Process Monitoring Of Complex Industrial Process Based On Probabilistic Graphical Model

Safety production is the basic guarantee for the development of process industry and an important condition for improving economic efficiency.As one of the most commonly used tools in the process industry,process monitoring methods provide a guarantee for the safety production of process industry.The complexity,nonlinearity,non-Gaussian distribution,dynamics and other characteristics of the modern process industry make the performance of traditional process monitoring methods degraded or even impossible to use.In order to adapt to the complex working conditions of the process industry and make better use of process knowledge,this paper proposes a process monitoring method based on graph model.Compared with the traditional methods,the graph model method can make full use of process knowledge and production data to achieve more accurate fault location and diagnosis.The specific research work is as follows:For the steady process,a process monitoring method based on hierarchical probability graph model is proposed.The graph model is drawn according to the conditional dependency between the process variables.According to the hierarchical structure of the graph model,the joint probability density of all the process variables is converted into the product of several conditional probability densities and a low-dimensional probability density,which are calculated by the nonparametric density estimation method.The probability densities are used as monitoring statistics to monitor the process.For the dynamic process,given the existing problems in the process monitoring of dynamic non-Gaussian systems,the state space model is used in this paper to describe the dynamic process and the state variable is introduced as the intermediate node to draw the graph model.Then the probability densities are obtained to realize process monitoring.Further,a fault-reconstruction-based iterative diagnosis method is proposed,and the optimal fault value for each possible fault variable is obtained by an iterative manner.After the corresponding correction,the impact of the fault on the variable is minimized,so as to achieve the separation of the fault variables.For the time-varying dynamic process,a recursive sparse dynamic principal component analysis method is proposed.The sample is sparsely processed by solving the sparse system matrix at each moment.The initial system model is obtained by the optimization algorithm based on the accelerated gradient descent,and the l_2,1,1 norm is introduced to achieve the purpose of row sparse;the system matrix at the next moment is iteratively obtained by the recursive Lasso algorithm.With the sparse processing,the redundancy problem of the data based on the recursive algorithm in PCA modeling is avoided.Then the dynamic PCA is used to model the sparse samples to realize the monitoring and fault diagnosis of the time-varying dynamic process.This method improves the computational efficiency,effectively avoids false alarms and can detect faults in time.Finally,on the basis of summarizing the full text,the challenges and development of the application of the graph model are expounded.