See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/320559743

Real Time Operation Model for Optimum Operation of Bukan Reservoir in

Lake Urmia Basin

Conference Paper · October 2017

CITATIONS READS

0 228

2 authors:

Keyhan Gavahi Jamshid Mousavi

University of Alabama University of Waterloo
25 PUBLICATIONS 472 CITATIONS 131 PUBLICATIONS 2,616 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Keyhan Gavahi on 03 October 2020.

The user has requested enhancement of the downloaded file.

Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

Real Time Operation Model for Optimum Operation of Bukan

Reservoir in Lake Urmia Basin

Kayhan Gavahi1, S. Jamshid Mousavi2

1- MSc. Student, School of Civil and Environmental Engineering, Amirkabir University of Technology
2- Professor, School of Civil and Environmental Engineering, Amirkabir University of Technology

Email: [email protected]

Abstract
This paper presents an adaptive forecast-based mathematical model for real-time operation of Bukan
reservoir located in Lake Urmia (LU) Basin. The model consists of three modules: 1) a forecasting module
that predicts future inflows to the reservoir up to the end of a year, 2) a reservoir operation optimization
module determining optimum reservoir releases and 3) an updating module that updates inflow forecasts and
optimal releases at the beginning of every time step. As the forecasting module, an adaptive neural-based
fuzzy inference system (ANFIS) is trained and used to forecast future inflows. Having these future inflows,
a linear optimization model is formulated and solved to find optimal reservoir releases in future time steps
of the year. The objective function of the model is to maximize total volume of water received by the lake
while supplying water for irrigation demands and instream flow. The model is tested for 2015-2016 water
year and its performance is assessed. Results obtained are promising in terms of volumes of water allocated
to different demand sectors and preventing downstream areas from flooding caused by uncontrolled spillage.
This is important as the system has experienced a severe flood in this particular year resulting in significant
damages.
Keywords: Lake Urmia, Bukan Reservoir, Real Time Operation, Optimization.

1. INTRODUCTION
Climate change and human related activities has severely affected total inflow received by the Lake
Urmia (LU) in the Northwest of Iran. Construction of various dams on the main rivers flowing to the lake and
domestic and agricultural developments, which has resulted in increasing water demands in the past decade, has
decreased the water level of the lake by 90% in September 2015. Climate change and increased irrigation demands
are two main reasons for reduction in inflows to the reservoirs and much less water received by the lake. In this
critical situation, the issue of improved operation of reservoiris to better meet the LU share from upstream
reservoir releases has become more significant. Most of these reservoirs are multipurpose reservoirs functioning
for water supply for downstream consumptive and non-consumptive demands and flood control. A number of
contradictory objectives exist for operation of reservoirs of the basin. The first one is to store water as much as
possible at the beginning of May to meet agricultural demands of future months up to about middle of September
(end of water year) and the other one requires water releases in order to reserve reservoir capacity for the incoming
flood. On the other hand, most of water releases made during irrigation season (May-September) will not reach
the LU, resulting in critical adverse conditions in the lake. In other words, retaining water behind dams in winter
months, to achieve higher reliability in meeting irrigation water demands, also causes significant reduction in the
annual volume of water received by the LU that ultimately culminate in water level decline of the lake being
drying up. Therefore, more careful and planned reservoir release schedule is needed accounting for the
aforementioned conflicting objectives. In this line, a real-time reservoir operation modeling approach benefiting
from forecasts of future inflows to reservoirs could be of help on how to best operate the basin's reservoirs [1].
The framework can provide insight on how to make best possible release decisions based on the most recent
knowledge of future inflows. Having the most recent state of the system, forecasting models can determine the
future state with a reasonable degree of uncertainty. Although the inherent forecast uncertainty will directly affect
the results of this kind of operational models, improvements is expected to attain because of step-by-step reduction
of forecast uncertainty as more information and data are received and the release decisions are updated [2], [3].
Two types of real-time reservoir operation models has been used in the literature [4]. Standard real-time
operation models in which a streamflow forecasting model estimates future inflows only one time up to a specified
short-run time horizon, and the decisions will be made based on these estimated predictions. This type of real-
time operation models is mostly used for flood management purposes [5]–[8]. The second type is adaptive real-
time operation models where the inflow-to-reservoir forecasts are updated step-by-step at the beginning of each

