中文摘要：

针对小功率LED失效时间预测需要大量失效数据、预测成本较高的情况,从工程实践的实际需求出发,提出了基于改进蛙跳算法（ISFLA）优化支持向量机（SVM）的小功率LED寿命预测模型。SVM使用径向基核函数（RBF）,以小功率LED的可靠度作为SVM输入,失效时间作为输出。为了减小蛙跳算法的计算成本,现引入进化阶段指标T加快收敛速度,并利用ISFLA优化核函数参数的选取,使得仅需少量训练样本即可建立SVM。该模型不仅能预测加速应力下的小功率LED失效时间,也可根据小功率LED在加速应力下的失效时间预测正常应力水平下的寿命。实验证明,在小样本条件下,该方法得到训练结果的相关系数为0.998,检验组误差小于3%,快速简洁地实现了小功率LED高精度寿命预测。

英文摘要：

The failure time prediction of low power LED requires a lot of failure data, and the pre- dicted cost is high. In view of the problems above, based on the improved shuffled frog leaping algorithm （ISFLA） , a model of the low power LED life prediction optimized by support vector machine （SVM） was presented. SVM uses the radial basis function （RBF） with reliability as the input and failure time as the output. In order to save calculating cost of the shuffled frog leaping algorithm, the evolutionary stage index T was introduced in algorithm to improve the constringeney speed. The method used ISFLA to opti- mize the kernel function parameters, so that only a few samples were needed to build SVM model. This method can predict the failure time of low power LED under an accelerated stress, and predict the life under normal stress with the failure time data on accelerated stress experiment. The experiment results show that under small samples conditions, the correlation coefficient of this method reaches 0. 998, and the inspection group error is less than 3%. The method quickly realized the prediction of high precision life of the small power LED.