Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems
An approach to forecasting the behaviour of dams according to the data of instrumental observations with regard to capabilities of automated monitoring systems has been proposed. The approach is based on the use of situational and inductive models, where the situational models correspond to selectiv...
Збережено в:
| Опубліковано в: : | Математичне моделювання в економіці |
|---|---|
| Дата: | 2017 |
| Автор: | |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
Інститут телекомунікацій і глобального інформаційного простору НАН України
2017
|
| Теми: | |
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/131920 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems / D.V. Stefanyshyn // Математичне моделювання в економіці. — 2017. — № 3-4(9). — С. 50-61. — Бібліогр.: 7 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| _version_ | 1859956133959041024 |
|---|---|
| author | Stefanyshyn, D.V. |
| author_facet | Stefanyshyn, D.V. |
| citation_txt | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems / D.V. Stefanyshyn // Математичне моделювання в економіці. — 2017. — № 3-4(9). — С. 50-61. — Бібліогр.: 7 назв. — англ. |
| collection | DSpace DC |
| container_title | Математичне моделювання в економіці |
| description | An approach to forecasting the behaviour of dams according to the data of instrumental observations with regard to capabilities of automated monitoring systems has been proposed. The approach is based on the use of situational and inductive models, where the situational models correspond to selective series of dynamics of observed data within limited time intervals and the inductive models which are constructed on model data derived from situational models simulate evolutions of diagnostic parameters.
Запропоновано підхід до прогнозування поведінки гребель за даними інструментальних спостережень з врахуванням можливостей автоматизованих систем моніторингу. Підхід ґрунтується на використанні ситуаційних та індуктивних моделей, де ситуаційні моделі відповідають вибірковим рядам динаміки спостережених даних на обмежених інтервалах часу, а індуктивні моделі, що будуються за модельними даними, які визначаються з ситуаційних регресійних моделей, відображають еволюції діагностичних параметрів.
Предложен подход к прогнозированию поведения плотин по данным инструментальных наблюдений с учетом возможностей автоматизированных систем мониторинга. Подход основан на использовании ситуационных и индуктивных моделей, где ситуационные модели соответствуют выборочным рядам динамики данных наблюдений на ограниченных интервалах времени, а индуктивные модели, строящиеся по модельным данным, которые определяются при помощи ситуационных регрессионных моделей, отображают эволюции диагностических параметров.
|
| first_indexed | 2025-12-07T16:19:40Z |
| format | Article |
| fulltext |
~ 50 ~
Mathematical modeling in economy, №3-4, 2017
UDC 504.75 + 004.942; 519.25; 621.791: 626/627
D.V. STEFANYSHYN
IMPROVING DIAGNOSTIC MODELS FOR FORECASTING
THE BEHAVIOR OF DAMS EQUIPPED WITH AUTOMATED
MONITORING SYSTEMS
Abstract. An approach to forecasting the behaviour of dams according to
the data of instrumental observations with regard to capabilities of
automated monitoring systems has been proposed. The approach is based
on the use of situational and inductive models, where the situational models
correspond to selective series of dynamics of observed data within limited
time intervals and the inductive models which are constructed on model
data derived from situational models simulate evolutions of diagnostic
parameters.
Keywords: automated monitoring systems, dams, dependent and
independent variables, instrumental observations, inductive and situational
models, long-term and real-time forecasting, monotonic and non-monotonic
data series of dynamics.
Introduction
The long experience the construction of dams shows that accidents on these
structures can lead to serious negative socioeconomic and environmental
consequences including those of catastrophic proportions. Therefore, the problems
of reliability and safety of dams are given considerable attention. At the
international level the main work in this sphere is conducted by the International
Commission on Large Dams (ICOLD). One of the most important challenges
which are solved by engineers to maintain reliability and safety of dams is the
creation of effective systems for monitoring of dams condition. The importance of
such systems for ensuring reliability and safety of dams was repeatedly emphasized
in the past. In particular, Bulletin 59 ICOLD (“Dam Safety – Guidelines” [1]) says
that the majority of damaged dams had no monitoring systems or those systems
were imperfect.
