APPLICATION OF MACHINE LEARNING TECHNIQUES TO DAMAGE ESTIMATION AND EMERGENCY SHUTOFF IN CITY GAS SUPPLY SYSTEM

Abstract

 Applicability of machine learning techniques has been examined for estimation of damage of low-pressure gas pipelines and decision making of emergency shutoff of city gas supply. A number of observation patterns of SI values, damage rate and shutoff patterns was generated by Monte Carlo simulation. The relationships between SI values and damage rate in training data was learned using support vector regression analysis. The relationships between SI values and shutoff status was learned using support vector machine. The results using test data suggests that the applied techniques can be promising tools for representing non-linear relationships among those factors related to damage estimation and shutoff decision.