论文部分内容阅读
为实现对液压泵特征参数的在线预测,提出一种贯序正则极端学习机(SRELM),并研究了基于SRELM的预测方法。SRELM根据结构风险最小化原理实现网络训练,其网络权值可随新样本的逐次加入而递推求解,具有泛化能力强与训练速度快的优点,因此适于特征参数的在线预测。基于SRELM的预测方法利用特征参数训练SRELM模型,以逐次增加新数据的方式对SRELM模型进行在线训练,并利用训练后的SRELM模型对未来时刻的特征参数进行外推预测。液压泵特征参数预测实例表明,基于SRELM的特征参数预测方法具有预测精度高与计算效率高的优点,其综合性能优于基于传统迭代式神经网络的预测方法与基于支持向量机的预测方法。
In order to realize on-line prediction of characteristic parameters of hydraulic pump, a sequential regular extreme learning machine (SRELM) is proposed and the prediction method based on SRELM is studied. SRELM realizes network training according to the principle of minimizing structural risk. Its network weights can be recursively solved with successive additions of new samples. It has the advantages of extensive generalization ability and fast training speed and is therefore suitable for on-line prediction of feature parameters. The SRELM-based prediction method trains the SRELM model by using the characteristic parameters, and the SRELM model is trained on-line by adding new data one by one. The SRELM model is used to extrapolate the characteristic parameters at a future moment. The example of characteristic parameters prediction of hydraulic pump shows that the characteristic parameter prediction method based on SRELM has the advantages of high prediction accuracy and high computational efficiency. Its comprehensive performance is superior to the traditional iterative neural network-based prediction and support vector machine-based prediction.