The problem with proper functioning of systems for monitoring of dams
condition is a complex one and its solution does not only lie in sphere of the
introduction of up-to-date equipment of automation and computerization. It should
be noted the modern automated monitoring systems (AMS) which are installed at
dams are not able to directly perform the functions of ensuring reliability and
ANALYSIS, EVALUATION AND FORECASTING
IN ECONOMY
Ó D.V. Stefanyshyn, 2017
~ 51 ~
Mathematical modeling in economy, №3-4, 2017
safety of the engineering facilities during operation. This is due to the fact that
monitoring of dams can never be sufficient enough to include all possible
influencing factors, important characteristics, parameters, elements and
components, the condition of which may affect the overall condition of dams. The
most modern types of instrumental control and samples of instrumentation installed
at dams allow for doing monitoring of a relatively small number of factors and
parameters. As a rule, monitoring is exercised to separate sections, cross-sections,
etc. In addition the most advanced AMS is incapable of ensuring the proper
modeling of dams’ condition yet, which would allow predicting the future behavior
of the waterworks. Adequately, they can only perform functions to storage of
relevant data and control the state of instrumentation.
In this case, a new task arises which consists in ensuring the processes of
modeling and forecasting of dams condition based on observational data under new
circumstances when data may be collected in the great amount. Previously, when
data were collected manually they were considered to be limited and insufficient to
build adequate mathematical models. But without improving of approaches and
methods for modeling and forecasting based on observational data, new
capabilities of automated monitoring of dams condition are substantially
minimized too. Practice shows that large amount of data does not always contribute
to the quality of traditional regression models, whose accuracy can degrade. It was
found that complex and well structured mathematical models based on
observational data in conditions of large arrays of input data do not provide desired
results [2]. In particular, the optimization principle, which lies at the basis of
construction of traditional mathematical models based on observational data,
requires for the systems under study to be in certain boundary limits. If there is a
need to have taken into account large amount of data, this principle cannot be easy
performed. As a result, challenges associated with the stability of solving
optimization problems can even arise in the simplest of cases. Increased quantity of
data in case of the traditional approach requires an increase in models dimension
by taking into account additional factors and non-linear effects, etc. This leads to
disruption the stability of complex models and they can not be used for forecasting
purposes.
1. Basic principles of technical diagnostics on dams and principal challenges of
regression modeling for forecasting purposes
The basic principles of technical diagnostics and monitoring of condition of
technical systems, which were formulated by R.A. Collacott [3], are as follows:
1) Consistency and regularity (continuity) of measurements for characteristics
which are selected as diagnostic parameters;
2) Detection of changes in behaviour of these parameters over time;
3) Predicting and forecasting of behaviour of the system which is under
monitoring with taking into account these changes.
Automated monitoring systems allow maintaining regular and systematic
measurement of diagnostic parameters and storing different data in sufficient
quantities to form representative data samples for any situation and any time
interval that can be considered in terms of monitoring changes in the environment
and in behavior of dams. However, the experience of operating the system at the
Kiev dam has showed that implementation of the two following above mentioned
~ 52 ~
Mathematical modeling in economy, №3-4, 2017
principles requires revision of traditional approaches to modeling based on
observational data which are accumulating in large volume due to the increased
capabilities of AMS. This is because the typical diagnostic models which are used
for predicting and forecasting of condition of dams which are in operation are
models of regression type.
Regression models determine the dependence of the mean value of some
random variable y , which is accepted as a diagnostic parameter or as a dependent
(endogenous or resulting) variable, from the other random variable x or the
several such variables mj xxxx ,...,,...,, 21 which are called independent
(exogenous, explanatory) variables. The choice of an adequate regression model is
based on the minimization of functional which is usually written as the sum of
squared deviations yye ii -= of the model values ),...,,( 10 jxaafy = of the
diagnostic parameter y from observed values iy , where jx , mj ,1= , are
independent parameters of the model with total number m :
( ) ( ) min,..., 10
2
11
2 ®=-=åå
==
aafyye
n
i
i
n
i
i , (1)
where ,..., 10 aa are required coefficients of the corresponding regression model. In
this case the structure of the model is considered known.
