关键词: 卷积神经网络, 长短期记忆, 风力发电机, 轴承损伤, 损伤状态预警, 趋势项消除, 卷积操作

Abstract: When the bearings of wind turbines are in a declining trend, the quality of the bearing status signal will decrease, resulting in a mixed sequence of damage state characteristics and a high coefficient of variation. A wind turbine bearing damage warning method based on convolutional neural network-long short-term memory ( CNN-LSTM) is proposed for this purpose. Based on the status signal of wind turbine bearings, set the fitting coefficient to fit and eliminate the trend term in the signal. The convolutional layer of convolutional neural network( CNN) is used to extract signal features, and the pooling layer is applied for dimensionality reduction to generate a sequence of damage state features, which is input into
long short-term memory( LSTM) to update the memory units and capture the temporal relationship of the damage state feature sequence. By classifying temporal relationships, calculating the risk value of bearing damage status signals, determining the damage status of wind turbine bearings, and dividing warning levels. The experimental results show that the deviation of bearing status signals is significantly reduced under the application of this algorithm, and the damage status is basically the same as the warning level. The coefficient of variation is low with a mean of 0. 03, achieving accurate warning of bearing damage status.

Key words: convolutional neural network, long short-term memory, wind power generator, bearing damage, damage status warning;trend item elimination;convolution operation