Abstract:
In order to ensure the safe and stable operation of production processes and avoid losses caused by faults, it is quite important to detect abnormal working conditions in time and diagnose them accurately. Aiming at the complexity of chemical processes, this paper proposes a parallel long and short-term memory network and convolutional neural network (PLSTM-CNN) model for fault detection in chemical production process. By combining the LSTM's ability to extract global features from time series data and the CNN model's ability to extract local features, this model can effectively reduce the loss of feature information and achieve a high fault detection rate. Meanwhile, by using one-dimensional dense convolutional neural network as the main body of CNN and combining the LSTM network's sensitivity to sequence information changes, a deeper network can be built while avoiding model over fitting. Besides, the maximum mutual information coefficient (MMIC) data preprocessing method is adopted to improve the local correlation of the data and improve the efficiency of the PLSTM-CNN model in detecting faults under different initial conditions. Finally, it is shown from the experiment results on Tennessee Eastman (TE) process that the PLSTM-CNN model is obviously superior to the traditional recurrent neural network in such indicators as average failure detection rate and false negative rate.