The use of regression analysis in modeling according to empirical data can be
right if certain requirements (boundary limits) are fulfilled, in particular [2]:
1) Data of observations for a dependent variable are random values which
follow normal distribution;
2) Independent variables mj xxxx ,...,,...,, 21 are measured with errors which
can be neglected compared to the error of dependent variable y ;
3) The factors mj xxxx ,...,,...,, 21 are random variables that are not correlated
to each other;
4) Random values ni yyyy ,...,,...,, 21 of the resulting variable y should be
obtained in the same conditions.
The modern regression analysis enables to simplify significantly the task of
modeling with using of empirical data for on-line diagnostics of dams condition
during their operation. This eliminates the need while modeling the causal
relationships between different variables of solving more complex problems of
structural and parametric identification of phenomenological models of processes
and phenomena that determine behaviour of dams using systems of equations for
the theory of elasticity, thermal conductivity, filtration theory, fluid mechanics, etc.
with the relevant conditions of uniqueness [4]. This approach to technical
diagnostics of dams on the basis of data of instrumental observations is the
common one in the international practice. However, if the data of observations are
heterogeneous, the construction of adequate regression models for forecasting
purposes can be a serious challenge even in simple cases [5].
Searching for unknown coefficients of regression models, according to (1), is
carried out so that the model in the statistical sense would better meet to empirical
~ 53 ~
Mathematical modeling in economy, №3-4, 2017
data. That is, to solve the problem (1) the principle of optimization is fulfilled with
taking account the compliance with the above boundary limits [2]. But in practice,
if the data are heterogeneous, these restrictions can not usually be performed. In
this case the increase in the number of observations can disturb the execution of
limit restrictions which modeling requires.
For example, Fig. 1 shows a dynamic series of daily observations of upstream
water level (UWL) which was taken as an independent variable (a), and the
scattering field of values of water level in a piezometer (PWL) (b) which was
considered as a dependent variable on random values of UWL. As we can see there
is a strong non-monotonicity of UWL values and there is a significant
heterogeneity of PWL values depending on UWL.
a)
101,00
102,00
103,00
104,00
31.10.2002 14.03.2004 27.07.2005 09.12.2006 22.04.2008 04.09.2009
Time, dates
U
W
L,
m
b)
97,50
98,00
98,50
99,00
99,50
101,00 101,50 102,00 102,50 103,00 103,50
UWL, m
PW
L,
m 2003-2009 years
Fig. 1 – Illustration of non-monotonicity in observations of upstream water
levels (UWL) (a) and heterogeneity of water levels in a piezometer (PWL)
depending on UWL (b)
Increasing the structure dimension of a regression model by introducing into
the model of additional independent variables cannot usually solve the problem of
heterogeneity of variance. The presence of correlation between different
independent variables in multivariate models (we know it as the multicollinearity
problem) may become an additional challenge. And we know that under
multicollinearity conditions the regression coefficients become highly unstable to
small changes in the data, which violates the stability of solutions in the search for
the unknown coefficients of regression models. Constructing such models like
autoregressive models, distributed-lag models, etc., does not always bring success
too.
2. A concept of situational regression models as the main idea of the new
approach to regression modeling for forecasting purposes
In short time intervals compared to the total duration of instrumental observations
it is easier to provide the monotony of observations series for variables of
regression models and the homogeneity of samples of data and the independence of
endogenous variables from the less significant factors [2, 4, 6]. It should be noted
~ 54 ~
Mathematical modeling in economy, №3-4, 2017
the main idea of regression analysis is that a suitable regression may take place if a
dependent variable y depends not only on variables ,...,1x mj xx ,..., , and the
variable y may depend on uncontrolled, unknown factors which form something
like a forecast background [7].
It can also be assumed, if in different periods of time these forecast
backgrounds are relatively homogeneous and the corresponding series of dynamics
of independent variables are monotonic (Fig. 2a), adequate regression models (Fig.
