Decision trees development for leak detection on gas transmission system using stationary model and machine learning from examples

One approach in solving the important problem of leak detection on gas transmission system, involving stationary simulation model and artificial intelligence techniques, is presented in the paper. This approach requires a lot of training examples, which were generated using data records from the real pipeline network and leak simulation technique proposed in the paper. Then feature (or attribute) extraction is presented with a complete set of attribute candidates. Leak classification rules were derived in the form of decision trees developed by the system for machine learning from examples. For some proposed detection strategies the accuracy of leak classification reached in DP Gas pipeline network is over 90%. Illustrative examples are included