Study on Fault Diagnosis Methods for Chillers Combining LSTM and 1DCNN
In chillers operation,the sensor collects data with time series characteristics.To accurately identify the fault data characteristics of chillers from the time dimension and address the problem of degrading fault diagnosis performance of chillers caused by covariate shift in neural networks,this paper proposed a deep learning fault diagnosis method combing One-dimensional Convolutional Neural Network(1DCNN)and Long Short-term Memory Network(LSTM).The fault experimental data of ASHRAE RP-1043 was used as the method model training and test samples to explore the influence of network model parameters on diagnostic performance.The model combines the advantages of 1DCNN to extract local features of samples and LSTM processing sample time series,and introduces layer normalization(LN)technology to solve the problem of network training overfitting caused by covariate shift in neural networks,effectively improving the performance of fault diagnosis of chillers.The experimental results show that the fault recognition accuracy of the proposed 1DCNN-LN-LSTM model reaches 98.85％,significantly improving the diagnostic performance compared with single deep learning models such as 1DCNN and LSTM.
NETL
NSTL
万方数据
