一种基于SVR的综合预测方法及应用

针对一类因变量具有复杂自变量、且不具备相同采样周期的预测问题，综合运用支持向量回归估计(SVR)、多元回归和主成分分析等多种数据分析技术，提出了一种综合预测方法，建立起了飞机故障率与其错综复杂的影响因素间的一种数学关系，并且采用航空装备质量控制的统计数据对所提出的方法进行了实验，预测结果显示了方法的有效性。在影响因素量化过程中，还引入了Pearson相关系数方法。

New Synthetic Prediction Method Based on SVR and Its Application

For the problem that the dependent variable has many independent variables and their sampling periods are also different, a predicting method is proposed by using synthetically the data analysis methods of support vector regression (SVR), multivariate regression and principal component analysis, etc. The method can be briefly described as follows: 1. Predicting with the independent variables which have dense sampling periods based on SVR, and then the results are synchronized to have the same sampling period with the dependent variable. 2. Amending the results by using another linear or non-linear method which includes SVR itself, with the rest independent variables which have the same sampling periods with the dependent variable. 3. In order to increase the predictive accuracy, three data processing methods (principal component analysis, standardization and normalization) are integrated. 4. Two approaches, error mean square line and small error probability, are also introduced to evaluating this synthetic method. By using the method, the mathematical relation between the aircraft's failure ratio and its anfractuous factors is first established. The results show that the method is efficient in predicting the aircraft's failure ratio. In the process of quantifying some influencing factors of the aircraft's failure ratio, the Pearson's correlation coefficient method is also adopted.