基于遗传算法SVM的电子元件寿命预测

摘要

针对电子元件在正常应力下的寿命预测，提出了基于遗传算法SVM的预测方法。首先进行多应力水平条件下的寿命实验，得到元件在各个应力下的失效时间，根据失效时间得出相应应力下的可靠性。然后将遗传算法与SVM相结合，建立预测模型，从而不仅可以预测同一应力下元件的寿命，可根据加速应力下元件的寿命来预测正常应力水平下的寿命。实验证明，在小样本条件下，该方法同神经神经网络相比，预测结果的精确度提高了14％，该预测方法能够更准确地预测出电子元器件的寿命。%To accurately predict the life of an electronic component under normal stress, a method based on genetic algorithm SVM is proposed. Firstly, after the test of accelerated life under several stress levels, the components time-out and corresponding reliability can be got. Then combining the genetic algorithm with SVM to build the pre-diction model, which can not only dope out the reliability under the same stress, but also can predict the compo-nents′life under normal stress level according to the life in accelerated life tests. Comparing with neural network this method can exactly predict the life of electronic components under the condition of small samples with improving the precondition accuracy by 14 percent, given in Figs. 1, 3 and 4 and Tables 2 and 3, and their com-parison.