The detection of unaccounted natural gas consumption: A neural networks and subscriber-based solution

Abstract

Energy is a paramount expense for nations, emphasizing the significance of energy access and conservation. This study proposes using neural networks and statistical methods to identify unaccounted natural gas consumption. Companies gauge gas usage through real-time calculations with A-type remote measurement stations (RMS-A) and monthly reflected bills. Unaccounted consumption, resulting in economic losses, stems from the disparity between these methods. The causes for this can include gas leaks, pressure changes, inactive subscribers, and faulty meters. This study focuses on unaccounted consumption at gas meters faulty meters. Big data, including subscriber data, meteorology, and calendar information, was used for predictions. Two neural network models estimated subscriber consumption, with gas meter replacement details influencing the predictions. Statistical techniques categorized subscribers based on MAPE values into safe, transition, and abnormal zones. Subscribers in the abnormal zone are pointed out to have a high risk of lost and unaccounted natural gas consumption.