基于EMD-KPCA-LSTM的抽水蓄能机组振动预测

Vibration Forecasting of Pumped Storage Units Base on EMD-KPCA-LSTM

针对抽水蓄能机组振动信号时间序列高度非线性、非平稳性等导致常规预测方法难以准确预测的问题,构建了结合经验模态分解(EMD)、由主成分分析(PCA)改进的核主成分分析(KPCA)和长短期记忆神经网络(LSTM)的抽水蓄能机组振动预测模型。该模型利用EMD算法首先将振动信号进行分解,再利用KPCA筛选出关键影响因子,最后通过LSTM对特征序列进行时间动态建模,实现对抽水蓄能机组振动预测。试验结果表明,所建模型相较传统的LSTM、EMD-LSTM等预测模型具有更好的预测效果,可以更精确地预测振动信号的变化趋势。

In the vibration trend prediction of pumped storage units,the conventional prediction methods are difficult to predict accurately due to the highly nonlinear and non-stationary vibration signal time series.In this paper,a vibration prediction model of pumped storage unit is proposed,which combines empirical mode decomposition(EMD),kernel principal component analysis(KPCA)improved by principal component analysis(PCA)and long short-term memory neural network(LSTM).The model uses EMD algorithm to decompose vibration signals,and the KPCA is used to screen out the key influencing factors.Finally,the LSTM is used to carry out time-dynamic modeling of feature sequences to realize vibration prediction of pumped storage units.Compared with the traditional LSTM,EMD-LSTM and other prediction models,the experimental results show that this model has better prediction effect and can predict the change trend of vibration signals more accurately.