Chemical Process Fault Detection And Diagnosis Method Based On Long Short-term Memory Network

As one of the pillars of Chinese industry,the chemical industry often has complex process and harsh production conditions,and the chemicals involved in the production process are often toxic,flammable,explosive and volatile.Therefore,it is very important to ensure the safety of chemical process.As one of the important means to ensure the safety of chemical process,fault detection and diagnosis has always been the focus of academic research.However,in the actual chemical production,due to the coupling and large-scale of production equipment,the historical data of chemical process is very complex,which often has the characteristics of high-dimensional,nonlinear and high noise.The traditional fault diagnosis methods do not show good diagnostic performance.Therefore,it is necessary to develop a kind of fault diagnosis technology with good performance.In recent years,due to the good data mining ability of neural network technology,many scholars have combined it with chemical fault diagnosis technology,and achieved good results.Based on the knowledge of chemical process fault detection and diagnosis and long short-term memory network,this paper proposes a chemical fault diagnosis method based on long short-term memory network.In this method,the data features are extracted layer by layer by continuously superimposing the hidden layer of the network,and then input into multiple classifiers for diagnosis and classification.During this period,the parameters are continuously optimized by Adam algorithm,and the over fitting phenomenon is suppressed by dropout layer and other means.Finally,the good diagnosis effect is achieved on TE process data set.In addition,based on the strong feature extraction ability of convolution neural network,this paper also proposes a chemical fault diagnosis method based on convolution long short-term memory network,and verifies it with TE data set.The results show that this method has the advantages of high diagnosis rate,low false alarm rate and high early warning ability,and can timely and effectively carry out fault early warning for TE chemical process.Finally,combined with the historical data of cyclohexanone production process in a chemical plant,the proposed method is verified and its performance is compared.The results show that the proposed method has good diagnosis performance for the actual working condition data,and the convolution long short-term memory network has the best performance.The fault diagnosis rate reaches 93.47%,and the false alarm rate for normal samples is only 0.3%.Therefore,the proposed method can basically ensure the smooth operation of the production section.