Predictions Of Abnormal Operating Conditions Of Steam Drums In Coal Gasification Plants Based On Gru-auto-encoder Neural Networks

The steam drum water level is considered as a key safety parameter in the process of the drum operations.The operating conditions of steam drum can reflect the operating performances of the steam drum process.Through the prediction of the abnormal operating conditions,the corresponding process variables can be monitored in time to reduce the risk of accidents,so as to ensure the safe operation of the plants.In response to the problem that it is difficult to obtain a large number of abnormal samples of steam drum operation data,methods for prediction of abnormal operating conditions of steam drum in coal gasification plants based on Gated Recurrent Unit&Auto-Encoder(GRU-Auto-Encoder)neural networks are proposed in this paper.The main research contents and achievements are presented as follows.1.Due to the lack of consideration of anomaly data timing dependence,the accuracy of reconstruction of traditional Auto-Encoder neural networks decreases with the increase of sequence length,which is not suitable for reconstructing long sequence anomaly data.By combining the long-term memory capability and excellent nonlinear fitting of GRU neurons,the GRU-Auto-Encoder neural networks are constructed.Considering the influence of the previous state information on the current input data in the reconstruction process to realize effective memory and reconstruction of long sequence anomaly data.The validity of anomaly data reconstruction is verified through Tennessee Eastman(TE)process data.2.Analyze the characteristics of steam drum data,a method for predicting the abnormal water level of steam drum in coal gasification plants based on GRU-Auto-Encoder neural networks is proposed.The predicted data are reconstructed by GRU-Auto-Encoder neural networks to achieve abnormal water level prediction.The effectiveness of the method is verified by steam drum operation data of an industrial coal gasification plant.3.Based on the predicted abnormal water level data of steam drum,a method for predicting the abnormal operating conditions of steam drum based on GRU-Auto-Encoder neural networks is proposed.GRU-Auto-Encoder neural network is used to learn abnormal features and generate abnormal samples.Thereby,the abnormal operating conditions are diagnosed for the predicted abnormal water level data.Through the method verification of steam drum operation data of industrial coal gasification plants,the prediction accuracy of abnormal operating conditions is effectively improved when the abnormal sample information is insufficient.