570
Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

time period of operation. It therefore has the chance to adapt itself to the most recent conditions and changes
occurred in the system.
In this paper an adaptive real-time operation model is proposed for Bukan reservoir and its downstream
system, which is the largest reservoir of the LU water basin built on the largest river of the basin, Zarinehrood,
which provides nearly 40% of the total inflow to the lake. It is a multipurpose reservoir with the main objectives
of water supply for agricultural, domestic and industrial uses, and controlling floods that occur mostly in spring
season. Thus, a release schedule specifically accounting for the LU share is needed as a necessary requirement for
the success of restoration plan of the lake. Therefore, the proposed modeling framework considers not only water
supply to various demand sectors, it but also takes into account the LU water demand and share, especially in
months prior to the crop season. The paper is organizing as follows: the proposed methodology is presented in
section 2 with three subsections explaining inflow forecasting module, reservoir operation module and updating
module, respectively. Section 3 introduces the case study and the data and information used followed by section
4 containing results obtained and discussing them. Finally, section 5 ends the paper by a summary and concluding
remarks.

2. METHODS AND MODELING TOOLS

We explained that the real-time operation model consists of three modules. The forecasting module
predicts the inflow to the reservoir up to a specified time horizon, T. Various types of forecasting methods
including physics-based, time-series modeling and data-driven methods can be used in order to forecast future
inflows to the reservoir. Having the future inflows, the reservoir operation module determines optimum releases
in future months based on the water demands and various constraints of the system. Simulation, optimization or
a combination of both could be used for this task. These two modules working together provide an efficient tool
for making decision on how to operate the system in near future based on the most recent knowledge available.
However, the longer the forecasting horizon of the prediction module, more uncertain future inflow forecasts will
be. Therefore, the reliability of decisions made on future releases diminishes rapidly by heading toward the final
time steps of the operation horizon (end of water year). This is where the last module plays an integral part by
updating the model predictions at the beginning of each time step based on the most recent information received
as the inputs for the forecasting module. The updating module also update the initial state of the system based on
observed inflows and the latest changes in downstream demands and conditions.
Figure 1. illustrates how these three modules work together. We present the details of each module in the
following subsections.

2.1. FORECASTING MODULE

An adaptive neural-based fuzzy inference system (ANFIS) model was developed and used as the
forecasting module. ANFIS uses a combination of artificial neural network (ANN) and fuzzy logic approaches. It
uses fuzzy inference system of Sugeno type to obtain final values of each decision variable [9], [10]. Using a
hybrid learning mechanism that consists of the back-propagation gradient descent and a least squares methods,
ANFIS determines the best possible set of parameters to map the given input-output set of data points. Thus model
is a data-driven model that identifies the inherent function within a system without explicitly knowing the physical
relationship between input and output variables.

571
Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

Figure 1. Flow diagram of the adaptive real-time reservoir operation model

Assuming a two dimensional input vector [x, y], the equivalent ANFIS structure that derives the output
function f will be a five-layer feed-forward network shown in Figure 2. More detailed presentation of ANFIS for
forecasting hydrological time-series can be found in the literature [11]–[15].

Figure 2. ANFIS structure for a two dimensional input vector [x, y] and output function f

2.2. RESERVOIR OPERATION MODULE

Figure 3. shows a schematic view of the Zarinerood River system including Bukan Dam. Based on the
sum of water demands for each sector that are directly supplied by the Bukan reservoir, we have developed a
linear mathematical model optimizing the system operations.

572
Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

Figure 3. A schematic view of water inflows and withdrawals to the Zarinerood River
downstream of Bukan Dam
The objective function of the proposed model is stated by equation 1. It is maximization of the benefits
gained by releasing water for environmental instream flow, agricultural uses and the LU water demand. Each of
these variables are accompanied by a priority coefficient. A larger coefficient for each user means a higher priority
for water allocation to that user during shortages.

