Online quantitative analysis of accidental contaminant
in water distribution system based on Dempster-
Shafer evidence theory
Xiang Wen, Dibo Hou*, Tianheng Feng, Pingjie Huang, Guangxin Zhang
State Key Laboratory of Industrial Control Technology
College of Control Science and Engineering, Zhejiang University
Hangzhou 310027, People’s Republic of China
houdb@zju.edu.cn
Abstract—Water quality pollution events in urban
distribution system always show the characteristics as abruptness
and variety, and the ability of qualitative detection of water
quality contamination events is limited in water quality early
warning. Aiming to these, the application of Dempster-Shafer
theory in water quality analysis is studied on, and the
quantitative detection method of contaminant in drinking water
based on water quality parameters is proposed. Combining with
experiment and simulation, the analysis results show that the
method could weaken the dependence of historical data
effectively. Comparing with regular cluster algorithm, the
method adjusts local non-linear variation between contaminant
concentration and water quality parameters and has better
performance.
Keywords—Evidence theory; quantitative analysis;
contaminant; water quality parameters
I. INTRODUCTION
Recently, contaminant accidents in water environment and
water distribution system happen occasionally [1]. Therefore,
the establishment of water quality early warning system is
highest priority now. Water quality early warning system is a
system which is able to predict sporadic or long-term water
pollution events and assist relevant functional departments in
making decisions through techniques as real-time monitoring
[2, 3]. Hence, how to monitor and analyze the water quality in
a distribution system quickly and effectively is a technical
problem which needs to solve instantly, and it’s also the key of
a water quality early warning system.
In recent years, researchers used hydrodynamics models
and multifactorial data analysis technologies comprehensively,
and water quality analysis technologies in urban water supply
were widely studied, the qualitative or quantitative analysis of
water quality in distribution system was mainly realized.
Reference [4] analyzed the variation of water quality
parameters with different contaminants, and the value of
variation was classified, also a linear fit was done between the
change of free chlorine and nicotine’s concentration. Another
further report was given by Hall et al. [5]. In this report, the
influence of conventional water quality parameters caused by
parts of contaminants was analyzed though experiments, and
the influence of the parameters of total and free chlorine,
turbidity, pH, conductivity, TOC, ORP, ammonia and nitrate
brought by potassium ferricyanide, glyphosate, malathion,
aldicarb, Escherichia coli, arsenic trioxide and nicotine was
concluded, also that particular pollutants could reflect by water
quality monitoring indicators was proved. And with different
technologies to measure the same parameter at different time,
similar results were obtained. Upon the researches of John Hall,
numbers of researchers studied on water quality early warning
based on conventional water quality parameters, including
experimental methods and calculated algorithm. In the research
of Szabo and cooperators, a single pass pipe was used to
simulate drinking water distribution system, with the study
between contaminants and water quality parameters, and
similar results were got as those in experiments before [6]. In
the research of Yang and cooperators, the diffusion progress
was modeled in a long pipe. With the injection of contaminant,
the concentration of contaminant was detected with equal time
intervals. The results of the research proved that the change of
water quality parameters is rapid after the injection of
contaminant and steady around the crest value. Also, linear
relationship between contaminants and water quality
parameters was given [7]. These researches proved that it’s
feasible to use experiments to take the place of real-time
detection for researching and the accuracy would not be
impacted on. In terms of algorithm, the cluster algorithm based
on Euclidean distance was used to analyze the water quality
qualitatively, and a relatively accurate result was obtained [8].
Facing to high-dimensional data, PCA algorithm and
multivariate nearest neighbor approach were used to reduce the
difficulty of analysis [9]. Reference [10] studied on the
relationship between contaminant concentration and the
variation of free chlorine with hydrodynamic models and time-
frequency method, and reliably identification was available
within noisy singles. All of the algorithms were used to
distinguish the anomalies of water quality, and well results
were got. However, these researchers mostly focused on the
detection of events and paid little attention to quantitative
analysis.
In practice, the quick detection request of contaminant
concentration is subsistent to make disposal in time, when
water pollution event has happened in a distribution system or
an external water source system and the contaminant is known.