1 Introduction: The Problem of Water Quality Events
Classification
According to ISO24522
1 (ISO standard for water quality events detection in drinking water and wastewater systems), a water quality event (henceforth WQE) is
defined as a situation in which “measurements of water quality are not according
to what they are expected to be.” This does not necessarily mean that these
measurements violate regulation rules.
Online monitoring uses several common measurements of water qualities in order
to ensure the safety of using it for drinking and sanitation. The most common
combination is free chlorine, turbidity, and pH as can be seen in several common
systems (http://ec.europa.eu/environment/water/water-drink/index_en.html). However, several other parameters were listed as can be seen in (https://www.epa.gov/
wqs-tech/water-quality-standards-handbook).
Given that, in some countries, the regulatory limit for measurement of turbidity is
1.0 NTU
2
, any water with measurement above this value is not recommended for
drinking. However, in a site in which the average measurement of turbidity is around
0.1 NTU with standard deviation of 0.05 NTU, a measurement of turbidity of 0.75
NTU may be considered abnormal and should be investigated even if it does not
violate regulation limits. If it is known that somewhere in the upstream of the
sampling point, a pipe repair or maintenance work was performed prior to the
event in which such turbidity was measured, this may explain this result and may
cause this event to be “expected.” The problem of identifying WQE becomes more
complicated when measurements include several parameters. In this case additionally to examine each measurement separately, a multiparameter approach should be
used. When using such measurements, it is advisable to use the likelihood of the
combination additionally to values of the individual sensors. See Amit and Brill [1]
and Mounce et al. [2].
Several general methods have been suggested in the past for identifying and
classifying multiparameter abnormal events in general cases. These general methods
include supervised methods such as regression or regression trees and methods that
make use of unsupervised learning such as clustering.
Examples for identifying abnormalities using distance-based or density-based
method have been demonstrated in the past in many areas. Knorr and Ng [3]
demonstrated distance-based methods. Further improvements were demonstrated
by Knorr and Ng [4], for distance-based clustering such as the kMean algorithm.
More recent work in this field is presented in Angiulli and Pizzuti [5] and in
Ramaswamy et al. [6]. The methods proposed in these works are also based on the
kNN algorithm (where kNN stands for multi-k-nearest neighbors). Bay and
1 ISO24522 is under publication procedures and will be available during winter 2019.
2 NTU are the standard units for measuring turbidity.
142
E. Brill
Classification
According to ISO24522
1 (ISO standard for water quality events detection in drinking water and wastewater systems), a water quality event (henceforth WQE) is
defined as a situation in which “measurements of water quality are not according
to what they are expected to be.” This does not necessarily mean that these
measurements violate regulation rules.
Online monitoring uses several common measurements of water qualities in order
to ensure the safety of using it for drinking and sanitation. The most common
combination is free chlorine, turbidity, and pH as can be seen in several common
systems (http://ec.europa.eu/environment/water/water-drink/index_en.html). However, several other parameters were listed as can be seen in (https://www.epa.gov/
wqs-tech/water-quality-standards-handbook).
Given that, in some countries, the regulatory limit for measurement of turbidity is
1.0 NTU
2
, any water with measurement above this value is not recommended for
drinking. However, in a site in which the average measurement of turbidity is around
0.1 NTU with standard deviation of 0.05 NTU, a measurement of turbidity of 0.75
NTU may be considered abnormal and should be investigated even if it does not
violate regulation limits. If it is known that somewhere in the upstream of the
sampling point, a pipe repair or maintenance work was performed prior to the
event in which such turbidity was measured, this may explain this result and may
cause this event to be “expected.” The problem of identifying WQE becomes more
complicated when measurements include several parameters. In this case additionally to examine each measurement separately, a multiparameter approach should be
used. When using such measurements, it is advisable to use the likelihood of the
combination additionally to values of the individual sensors. See Amit and Brill [1]
and Mounce et al. [2].
Several general methods have been suggested in the past for identifying and
classifying multiparameter abnormal events in general cases. These general methods
include supervised methods such as regression or regression trees and methods that
make use of unsupervised learning such as clustering.
Examples for identifying abnormalities using distance-based or density-based
method have been demonstrated in the past in many areas. Knorr and Ng [3]
demonstrated distance-based methods. Further improvements were demonstrated
by Knorr and Ng [4], for distance-based clustering such as the kMean algorithm.
More recent work in this field is presented in Angiulli and Pizzuti [5] and in
Ramaswamy et al. [6]. The methods proposed in these works are also based on the
kNN algorithm (where kNN stands for multi-k-nearest neighbors). Bay and
1 ISO24522 is under publication procedures and will be available during winter 2019.
2 NTU are the standard units for measuring turbidity.
142
E. Brill