2b) may be constructed. Henceforth, we will call these suitable models as
situational models. The situational models can be relatively simple. These can be
single-factor models [7], which show how an endogenous variable y depends on
one the most important exogenous variable x .
15.05.2000
25.04.2000
17.02.2000 03.04.2000
14.04.2000
01.02.2000
а)
115,00
120,00
125,00
25.01.2000 24.02.2000 25.03.2000 24.04.2000 24.05.2000
Time, datesU
W
L,
m
b)
y 4 = 4E-63x 31
R 2 = 0,7931
y 5 = 3E-17x 8,8944
R 2 = 0,972
y 6 = 0,0018x 2,2926
R 2 = 0,1462
y 3 = 5E-58x 28,496
R 2 = 0,9166
20,00
40,00
60,00
80,00
100,00
120,00
115,00 116,00 117,00 118,00 119,00 120,00 121,00 122,00 123,00
UWL х , m
Se
ep
ag
e
di
sc
ha
rg
e
y,
l/
s
17.02.00-03.04.00 01.02.00-17.02.00 03.04.00-14.04.00 15.04.00-15.05.00
4 5 6 3
Fig. 2 – Data of observations of upstream water level (UWL) (a)
and situational regression models for seepage discharge values through
the dam foundation depending on UWL (b)
We suppose that upstream water level (UWL) for dams could be the principal
independent variable x for situational modeling. This is the most suitable and
convenient independent variable and the only independent variable which can be
controlled if it is necessary.
We should emphasized that the main thing there is that situational models
must adapt to particular situations (forecast backgrounds, etc.) that take place
within limited time periods. It is very important they were the most adequate
models to these situations.
In fact, the corresponding situational models, which are based on limited data,
reflect different phase states of the dam as a dynamical system on respective time
intervals. In the simulation we can get sets of adequate situational models that
appropriately evolve over time (Fig. 2b). Although the transitions between the
~ 55 ~
Mathematical modeling in economy, №3-4, 2017
nearest situational models which define adjacent phase states of the dam as a
dynamic system can be non-monotonic [4, 7], the prediction of future condition of
the dam can be based on monitoring the evolutions of these situational models.
This is the main idea of such simulation to obtain situational models.
3. Inductive models and forecasting future states of dams
Inductive diagnostic models are models obtained on base of generalization of
results of construction of situational diagnostic models (Fig. 3a). In the most
general case, inductive models are models of “levels” (Fig. 3b). These models,
which are constructed with using of results of situational modeling at separate time
intervals, can spread on the entire period of observations.
a)y = UWL - PWL
y 1 = 0,9942x - 95,305
R 2 = 0,9773
y 2 = 0,9219x - 87,878
R 2 = 0,9914
y 3 = 1,1284x - 108,97
R 2 = 0,9784y 4 = 0,7836x - 73,627
R 2 = 0,8651
y 5 = 0,9451x - 90,115
R 2 = 0,9108
y 6 = 0,8318x - 78,515
R 2 = 0,9922
6,10
6,20
6,30
6,40
6,50
6,60
6,70
101,80 101,90 102,00 102,10 102,20 102,30 102,40 102,50 102,60
UWL x , m
H
ea
d
de
cr
ea
se
у
,
m
16.11.02-24.11.02 29.11.02-10.12.02 11.12.02-27.12.02
18.01.01-29.01.03 30.01.03-11.02.03 12.02.03-23.03.03
1 2 3
4 5 6
b)
y m ,1 = -1E-05t 2 + 0,8743t - 16482
R 2 = 0,9611
y m ,2 = -1E-05t 2 + 0,9679t - 18241
R 2 = 0,9861
6,30
6,40
6,50
6,60
30.11.2002 30.12.2002 29.01.2003 28.02.2003 30.03.2003
Time t , dates
y m
, m
UWL = 102,2 m UWL= 102,3 m 1 - UWL = 102,2 m 2 - UWL = 102,3 m
Fig. 3 – An example of a set of corresponding situational models (a)
and two inductive models of “levels” (b) for dependencies of head decrease
values on upstream water level (UWL) at a site of seepage between
upstream and a piezometer
The structure of an inductive model is determined by particularities of
behaviour of time series of simulated data obtained from corresponding situational
models. In general, results of situational modeling may represent non-stationary
(Fig. 3b) or stationary (quasi-steady) (See below Fig. 4a) time series of modeling
data with presence or absence of trends respectively.
