的预测分析

针对露天采矿爆破振动对民房破坏的预测问题，采用主成分分析(PCA)和极限学习机(ELM)方法，选取爆破振幅、主频率、主频率持续时间、灰缝强度、砖墙面积率、房屋高度、屋盖形式、圈梁立柱、施工质量、场地条件10个主要影响因素。引入相关性分析在主成分分析过程中，对相关性高的指标进行降维，把得到的3个综合因子和爆破振幅、主频率、主频率持续时间、砖墙面积率作为输入变量，构建露天煤矿PCA-ELM预测模型。选取露天矿实际爆破过程中测量的100组数据作为模型训练样本，用另外20组数据作为测试样本进行预测。结果表明:对民房破坏影响因素中灰缝强度、房屋高度、屋盖形式、圈梁立柱、施工质量、场地条件之间具有较高的关联度。该模型处理高维数据时较传统的ELM算法具有预测精度高、稳定性好等特点，可准确预测爆破振动对民房的破坏程度，误判率为1/20。

Predicting analysis on damage to residential house by blasting vibration in open pit mining based on PCA-ELM model

For the predicting problem of damage to residential house by blasting vibration in open pit mining, by adopting the principal component analysis (PCA)and extreme learning machine (ELM) method, 10 major influencing factors were selected, including blasting amplitude, main frequency, duration of main frequency, mortar joint strength, brick wall area ratio, building height, roof form, ring beam column, construction quality and site conditions. By introducing correlation analysis into PCA process, dimension reduction was carried out on the indexes with high correlation. Taking the obtained 3 comprehensive factors and blasting amplitude, main frequency, duration of main frequency blasting and brick wall area ratio as the input variables, the PCA-ELM prediction model of the open-pit coal mine was established. 100 groups of data measured at the process of blasting in open-pit mine were selected as the training samples of the model, and the other 20 groups of data were selected as the test samples to perform prediction. The results showed that in the influence factors of damage to residential house, there exists a higher correlation among the mortar strength, height of building, roof forms, beam column, the quality of construction and site conditions. Compared with the traditional ELM algorithm, at the time of processing high-dimensional data, the model has the characteristics of high accuracy, good stability and so on, and it can accurately predict the damage extent of blasting vibration on the houses with the misjudgment rate as 1/20.