Parameter screening and optimized gaussian process for water dew point prediction of natural gas dehydration unit

Petrochemical equipment is characterized by continuity, large scale, complexity of processes, and critical operation conditions. Based on auto-collected monitoring parameters, online prediction of critical process parameters can be used to maintain high reliability of petrochemical equipment, which is generally unpractical due to interference parameters and the difficulty in establishing prediction models. In this paper, a process parameter online prediction method for petrochemical equipment is proposed. Firstly, sensitive parameters are selected applying gradient boosting decision tree (GBDT). Then optimized Gaussian process (GP) is utilized to develop a mapping model in order to inference process parameter from auto-collected parameters. Natural gas water dew point online prediction method for triethylene glycol (TEG) dehydration unit is investigated. The effectiveness of the proposed method is verified on production data of a natural gas dehydration station. The method proposed provides a promising solution for process parameter prediction for petrochemical processes as well as other similar scenarios.