Z  t 1 Co Rot  Ca Rat  CL RLt

n
Max (1)
In above equation n is the total number of time steps, C o is the priority coefficient for environmental
instream flow of Zarineroud River, C a refers to priority coefficient for agricultural demands, and C L is the
priority coefficient for water received by the Lake Urmia. Rot , Rat and RLt are allocations for Zarineroud
River, irrigation demands and the Lake Urmia, respectively. This objective function ensures that water is allocated
first to minimum instream flow and minimum obligatory irrigation demands decreased by 40% compared to
business-as- usual irrigation demands. After that the next priority is for the LU. Constraints of the mathematical
program consist of water balance equations, upper bounds on water allocation values that must be less than the
required demands and physical constraints regarding capacity of the reservoir and the downstream channels,
especially at the inlet of Lake Urmia where fuse plugs are installed. They facilitate the release made for the lake
reach the water body of the lake and prevent it from losing through seepage and evaporation in the buffer zone
adjacent to the lake. Below is the set of constrains represented by equations 2 to 7:
S1  Sn (2)
St 1  St  Qt  Rat  Rdt  Rit  Rot  RLt  Et  spilt (3)
Rat  Dat , Rd t  Dd t , Rit  Dit , Rot  Dot (4)
RLt  Rot  Cap fuze (5)
S min  S t  S max (6)
Rat Rat 1
 (7)
Dat Dat 1
where S t and Qt are the beginning-of-month reservoir storage and inflow to reservoir in month t ,
respectively. Et and spil t are evaporation and spillage from the reservoir, respectively. Da , Dd , Di t t t

and Dot are agricultural, domestic, industrial and minimum instream flow requirements, respectively. S m ax
and S min are respectively upper and lower bounds on the reservoir storage volume, and Cap fuze is the

573
Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

capacity of the structures built at the inlet of the lake. Equation 7 ensures that if there exist any shortages and the
annual irrigation demand cannot be fully met, shortages are distributed proportionately among irrigation months
in an irrigation season [16]. Solution of this mathematical program contains optimum releases and allocation to
different users.

2.3. UPDATING MODULE

According to figure 1, at the beginning of each time step, updating module will provide the ANFIS model
with the most up-to-date observed data. These data include inflow to the reservoir, changes in downstream
demands and also changes in downstream conditions that may result in changing the reservoir operation module
coefficients. It also enables the ANFIS model to be trained once again in order to adapt itself and its parameters
to observed inflows of previous time steps and therefore obtain higher performance in future inflow predictions.
After forecasting future inflows of the next time steps up to the end of the operation horizon, T, the presented
linear reservoir operation optimization model determines optimum releases and water allocations to different
demands up to the end of the operation horizon. Since the forecasted inflows and therefore releases for more
distant time steps are more uncertain and less reliable, only the reservoir release and water allocation for the
immediate next time step ahead will be implemented. Having the observed inflow, the updating module simulates
the system with actual values of inflow to the reservoir in order to determine the initial state of the system for the
next time step.
Then if the model has not reached the final time step, the updating module collects the observed inflows
and determines the initial state of the system, and forecasting and reservoir operation models are employed again.
The procedure is repeated until reaching the end of operation time horizon that is the beginning of the next water
year.

