Artificial Neural Network Modelling and Simulation for Gas Leak Detector Performance in Geregu Gas Turbine Power Plant

Abstract

This study is aimed at reducing false alarms in Geregu gas power plant by integrating an Artificial Neural Network (ANN) with the conventional calibrated detectors to improve their performance. The impact of industrial gas leakage to human and environment has necessitated the use of detectors to monitor their presence and concentration. However, the inability of the detectors to differentiate actual gas leaks from unanimous gas figments results to false alarms. Primary causes of false alarms were identified with ambient seasonal factors and human errors as leading indicators among others. An ANN with five inputs, two hidden layers of six and two neurons with an output was developed, trained and validated using the plants historical alarm data from 2019 to 2024. The results obtained shows false alarms occurred almost 100% in January and December which indicates the great influence of ambient condition during harmattan periods. Slight traces of human operation error were also observed. With an epoch of 65, the ANN produced an almost 92.2% output matching with a mean absolute error (MAE) of 0.371, root mean square error (RMSE) of 0.304 and R² representation of 87%. The model thus, demonstrate the effectiveness of gas detectors integrating with artificial intelligence (AI) to predict true gas leak from unanimous causes of false alarm. The ANN, however, cannot completely take the place of conventional calibrators, but will improve the overall system performance.