Forecast model for product design time based onprobabilistic support vector regression
基于概率支持向量回归的产品设计时间预测模型

This paper proposed a forecast model based on probabilistic support vector regression (PSVR) to overcome the problems of small samples and heteroscedastic noise in design time forecast, and provided useful information in addition to the forecast value. Firstly, it designed probability constraints on the basis of heteroscedastic regression model, and made sure that for every sample the forecast value was in a neighborhood of the target value with high probability. It formulated the optimization objective in the form of support vector regression with parameter-insensitive loss function, and proposed PSVR. Then, it embedded prior knowledge of maximum completion time into the constraints of PSVR, and provided the size of the neighborhood of the target value. It applied the combination of genetic algorithm and cross validation to determine the relevant parameters of PSVR. Finally, it analyzed the application in injection mold designs. The results verify that the time forecast method based on PSVR can simultaneously provide effective forecast values and forecast intervals.