LAUSR

Abstract In this paper, the authors propose the use of time windows to improve the detection of fuel leaks in petrol stations. They employ two-class supervised classifiers that work with feature sets containing representative variables taken from station inventory books that indicate the presence of leaks. Fuel leaks in petrol stations with underground tanks pose a serious problem from an environmental standpoint. Large leaks are very evident, and are therefore detected quickly without the need to use a specific procedure. Small leaks, however, tend to go unnoticed, and if no detection techniques are employed, they are only identified once environmental damage has been done. This makes detecting the leak in the shortest time possible as important as ascertaining when the leak started. The authors show how the use of time windows, which entails having the classifier work with information accumulated over several days, can be used to efficiently resolve the proposed problem, fully complying with the applicable regulation.