时序数据驱动的化工过程风险动态预警研究

Chemical process risk warning based on CNN LSTM and fuzzy mathematics

对化工过程进行在线监测与动态风险预警是降低事故发生的有效途径。提出了一种基于深度学习时序预测与模糊数学定量风险评估相结合的预警方法。针对化工过程数据的动态性、时序性、非线性强,且预测周期短等问题,将卷积神经网络(Convolutional Neural Networks,CNN)与长短期记忆网络(Long Short-Term Memory,LSTM)模型结合形成深度学习时序预测模型,实现过程参数108 min的超前预测。将该方法应用于合成氨过程,对温度、压力、流量、氢氮比等6个风险参数进行预测。结果表明,该预测方法具有较高的预测精度,其线性回归相关系数及均方根误差表明所提出的方法具有非常高的精度。同时利用三角模糊数对时序预测结果进行风险评估,得到时序风险变化曲线,实现了化工过程风险预警。研究对使用人工智能和大数据实现过程控制和风险预警进行了有益探索,为实现化工过程的超前预警提供参考。

Online monitoring and dynamic risk warning of chemical processes are effective ways to reduce accidents.In this study,an early warning method combining time series prediction based on deep learning and quantitative risk assessment by fuzzy mathematics was proposed for the first time.The model predicted the values of chemical process parameters in a certain period in the future and used triangular fuzzy numbers for quantitative risk assessment to achieve risk classification of the predicted data.For the dynamic,time-series,strong nonlinearity,and short prediction period of chemical process data,a Convolutional Neural Network(CNN)was combined with a Long Short-Term Memory(LSTM)network model to form a deep-learning time series prediction model and achieved a 108-minute advance prediction of process parameters.Taking the synthetic ammonia process as the research object,through expert review,six risk parameters such as temperature,pressure,flow rate,and hydrogen to nitrogen ratio were selected through expert review,and each group of time series data was set to 1080 minutes.Sequence data were extracted from the Distributed Control System(DCS)to predict 6 risk parameters.The Support Vector Machine(SVM),CNN,and LSTM models were used to compare and analyze the prediction results,which showed good reliability and accuracy of the method for prediction results of the synthetic ammonia process parameters.Besides,the triangular fuzzy number was used to evaluate the risk of the time series prediction results and obtained the time-series risk change curve to realize the chemical process risk early warning.The method proposed in this study is a useful exploration of the use of artificial intelligence and big data to achieve process control and risk early warning,which is of great significance for obtaining advanced early warning of chemical processes.