Design of a predictive emission monitoring system for natural gas plant using artificial neural network.

摘要

The primary objective of this work is to design a predictive emission monitoring system (PEMS) for a natural gas processing unit from an existing natural gas plant using artificial neural networks (ANN). The processing unit of interest was the multi stage compression unit composed of three primary compression stages which are considered one of the major GHGs emission sources in the plant. The modelling system was designed so as to predict the emission rate of CO2, CH4 and N2O from each emission source individually. The modelling system consisted of three network models each predicting the generated emissions individually rather than creating a single network that predicts the generated emissions from each source simultaneously. Moreover, this work contrasted the effect of utilizing three network structures namely multi-layer perceptron, cascade feed forward and generalized regression networks. Along with various network related parameters such as training algorithm, activation function and number of neurons in hidden layer. The designed networks for each emission source were contrasted to linear and non-linear regression models. It was found that the performance of ANN to all sub-models was far more superior to linear and non-linear regression models, due to its ability to capture the behaviour of the intended relationship between process parameters and emission rates of the three criteria pollutants. Optimal models for each emission sources based on ANN were found through trial and error and adjusting network related parameters. This assisted in establishing some general set of criteria towards the design of PEMS models using ANN for future works. Moreover, the results of this work can assist in future works aimed at designed more universal ANN based PEMS models that can be utilized for different operating conditions and process configurations.