If trends have high coefficients of determination, inductive models can be
described by these trends (Fig. 3b). Then general inductive models will consist of
corresponding functions which show trends and random “balances” after the
~ 56 ~
Mathematical modeling in economy, №3-4, 2017
extraction of these trends. If results of situational modeling give stationary (quasi-
steady) time series (there are no trends) (Fig. 4a), inductive models can be
represented as regressions (Fig. 4b). In these cases general inductive models will
consist of corresponding regressions and random “balances”. If trends in dynamic
series of results of situational modeling of variable y have small coefficients of
determination, an inductive model of i -level for y can be presented as
composition of a selected trend ( ))(tyT i (Fig. 5a) and a regression R for
“balances” ( ))(tyTyy iii -=D which are random values of the dependent variable
(Fig. 5b):
( ) ( ))ˆ()()( Liii xyRtyTty D+= . (2)
If it is necessary, we can use a new explanatory variable Lx̂ for modeling the
regression of “balances” iyD . We may take into account a transport lag between
the “balances” and the variable Lx̂ too. In more complex cases if we need to take
into account autocorrelation of the “balances”, in addition some cyclical
components or deterministic components of corresponding series can be considered
in inductive models.
a)
y m = 487,94e-4E-05t
R 2 = 0,0259
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
06.12.1999 05.11.2001 06.10.2003 05.09.2005 06.08.2007 06.07.2009
Time t , dates
Se
ep
ag
e
di
sc
ha
rg
e
у m
, l
/s
b)
x m = (x 1 + x n )/2
y m = -3,2907x m
2 + 797,94x m - 48265
R 2 = 0,9002
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
116,00 117,00 118,00 119,00 120,00 121,00 122,00
UWL х m , m
Se
ep
ag
e
di
sc
ha
rg
e
у m
, l
/s
Fig. 4 – Stationary series of dynamics (a) of results situational modeling
of seepage discharge values in conditions of stationary oscillations of
upstream water level (UWL) and a corresponding inductive model
in form of the regression (b)
~ 57 ~
Mathematical modeling in economy, №3-4, 2017
Forecasting future states and behavior of dams is based on the method of
extrapolation and is carried out in two main forms:
1) Real-time forecasts;
2) Long-term forecasts.
Real-time forecasts are made for the purpose of rapid assessment dams’
condition in changed situations (See below Fig. 6) and performed by means of new
situational models which require adjustments to previous situational models due to
new data with extrapolation into region of expected values of independent
variables. If new data comes, real-time forecasts may be constantly corrected.
Observed values which differ significantly than situational models show can
indicate changes of forecast backgrounds (Fig. 6).
y = UWL - PWL
a)
y m ,1 = 16,731e-2,47E-05t
R 2 = 0,3372
y m ,2 = 16,238e-2,32E-05t
R 2 = 0,3739
y m ,3 = 15,799e-2,18E-05t
R 2 = 0,378
6,00
6,50
7,00
10.10.2006 06.08.2007 01.06.2008 28.03.2009 22.01.2010
Time t , dates
y m
, m
1) UWL = 102,2 m 2) UWL = 102,4 m 3) UWL = 102,6 m 1 2 3
x m ,L = (x 1,L +
x n ,L )/2
b)
y 1 = -0,25082x m ,L
2 + 51,288x m ,L - 2621,8
R 2 = 0,5118
y 2 = -0,169325x m ,L
2 + 34,564x m ,L - 1763,8
R 2 = 0,59 y 3 = -0,0878185x m ,L
2 + 17,836x m ,L - 905,51
R 2 = 0,5982
-0,30
-0,20
-0,10
0,00
0,10
0,20
0,30
102,00 102,20 102,40 102,60 102,80 103,00 103,20
x m ,L , my
- b
al
an
ce
s o
fy
m
, m
1) UWL = 102,2 m 2) UWL = 102,4 m 3) UWL = 102,6 m 1 2 3
Fig. 5 – An example of non-stationary series of dynamics of results
of situational modeling of head decrease values with relatively small
coefficients of determination of trends
Long-term forecasts (Fig. 7) are usually based on inductive models which
include trends and regressions for their balances (Fig. 5) but simple regression
models (Fig. 4) may be used too. The expected results of the long-term forecasting
are determination a new situational diagnostic model which corresponds to
expected situation in the nearest future period (Fig. 7a) or series (variety of
options) of situational diagnostic models which can correspond to several possible
situations in the future period (Fig. 7b).
