为准确预测地震死亡人数，提出了基于主成分分析法（PCA）和粒子群算法（PSO）优化的支持向量机（SVM）模型。首先利用主成分分析法对地震死亡人数7个影响因子中的6个进行数据降维，同时对第7个发震时刻因子单独进行区间分类，然后对提取出的主成分进行归一化处理，将归一化的主成分数据作为支持向量机的输入向量，通过粒子群算法寻优获得最优支持向量机模型参数，最终建立基于PCA-PSO-SVM的地震死亡人数预测模型，并对5组样本进行死亡人数预测，同时对比分析包含和不包含发震时刻因子的2种情况下的模型预测效果。结果表明：在不考虑发震时刻因子的情况下，使用PCA-PSO-SVM模型的最小误差、最大误差和平均误差分别为0.85%、20%、10%，其平均误差相比PSO-SVM、SVM模型分别降低2.08%、2.28%；输入向量加入发震时刻因子分类数据后，PCA-PSO-SVM模型的最小误差、最大误差和平均误差分别为0.25%、20%、7.18%，其平均误差相比PSO-SVM、SVM模型分别降低3.34%、3.50%。因此，加入发震时刻因子后3种模型的平均误差明显降低，同时由于PCA-PSO-SVM模型进行主成分降维处理，能够明显提高运行效率和预测精度，故降低了模型复杂度。
In order to predict earthquake casualties accurately, support vector machine (SVM) model optimized by genetic algorithm (PSO) based on principle component analysis (PCA) was proposed. Making the data dimension reduction to 6 factors from 7 impact factors of earthquake casualties using PCA, classifying the origin time of earthquake by intervals, normalizing the extracted principal components which were used as input vectors of support vector machine and optimizing the best SVM parameters using PSO, finally the prediction model for earthquake casualties based on PCA-PSO-SVM was established which was used to predict the casualties of 5 samples. The prediction model results considering the earthquake origin time factors or not were compared. The result shows the minimum error, maximum error and average error of PCA-PSO-SVM model were 0.85%, 20% and 10% respectively without considering the earthquake origin time factor. Compared with PSO-SVM model and SVM model, the average error of PCA-PSO-SVM model is reduced by 2.08% and 2.28% respectively. After the classified data of origin time factor was added in input vectors, the minimum error, maximum error and average error of PCA-PSO-SVM model were 0.25%, 20% and 7.18% respectively. Compared with PSO-SVM model and SVM model, the average error of PCA-PSO-SVM model is reduced by 3.34% and 3.50%, respectively. Therefore, the average error of three models was reduced obviously after adding the earthquake origin time factor, and PCA-PSO-SVM model can improve the operation efficiency and prediction accuracy obviously and reduced the complexity of the model duo to dimension reduction.
2019,40(5): 41-47 收稿日期：2018-11-30
DOI：10.3969/j.issn.1003-3246.2019.05.006
基金项目：河北省地震科技星火计划（项目编号：DZ20160405023）
作者简介：刘立申(1974-),男,高级工程师,主要从事地震分析与观测研究工作。E-mail:270882786@qq.com
*通讯作者：王晨晖(1992-),男,硕士研究生,主要从事地震观测研究工作。E-mail:caesar621@163.com
参考文献：
何明哲,周文松. 基于地震损伤指数的地震人员伤亡预测方法[J]. 哈尔滨工业大学学报,2011,43(4):23-27.
史才旺,何兵寿. 基于主成分分析和梯度重构的全波形反演[J]. 石油地球物理勘探,2018,53(1):95-104.
王军. 支持向量机在地震与爆破识别中的应用[J]. 地震地磁观测与研究,2018,39(3):181-188.
王岩,李彤霞,钱蕊,等. 主成分分析法在2013年灯塔MS 5.1地震预测中的应用[J]. 地震地磁观测与研究,2017,38(5):44-48.
文翔,周斌,阎春恒. 遥感分类方法在建筑物震害提取中的应用(以玉树地震为例)[J]. 地震地磁观测与研究,2014,35(5/6):134-143.
吴新燕,吴昊昱,顾建华. 1999年以来地震生命损失评估研究新进展[J]. 震灾防御技术,2014,9(1):90-102.
谢玮,王彦春,刘建军,等. 基于粒子群优化最小二乘支持向量机的非线性AVO反演[J]. 石油地球物理勘探,2016,51(6):1 187-1 194.
张文路,蒋欢军. 地震人员伤亡评估方法与模型研究综述[J]. 结构工程师,2016,32(3):181-191.
张莹,郭红梅,尹文刚,等. 基于多因素的地震灾害人员伤亡评估模型研究[J]. 震灾防御技术,2017,12(4):870-881.
周德红,冯豪,程乐棋,等. 遗传算法优化的BP神经网络在地震死亡人数评估中的应用[J]. 安全与环境学报,2017,17(6):2 267-2 272.
