基于GAF-CNN的机组振动信号特征提取方法研究

Research on Feature Extraction Method of Hydropower Unit Vibration Signal based on GAF-CNN

随着水电机组运行时间增长,机组运行数据不断增加,在机组健康状态识别过程中存在着健康样本过多、特征参数不明显等问题。本文结合格拉姆角场(GAF)与卷积神经网络(CNN)在特征表达与提取方面的优势,对水电机组健康运行数据进行处理。通过格拉姆角场将机组振动信号进行编码并生成相应特征图像,进而将其输入至卷积神经网络(CNN)模型以达到特征提取及分类的目的。使用仿真数据与机组实测数据,将GAF-CNN模型与传统长短期记忆(LSTM)网络模型进行对比,结果表明,GAF-CNN模型的特征提取方法具有更高的准确度与鲁棒性,在面对更长时间序列数据时依然能保持良好的准确度与抗噪性能,为水电机组健康评估模型性能提升提供数据基础。

As the operational lifespan of hydropower units increases,the volume of operational data correspondingly expands.This growth presents several challenges in identifying the health status of these units,including an overabundance of health samples and ambiguous characteristic parameters.This study proposes a method that combines the strengths of the Gram Angular Field(GAF)and Convolutional Neural Network(CNN)in feature representation and extraction to process the healthy operational data of hydropower units.The vibration signal of the unit is encoded by the GAF to generate a corresponding feature image,which is then input into the CNN model for feature extraction and classification.The performance of this GAF-CNN feature extraction method is compared with the traditional Long Short-Term Memory(LSTM)network model using both simulated and measured unit data.The results demonstrate that the feature extraction method of GAF-CNN model has higher accuracy and robustness,and can still maintain good accuracy and anti-noise performance in the face of longer time series data,which provides data basis for the performance improvement of hydropower unit health assessment model.