一种基于PSO-SVR的软件可靠性评估方法

摘要

软件可靠性建模时,如果简单地利用支持向量回归机制建模,就有可能由于支持向量回归(SVR)自身参数选择难以及实验数据本身的不确定性,从而导致预测结果不理想、精度低等缺陷.因此,借鉴粒子群优化算法(PSO)多参数寻优的优势,将PSO与SVR优化算法相结合,利用分层聚类算法对初始实验数据进行归一化处理,剔除异常数据,构建基于PSO-SVR的软件可靠性评估方法,从而提高软件模型的预测精度.实验结果表明,基于PSO-SVR方法的预测模型其预测精度高,更适应实际软件应用环境.%In the process of software reliability modeling, if the support vector regression mechanism is simply used, it is likely that the difficulty in parameter selection and the uncertainty in experimental data themselves may lead to the realistic predicted results and low accuracy. For this reason, it is necessary to reference the multi -parameter optimization advantages of particle swarm algorithm to combine PSO and SVR optimization algorithm, to normalize the initial experimental data with hierarchical algorithms, and to reject the abnormal data and construct a software reliability assessment method based on PSO-SVR. Experimental results show that the prediction model based on PSO-SVR has a high precise and a suitable environment for software application.