基于免疫优化的产品系统可靠性参数区间预测方法

Intervals prediction of system reliability parameters based on immune optimization

针对当前系统可靠性预测方法的不足,提出一种基于免疫优化的区间预测方法.构建可靠性参数预测神经网络模型,由极端学习机(ELM)算法进行训练,并以修正BFGS法对网络的左侧权值和阈值进行调整.在由该网络得到的预测点值和网络权重的基础上,根据非线性回归模型构建可靠性参数的区间预测值(PI),PI的质量取决于网络结构和权衰减调节参数.结合PI的覆盖率和平均区间比例长度提出一种新的PI综合评价指标,以此衡量PI的质量;引入免疫优化算法优化区间预测值和网络结构,以最小化综合评价指标为成本函数,寻求决策变量,即网络隐层神经元个数和权衰减调节参数的最优值.将提出的方法和理论应用于某系列数控车床的可靠性参数平均无故障时间的预测,证明了其预测性能优于传统方法.

Aiming at the deficiencies of system reliability prediction, an interval prediction method based on immune optimization algorithm was put forward. First, a prediction neural network of reliability parameters was constructed, which was trained by extreme learning machine （ELM） algorithm. The improved BFG＇S technique was used to optimize left weights and biases of the network. Then nonlinear regression model was used to construct prediction interval（PI） for reliability parameters based on its point value derived from the trained neural network and the weights of network. So the PI quality depends on the network structure and the weight decay regularizing factor. The immune algorithm was adopted to automate the neural network model selection and adjustment of the weight decay regularizing factor. Model selection and parameter adjustment were carried out through minimization of the PI based cost function called coverage and proportional length based criterion（CPLC）, which combines the coverage probability and the mean interval proportional length of PI. Finally, the proposed theory and method was applied to predict the reliability parameter--mean time between failure（MTBF） of computer numerical control（CNC） lathes, which proved the that prediction performance of the method was better than that of the traditional methods.