~ 58 ~
Mathematical modeling in economy, №3-4, 2017
The accuracy of long-term forecasts which are made on the basis of inductive
models can be greatly improved if the inductive models are based on results of
situational modeling of past periods which are presented by homogeneous and
interrelated clusters of the relevant data with taking into account behavior of
independent variables and transport lags.
First, we should pay attention to behavior of upstream water level (UWL) (See
below Fig. 8). In particularly, some following typical modes of behaviour of
upstream water level (UWL) affecting dams can be allocated:
1) Slow rise of UWL;
2) Rapid rise of UWL;
3) Stationary fluctuations of UWL;
4) Slow lowering of UWL;
5) Rapid lowering of UWL.
If some transport lags exist between corresponding data of previous and next
periods, forecasts can be unambiguous (Fig. 7a). Otherwise, we get several
possible long-term forecasts concerning future periods (Fig. 7b).
a)
25.03.2004
26.03.2004 -
Change of
forecast
background
y = UWL - PWL
28.03.04
27.03.04
26.03.2004
Situational model of past period:
y 20 = 0,9278x - 88,364
R 2 = 0,9378
Real-time forecast:
y 21.0 = 1,0119x - 96,79
R 2 = 0,9963
6,30
6,40
6,50
6,60
6,70
6,80
101,90 102,00 102,10 102,20 102,30 102,40 102,50
UWL x , m
H
ea
d
de
cr
ea
se
y,
m
17.02.04-25.03.04 26.03.04-28.03.04 20 21.0
b)
11.03.01-
Change of
forecast
background
Situational model
of past period:
y 11 = 6E-124x 60,367
R 2 = 0,8223
Real-time forecast:
y 12.0 = 4E-34x 16,97
R 2 = 0,7822
A final new model:
y 12 = 1E-40x 20,17
R 2 = 0,9527
0,00
20,00
40,00
60,00
80,00
100,00
120,00
140,00
114,00 116,00 118,00 120,00 122,00
UWL х , мSe
ep
ag
e
di
sc
ha
rg
e
y,
l/
s
15.02.01-05.03.01 11.03.01-22.03.01 11.03.01-20.04.01
11 12.0 12
Fig. 6 – Examples of real-time forecasts: a) how head decrease values at
a site of seepage between upstream and a piezometer can depend on
upstream water level (UWL); b) how seepage discharge values can depend
on upstream water level (UWL)
A general diagnostic model of an appropriate diagnostic parameter of a dam
can be presented as a family of situational diagnostic models which are adapted to
~ 59 ~
Mathematical modeling in economy, №3-4, 2017
individual time intervals or as a family of inductive diagnostic models which allow
producing situational diagnostic models for periods in the future. So, forecasts are
performed on the basis of situational diagnostic models and by means of
monitoring for evolutions of these models in time.
