Identification of axis orbit type of hydropower unit based on convolution neural network
XU Jingjun1, ZHENG Yuan2*, YU Yang3, PAN Hong1, TANG Wei2
1. College of Energy and Electrical Engineering, Hohai University, Nanjing, Jiangsu 211100, China; 2. Institute of Innovation, Hohai University, Nanjing, Jiangsu 210098, China; 3. China Water Northeastern Investigation, Design & Research Co. Ltd., Changchun, Jilin 130021, China; 4. College of Water Conservancy and Hydropower Engineering, Hohai University, Nanjing, Jiangsu 210098, China
Abstract:Aiming at the problem of low efficiency and manual participation in fault type identification of conventional hydraulic units. With the help of rich information contained in the images of shaft center track and on the basis of the introduction of fine-grained model to distinguish the fault severity. An intelligent identification method of shaft center track type of hydraulic units based on convolution neural network was proposed. In this method, four kinds of fault severity evaluation criteria and the corresponding two kinds of fine-grained database of shaft orbit of hydropower units were established. A convolution neural network model with improved parameters of over convolution layer and pooling layer was used to simulate the database, and the comparison analysis was made with full connection network. The results show that the recognition rate of fault type and severity of the model is 98.75% and 98.33% respectively. The dimensionality index of fine-grained classification proposed by this method has better state description ability than dimensionless one index. It is also in line with the development trend of fault diagnosis of hydraulic units. Therefore, the axis trajectory identification algorithm based on convolution neural network technology has important value for fault diagnosis of hydraulic units.
徐晶珺,郑源*,于洋,潘虹,唐魏. 基于卷积神经网络的水电机组轴心轨迹类型识别[J]. 排灌机械工程学报, 2021, 39(5): 471-476.
XU Jingjun,ZHENG Yuan*,YU Yang,PAN Hong,TANG Wei. Identification of axis orbit type of hydropower unit based on convolution neural network. Journal of Drainage and Irrigation Machinery Engin, 2021, 39(5): 471-476.