中文摘要：

为了采用非实验的方法对安全物质学的研究内容及研究方法进行初探,基于定量结构-性质关系法,选择13种与有机过氧化物热危险性的影响因子密切相关的描述符,分别对起始分解温度T0和分解热△H的实验数据进行多元线性回归、偏最小二乘和支持向量机回归分析,从而获得3种相应的预测模型。对比T0与△H的实验值和预测值,结果发现：SVM预测模型的精度高于PLS预测模型,MLR预测模型的精度最低;同种预测模型对分解热的预测结果均优于起始分解温度。此外,分析各预测模型的稳定性数据发现：MLR模型的预测过程发生了过拟合现象,不具备预测能力;PLS模型的交互验证系数均大于0.5,具备较稳定的预测能力;SVM模型的交互验证系数均大于0.9,具备非常稳定的预测能力。

英文摘要：

In order to explore the research contents and methods of safety material science by using the non- experimental methods,based on the method of quantitative structure- property relationship,13 kinds of descriptor being closely related to the influence factors about thermal hazard of organic peroxides were selected. The experimental data of initial decomposition temperature T0 and decomposition heat △H were analyzed by multiple linear regression（ MLR）,partial least squares（ PLS）and support vector machine（ SVM） regression,so as to obtain three kinds of corresponding prediction model. The experimental values and prediction values of T0 and △H were compared,and the results showed that the prediction model of SVM had higher accuracy than PLS,and the prediction accuracy of MLR model was the lowest. The prediction results of decomposition heat were better than the results of initial decomposition temperature by the same prediction model. In addition,through a detailed analysis on the stability data of each prediction model,it showed that the over- fitting phenomenon happened in the prediction process of MLR model,without the prediction ability. The cross validation coefficient of PLS model was greater than 0. 5,with the stable prediction ability. The cross validation coefficient of SVM model was greater than 0. 9,with the very stable prediction ability.