It should be noticed the simple mathematical models (trends and regressions)
may be used as situational and inductive diagnostic models for dams where
automated monitoring systems are installed. Such models and combinations thereof
are the best to adapt to new data and changes of forecast backgrounds. When
choosing the diagnostic models of a dam, it is also allowed using of modified
diagnostic parameters and different data processing procedures which are aimed at
the construction of adequate situational diagnostic models for forecasting of dams
condition to assess their reliability according to the data of instrumental
observations.
a)
Field data
y 83 = 0,8787x - 83,583
R 2 = 0,993
Forecast for the next period
y 83(P) = 0,9256x - 88,376
Model of previous period data
y 82 = 0,8629x - 81,996
R 2 = 0,966
6,00
6,10
6,20
6,30
6,40
6,50
6,60
6,70
102,00 102,10 102,20 102,30 102,40 102,50 102,60 102,70
UWL x , mH
ea
d
de
cr
ea
se
y,
m
19.11.09-14.01.10 10.09.09-18.11.09 83 Forecast 83 82
x m,L b) Field data
01.12.11-15.12.11
y = 1,0459x - 100,74
R 2 = 0,9572
x m,L = 102,625 m
6,00
6,20
6,40
6,60
6,80
7,00
102,00 102,20 102,40 102,60 102,80
UWL x , mH
ea
d
de
cr
ea
se
y,
m 102,4 m
102,5 m
102,625 m
102,8 m
Fig. 7 – Examples of long-term forecasts how head decrease values at a site
of seepage between upstream and a piezometer depend on upstream water
level (UWL): a) an unambiguous forecast; b) some forecast options
~ 60 ~
Mathematical modeling in economy, №3-4, 2017
Slow rise of
UWL
Rapid rise of
UWL
Rapid rise of
UWL
Stationary
fluctuations of
UWL
20,00
40,00
60,00
80,00
100,00
120,00
115,00 116,00 117,00 118,00 119,00 120,00 121,00 122,00 123,00
UWL х , mSe
ep
ag
e
di
sc
ha
rg
e
y,
l/
s
17.02.00-03.04.00 01.02.00-17.02.00 03.04.00-14.04.00
15.04.00-15.05.00 4 5
6 3 Forecast (6)
Forecast (3) Forecast (5) Forecast (4)
Fig. 8 – Examples of long-term retrospective forecasts how seepage
discharge values depend on upstream water level (UWL)
Conclusions
1. A new approach to forecast condition and behavior of dams according to data of
instrumental observations with regard to capabilities of automated monitoring
systems installed on hydraulic structures has been proposed. The approach is based
on the use of situational and inductive models of regression type where situational
models correspond to selective series of dynamics of observed data within limited
time intervals and inductive models which are constructed with model data derived
from situational regression models enable to take into account evolutions of
diagnostic parameters in time.
2. The simple mathematical models (trends and regressions) may be used as
diagnostic models of dams if these models are easy adapted to new data and
changes of corresponding forecast backgrounds. The accuracy of long-term
forecasts which are made on the basis of inductive models can be greatly improved
if the inductive models are based on results of situational modeling of past periods
which are presented by homogeneous and interrelated clusters of the relevant data
with taking into account behaviour of independent variables and transport lags.
REFERENCES
1. Dam safety – Guidelines. ICOLD, Bulletin No. 59, 1989.
2. Kuhn M. Applied Predictive Modeling / M. Kuhn, K. Johnson. – New York: Springer
Science+Business Media, 2013. – 600 p.
3. Collacott R.A. Structural Integrity Monitoring / R.A. Collacott. London – New York;
Chapman and Hall, 1985. 455 p.
4. Stefanyshyn D.V. Prediction of indexes of dynamic system with use of observational
data as time series / D.V. Stefanyshyn // Problems of decision making under uncertainties
(PDMU-2013). Abstracts of XXII International Conference. September 23-27. Foros-Yalta,
Ukraine. Kyiv, 2013. – P. 31-32.
5. In-Soo Jung. Interpreting the Dynamics of Embankment Dams through a Time-Series
Analysis of Piezometer Data Using a Non-Parametric Spectral Estimation Method / In-Soo
~ 61 ~
Mathematical modeling in economy, №3-4, 2017
Jung, M. Berges, J.H. Garrett Jr, Ch.J. Kelly // Computing in Civil Engineering, 2013. –
P.P. 25-32.
6. Hamilton J.D. Time series analysis / J.D. Hamilton. Princeton University Pr. Princeton. –
N. J.: 1994. – 782 p.
7. Stefanyshyn D.V. A Method of Forecasting of Indices of Dynamic System that evolves
slowly based on Time Series Analysis / D.V. Stefanyshyn // ICIM 2013. Proc. of 4th Int.
