Multiobjective Calibration of Event-Detection Systems

Abstract

CANARY, developed by the U.S. Environmental Protection Agency (USEPA), is a freeware designed for contamination events detection in water distribution systems. CANARY has several imbedded statistical methods for analyzing water quality data to detect contamination events. The imbedded methods require calibration of their parameters to achieve good performance. However, the multiobjective nature of the problem creates a conflict between high sensitivity which results in good detection but with many false alerts, and low sensitivity which results in a small number of false alarms but with poor detection. In this work, a MOGA-CANARY add-in for CANARY autocalibration is introduced. This tool could be used by CANARY users to find the optimal parameter configuration that fits their system. MOGA-CANARY gives the users the whole set of Pareto optimal system configurations based on their defined objectives. The results of MOGA-CANARY are compared with existing manual calibration methods proposed in CANARY documentation.