基于ELM-SVR模型的装备关键部件寿命预测

Life Prediction of Key Components of Equipment Based on ELM-SVR Model

装备关键部件数量众多,其性能决定了武器系统的健康寿命。为解决单一模型预测精度不高的问题,提出了基于ELM-SVR模型的剩余寿命预测方法。ELM模型为单隐层网络,输出拟合采用样本权重相同线性回归方法,存在特征辨识度不高,预测泛化能力不强的问题,将ELM模型与SVR模型融合,在ELM神经元基础上使用核函数进一步提高特征的辨识度,以软间隔拟合函数代替线性回归对输出进行拟合,突出了样本权重,有效提升了剩余寿命预测的泛化能力。依据多个装备关键部件的综合性能指标对其进行剩余寿命预测,实验结果表明:与ELM、SVR、KELM模型相比,平均预测精度分别提高了57.14%、44.44%、11.76%,可见ELM-SVR模型具有一定的泛化能力与鲁棒性,可显著提高剩余使用寿命预测精度。

There are a large number of key components in the equipment,and their performance determines the healthy life of the weapon system.In order to solve the problem that the prediction accuracy of a single model is not high,a RUL prediction method based on the ELM-SVR model was proposed.The ELM model is a single-hidden-layer network,and the output fitting adopts the linear regression method with the same sample weight,which has the problems of low feature recognition and weak prediction generalization ability.The ELM model and the SVR model were fused.On the basis of ELM neurons,the kernel function further improves the identification of the features,and the soft interval function was used to replace the linear regression to fit the output,which highlights the sample weight and effectively improves the generalization ability of the remaining life prediction.Based on the comprehensive performance indicators of multiple key components of the equipment,the remaining useful life(RUL)was predicted.The experimental results show that,compared with the ELM,SVR,and KELM models,the average prediction accuracy is increased by 57.14%,44.44%,and 11.76%,respectively.It can be seen that ELM-SVR The model has certain generalization ability and robustness,which can significantly improve the prediction accuracy of RUL.