基于灰色神经网络的设备剩余寿命在线预测

摘要

To solve practical problems that some residual life of highly reliable complex equipment is usually associ-ated with common degradation of multi-performance parameters,a prediction model of residual life,which is the multi-variable grey compensating RBF neural network prediction model,was constructed.Firstly,MGM(1,n)with optimized background value was constructed,and initial predictive values of original data sequences were obtained. Secondly,the mapping relationship between residual sequences and original data sequences was established to train RBF neural networks.Finally,the multi-variable grey compensating RBF neural network prediction model was con-structed to combine improved MGM(1,n)with RBF neural network.Results of case study indicate that the presen-ted method effectively improve the prediction accuracy when comparing with the single prediction model.%为解决高可靠复杂设备的剩余寿命通常与多个性能参数共同退化相关的实际问题，提出一种多变量灰色误差神经网络预测方法。首先，建立经过背景值优化的多变量灰色预测模型 MGM（1，n），并得到原始数据序列的初始预测值。然后，利用神经网络建立残差序列与原始数据序列之间的映射关系，训练 RBF 神经网络。最后，将改进的 MGM（1，n）模型和 RBF 神经网络集成，建立多变量灰色误差神经网络预测模型。实例计算结果表明，与单一预测模型相比，该方法能够有效提高预测精度。