Using Artificial Neural Network to Predict Gas Flare Volume, Gas Composition, and Temperature from Gaseous Emission.

Abstract

Abstract Gas emission from gas flaring is known to have deleterious effects on the environment and they constitute a major source of global warming. Flare gas volume, temperature, gas composition and other meteorological factors are major parameters in gas flaring processes. Using Artificial Neural Network, a model was developed to estimate gas flare volume, gas composition and flare temperature from gaseous pollutants using. Data used for the study were from oil prolific Niger-Delta region of Nigeria. Air quality index parameter, gas flared volume, temperature and composition between 2013-2017 were used in developing the ANN model using the Neural Training toolbox (nntool) of the Matrix Laboratory (R2019a MATLAB) mathematical software. An 8-6-3 network architecture was adopted. It consists of eight input parameters (suspended particulate matter, carbon monoxide, sulphur oxide, nitrogen oxide, volatile organic compounds, hydrogen sulphide, methane, and ammonia), six hidden layers and three output parameters (gas flared volume, gas composition and temperature) using 1286 dataset for each input and output parameter. Multiple-input multiple-output (MIMO) neural network using supervised learning algorithm (Levenberg-Marquardt) to train the network was adopted in model development. 75% (880 data points) of the data was set aside for the training of the model at its developmental stage, 10% for test data set and 15% for the validation data set. From the models’ prediction, it was observed that the developed model predicted excellently and performed well when tested with new set of data which was not a part of the developmental dataset with a coefficient of determination of 0.99999918, a root mean square error of 0.009029, an absolute average percentage relative error of 0.0362% for Gas Flare Volume, Composition and Temperature respectively. The outcome of this study presents a reliable and speedy tool for forecasting of gas flare volume, composition, and temperature in the absence of conventional methodologies.