Research On Fault Diagnosis And Dynamic Risk Assessment Of Chemical Process

The production technology in chemical process is complex and changeable,and the coupling between variables is strong.Once a fault occurs in one production link,the chain effect between variables will occur,resulting in serious accident consequences.Therefore,how to effectively improve the risk control level of chemical accidents,timely determine the cause of chemical process failure and accident path,is the key to ensure the safe operation of chemical production process.In this paper,the fault diagnosis and dynamic risk assessment of chemical process are studied.The main work of this paper is as follows:（1）Two improved data preprocessing methods,IFA feature extraction and RFRFECV feature selection,are proposed in this paper.Aiming at the characteristics of high dimension and coupling of fault data,IFA determines the common factor number of factor analysis based on variance contribution rate of principal component analysis,and obtains new features by spatial mapping of original features.RF-RFECV selects the most representative feature combinations with good classification performance from the original features into subsets based on random forest combined with cross-validation recursive feature elimination algorithm.The feature dimension reduction of fault data is realized effectively.（2）A binary fault diagnosis model based on PSO-KSVM is proposed in this paper.On the basis of the gaussian kernel support vector machine（KSVM）model,particle swarm optimization（PSO）is used to optimize the model parameters for the problem that the penalty parameter C and the kernel function parameter σ are not easy to determine.The simulation results of Tennessee-Eastman（TE）process show that the KSVM model with optimized PSO parameters has better fault diagnosis ability.（3）A hierarchical multi-classification fault diagnosis model based on IH-PSOKSVMs is proposed in this paper.Aiming at the difficulty and low accuracy of traditional KSVM multi-classification model modeling,RF-RFECV is combined to make feature selection for each type of fault data.By introducing k-means algorithm and similarity measurement method,the class center point and difference of each fault feature subset are determined,and a hierarchical classification model is built.The simulation results of TE process show that the IH-PSO-KSVMS model takes into account the advantages of simple structure and stable performance of traditional models,and meets the diagnosis requirements of multiple types of faults.（4）A dynamic risk assessment method for tank leakage based on Bayesian network is proposed in this paper.Aiming at the problem that traditional Bow-Tie analysis method（BT）cannot capture the dynamic risk characteristics of chemical process.Taking the tank leakage event in the automatic feeding system of a pharmaceutical company as an example,the BT model is mapped to Bayesian network（BN）to fully combine the advantages of both.By introducing Euclidean distance and improved A-Star algorithm,the key risk factors of the system and the most likely path of tank leakage are identified,and the risk control measures are given from three parts: technology,management and individual training.