Operating conditions of steam drums significantly reflect operating performances of coal gasification processes. It is crucial to accurately diagnose the abnormal operating conditions and pay attention to the associated process variables in time. In response to the problem that it is difficult to obtain a large number of abnormal samples of steam drum operating data, which are insufficient for abnormal features learning with traditional methods, an Auto-Encoder neural network based on gated recurrent unit (GRU-Auto-Encoder) is proposed in this paper. The method generates abnormal samples by considering the temporal dependence of data before operating data which contain the generated abnormal samples are provided to GRU neural network for extracting operating conditions features in a deeper and dynamic manner, helping analyze root causes of abnormalities and monitor operating conditions. The effectiveness of method is demonstrated by experiments with operating data of an industrial coal gasification plant.

汽包的运行条件显着反映了煤气化过程的运行性能。准确诊断异常工况并及时注意相关的过程变量至关重要。针对难以获得大量蒸汽鼓运行数据异常样本的问题，而这些样本不足以使用传统方法来学习异常特征，因此基于门控递归单元的自动编码器神经网络（GRU-Auto-本文提出了编码器）。该方法通过在将包含所生成的异常样本的操作数据提供给GRU神经网络以更深和动态地提取操作条件特征之前，考虑数据的时间依赖性来生成异常样本，帮助分析异常的根本原因并监控运行状况。通过工业煤气化装置的运行数据实验证明了该方法的有效性。