论文部分内容阅读
为实现对机载设备工作状态的在线状态预测,提出了一种稀疏核增量超限学习机(ELM)算法。针对核在线学习中核矩阵膨胀问题,基于瞬时信息测量提出了一个融合构造与修剪策略的两步稀疏化方法。通过在构造阶段最小化字典冗余,在修剪阶段最大化字典元素的瞬时条件自信息量,选择一个具有固定记忆规模的稀疏字典。针对基于核的增量超限学习机核权重更新问题,提出改进的减样学习算法,其可以实现字典中任一个核函数删除后剩余核函数Gram矩阵的逆矩阵的前向递推更新。通过对某型飞机发动机的状态预测,在预测数据长度等于20的条件下,本文提出的算法将预测的整体平均误差率下降到2.18%,相比于3种流形的核超限学习机在线算法,预测精度分别提升了0.72%、0.14%和0.13%。
In order to realize the on-line state prediction of the working status of airborne equipment, a sparse kernel incremental overrun learning machine (ELM) algorithm is proposed. Aiming at the problem of nuclear matrix expansion in nuclear online learning, a two-step thinning method of fusion construction and pruning strategy is proposed based on instantaneous information measurement. By minimizing dictionary redundancy in the construction phase and maximizing the instantaneous conditional self-confidence of the dictionary elements during the pruning phase, a sparse dictionary with a fixed memory size is chosen. Aiming at the kernel weight renewal problem of kernel-based incremental overrun learning machine, an improved algorithm of reducing the sample size is proposed, which can realize the forward recursive updating of the inverse matrix of the remaining kernel function Gram matrix after any one kernel function in the dictionary is deleted. By predicting the state of a certain type of aircraft engine, the proposed algorithm reduces the overall average error rate to 2.18% when the predicted data length is equal to 20. Compared with the three overmanuated kernel overrun learning machines Algorithm and prediction accuracy have been improved by 0.72%, 0.14% and 0.13% respectively.