水电能源科学

为了充分利用水电机组振动信号资源，建立更高效的故障诊断模型，提出利用主成分分析(PCA)对振动数据进行降维，基于粒子群算法(PSO)优化目标维度和卷积神经网络(CNN)参数的故障诊断模型。首先将多通道的原始振动数据进行通道层面的降维，再将降维后数据输入CNN网络进行故障诊断分类；其次采用PSO对目标维度和CNN模型中部分关键参数进行寻优，实现信号自适应降维，构建更高效的模型；最后基于寻优结果进行数据降维和模型深入训练，获得最优诊断模型，输出诊断结果。基于某水电机组不同工况下的实测振动数据进行试验对比和分析，验证了所提方法具有较高的诊断精度和稳定性。

In order to make efficient use of the vibration signals of hydropower units and establish a more efficient fault diagnosis model, this paper proposes a fault diagnosis model that uses principal component analysis(PCA) to downscale the vibration data and optimizes the downscale target dimension and the parameters of convolutional neural network(CNN) based on particle swarm algorithm(PSO). Firstly, the original multi-channel vibration data are downscaled at the channel level, and then the downscaled data are input into the CNN network for fault diagnosis and classification. In order to achieve adaptive dimensionality reduction and construct a more efficient model, PSO is used to optimize the target dimensions and the key parameters of the CNN model. Finally, based on the optimization results, the data reduction and in-depth training of the model are carried out to obtain the optimal diagnostic model and output the diagnostic results. The experimental comparison and analysis of the measured vibration data of hydropower units under different working conditions demonstrate that the proposed method has high diagnostic accuracy and stability.