Conf. on Inductive Modelling; Kyiv, Ukraine, September 16-20, 2013. – pp. 221-224.
Been received for revision 26.05.2017.
|
| id | nasplib_isofts_kiev_ua-123456789-131920 |
| institution | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| issn | 2409-8876 |
| language | English |
| last_indexed | 2025-12-07T16:19:40Z |
| publishDate | 2017 |
| publisher | Інститут телекомунікацій і глобального інформаційного простору НАН України |
| record_format | dspace |
| spelling | Stefanyshyn, D.V. 2018-04-06T08:40:54Z 2018-04-06T08:40:54Z 2017 Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems / D.V. Stefanyshyn // Математичне моделювання в економіці. — 2017. — № 3-4(9). — С. 50-61. — Бібліогр.: 7 назв. — англ. 2409-8876 https://nasplib.isofts.kiev.ua/handle/123456789/131920 504.75 + 004.942; 519.25; 621.791: 626/627 An approach to forecasting the behaviour of dams according to the data of instrumental observations with regard to capabilities of automated monitoring systems has been proposed. The approach is based on the use of situational and inductive models, where the situational models correspond to selective series of dynamics of observed data within limited time intervals and the inductive models which are constructed on model data derived from situational models simulate evolutions of diagnostic parameters. Запропоновано підхід до прогнозування поведінки гребель за даними інструментальних спостережень з врахуванням можливостей автоматизованих систем моніторингу. Підхід ґрунтується на використанні ситуаційних та індуктивних моделей, де ситуаційні моделі відповідають вибірковим рядам динаміки спостережених даних на обмежених інтервалах часу, а індуктивні моделі, що будуються за модельними даними, які визначаються з ситуаційних регресійних моделей, відображають еволюції діагностичних параметрів. Предложен подход к прогнозированию поведения плотин по данным инструментальных наблюдений с учетом возможностей автоматизированных систем мониторинга. Подход основан на использовании ситуационных и индуктивных моделей, где ситуационные модели соответствуют выборочным рядам динамики данных наблюдений на ограниченных интервалах времени, а индуктивные модели, строящиеся по модельным данным, которые определяются при помощи ситуационных регрессионных моделей, отображают эволюции диагностических параметров. en Інститут телекомунікацій і глобального інформаційного простору НАН України Математичне моделювання в економіці Analysis, evaluation and forecasting in economy Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems Удосконалення діагностичних моделей для прогнозування поведінки гребель, обладнаних автоматизованими системами моніторингу Совершенствование диагностических моделей для прогнозирования поведения плотин, оборудованных автоматизированными системами мониторинга Article published earlier |
| spellingShingle | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems Stefanyshyn, D.V. Analysis, evaluation and forecasting in economy |
| title | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| title_alt | Удосконалення діагностичних моделей для прогнозування поведінки гребель, обладнаних автоматизованими системами моніторингу Совершенствование диагностических моделей для прогнозирования поведения плотин, оборудованных автоматизированными системами мониторинга |
| title_full | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| title_fullStr | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| title_full_unstemmed | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| title_short | Improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| title_sort | improving diagnostic models for forecasting the behavior of dams equipped with automated monitoring systems |
| topic | Analysis, evaluation and forecasting in economy |
| topic_facet | Analysis, evaluation and forecasting in economy |
| url | https://nasplib.isofts.kiev.ua/handle/123456789/131920 |
| work_keys_str_mv | AT stefanyshyndv improvingdiagnosticmodelsforforecastingthebehaviorofdamsequippedwithautomatedmonitoringsystems AT stefanyshyndv udoskonalennâdíagnostičnihmodeleidlâprognozuvannâpovedínkigrebelʹobladnanihavtomatizovanimisistemamimonítoringu AT stefanyshyndv soveršenstvovaniediagnostičeskihmodeleidlâprognozirovaniâpovedeniâplotinoborudovannyhavtomatizirovannymisistemamimonitoringa |