Applying pattern classification techniques to the early detection of fuel leaks in petrol stations

Leaks below the surface are one of the most serious problems in service stations with underground fuel storage tanks. These leaks result in pools of fuel, which flows into both the ground and the aquifers, polluting ecosystems and damaging them severely. In this paper, pattern classification techniques are used to carry out the early detection of fuel leaks in petrol stations. Early detection is crucial from an environmental point of view. The use of these classification methods requires properly selecting those variables that suitably represent the objects to classify, which in this case are the days when the petrol station is operative. In our study we use actual data provided by Repsol (a Spanish energy company) to construct these objects, which are then classified into two possible categories: “day without leaks” or “day with leaks”, applying both supervised and unsupervised classifiers. Finally, three different combinations of “classifier + feature group” are proposed as possible solutions for the problem of the early detection of fuel leaks in service stations.