Research On State Inspection Of Industrial Methanol Production Process

Methanol is an important chemical raw material and its applications cover all areas of modern production.As a major coal producing country,China uses the way prepare methanol from coal extensively,so it is very important to ensure the safety and efficiency of the methanol production process.This article first makes a summary of the data collection and modeling methods of industrial methanol production process.Subsequently,probabilistic principal element analysis is used as the entry point for datadriven modelling,and a new regularisation method is proposed to address the problem that the space of latent variables remains interspersed with a large number of irrelevant variables after dimensionality reduction.Unlike traditional probabilistic models,this scheme introduces a mixed prior with spike and slab components to indicate the sensitivity of latent variables to anomalous states.It is able to express more information about the original space while ensuring spatial sparsity.Subsequently,this regularisation scheme is imposed on the latent variable and load matrix respectively,a Bayesian inference scheme based on the expectation maximisation framework is proposed,and a fault detection and diagnosis method based on the inference model is developed.Finally,the two methods are validated separately using industrial production data from the methanol plant of China National Petroleum Corporation（CNPC）as a basis.The experiments demonstrate the superiority of the proposed method over conventional solutions in three directions: data filtering,fault detection and fault localisation,by way of comparison with conventional solutions.Accurate fault diagnosis and localisation is possible when faults exist in a narrow time domain,when variables are interrelated or have delayed characteristics.And in the vast majority of cases,it shows the lowest false alarm and missed alarm rates.It shows that the latent variable space obtained after regularisation is not only more sensitive to the relevant variables,but also eliminates the influence of irrelevant variables.