Minimum volume ellipsoid classification model for contamination event detection in water distribution systems

Network hydraulics inclusion in water quality event detection using multiple sensor stations data

“…It should be noted that the local threshold, generating the local event alert, is 0.7, which is higher than in Oliker and Ostfeld (2014). The increase of the local threshold is aimed at reducing positive false alarm.…”

“…This section is based on Oliker and Ostfeld (2014). For each sensor's dataset the MVE classifier produces a binary sequence which includes the normal or outlier classification for each time step measurements vector.…”

“…Each data stream is analyzed by a minimum volume ellipsoid (MVE) classifier trained on identifying suspiciously exceptional measurements. The application of the MVE for outlier detection is described in detail in Oliker and Ostfeld (2014). Briefly, the high dimensional ellipsoid (corresponding to the number of measured water quality parameters by each sensor station) is constructed according to the training dataset.…”

“…The coefficients of the measure (i.e., the 0.25 and 0.75 figures) were determined through a calibration process (see Oliker and Ostfeld, 2014). …”

“…The training datasets were used as "off-line" data for the construction of all initial ellipsoids (Oliker and Ostfeld, 2014). The testing datasets were left un-touched for simulating the "on-line" realtime operation of the model for enabling its assessment.…”