3. CASE STUDY
Lake Urmia (LU) water basin is located in northwest part of Iran. The basin takes its name from Urmia
Lake located at the center of water basin. The lake suffers from dramatic decline in its water level and volume of
water stored in it. In September 2015, the lake had only 10% of its potential storage capacity, a situation that
seems to get worse if proper actions are not taken that could culminate in total loss of the lake. The largest river
in the basin is Zarinehrood River. This river starts from Zagros mountain range and drains to Lake Urmia after
flowing for about 350 km distance. It passes through three provinces of Kordestan, West and East Azarbaijan and
contributes to about 41% of total surface water inflow to the lake. The drainage area of Zarinehrood River is about
12,000 km, and it crosses two cities of Shahindej and Miandoab [17]. Bukan reservoir as the largest reservoir in
the basin has been constructed on this river for supplying water to agricultural, domestic and industrial water
users. It has 810 million cubic meters (MCM) storage capacity and the long-term average annual inflow to the
reservoir is about 1600 MCM. The outflows are distributed by the Norozlu diversion dam to about 85,000 hectares
of Miandiab farming lands, and to three major cities such as Tabriz and its suburbs.
There are eight major dams being under operation in the basin discharging water to the lake and other
demand points. The operation policies based on which the dams are operated directly affects the amount of water
received by the lake throughout a year. Thus, it is important to use operation methods to account for downstream
needs as well as considering the water required by the lake in order to survive. It should be noted that temporal
variation of water released for LU is as important as the total quantity of water received. Releasing high amount
of water in a shorter period of time not only can cause damages in downstream urban and agricultural areas, it but
also results in damaging a number of structures and fuse gates built to facilitate transferring water to the main
water body of the lake and prevent it from losing by evaporation.

4. RESULTS AND DISCUSSION

The proposed adaptive real-time model is tested for planning the operation of Bukan reservoir for a
typical year. The operation horizon was considered to be one year with monthly time steps. However, the model
has the ability to be used for any desired time horizon and time steps as long as the two first modules are adopted
with the desired reliability. Using the 34 years available historical inflow time series, we trained and validated an
ANFIS-based inflow forecasting model. Three previous monthly inflows were used as input to the model to
forecast the future inflows to the reservoir. For example, if we are at the beginning of September, the ANFIS
model uses inflows in June, July and August and predicts inflows to the reservoir during September. Having
estimated inflows to the reservoir in September, August and July as inputs to the ANFIS model, the model will
forecast October's inflow and so on. This procedure is continued up to the last month of the planning horizon until

574
Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

inflow values of all future months are predicted. Note that the predicted inflow for September has a certain amount
of error and uncertainty. This error will propagate through inflow forecast of all future months, therefore, forecast
errors are accumulated as we go ahead. In this regard, the updating module helps us significantly reduce the
forecast error and adapt the model to inflow changes by involving actual observed values into the model at the
beginning of each time step.
Table 1. reports water demands for various users to be supplied exclusively by Bukan reservoir.
Agricultural demands are about 58% of total water demand. Noting that the storage capacity of the reservoir is
810 Million Cubic Meters (MCM), volume of withdrawal for irrigation purposes is relatively considerable. Also
95% of annual irrigation demand belongs to the main crop season, i.e. May to September. Interestingly, only less
than 15% of total annual inflow occurs in this period; therefore, the role of the reservoir storage to regulate natural
inflows would be essential. Because of high water demands and farmers tendencies to use water in the crop season,
releases made to reach LU ( R Lt ) are more likely to reach the lake during time periods other than the crop season.

Table 1- Demands for various sectors downstream of Bukan Dam (all values are in
MCM).
Oct Nov Dec Jan Feb Mar Apr May June Jul Aug Sep sum
Agriculture
1.9 18.3 0 0 0 0 0 51.2 90.9 81.6 99.5 72.8 416.2
Demand (Da)
Domestic Demand
11.5 10.2 9.8 9.8 10.3 10.5 10.5 14.1 12.5 13.5 14 15.1 141.7
(Dd)
Industrial Demand
0.5 0.5 0.5 0.5 0.5 0.5 0 0 0 0 0 0 3
(Di)
Environmental
18.1 10 9 5 13.5 13.5 20 13.5 13.5 13.5 13.5 13.5 156.6
Demand (Do)

According to high priority of meeting domestic and industrial water demands, the reservoir releases for
this type of demands are incorporated as rigid constraints of the reservoir operation module in the model (equation
4), so they must be met regardless of yearly variations of inflow to the reservoir.
The proposed model is tested for year 2015-2016 with monthly inflows shown in figure 4. It can be seen
that inflow to reservoir in April is 517 MCM, which is 64% of total capacity of Bukan reservoir. This huge amount
of input resulted in uncontrolled spillage and heavy damages downstream. The problem was mostly because of
not having a suitable tool to forecast this event in advance and operate the reservoir smartly in such a critical
situation. In this case, inflows in months prior to the crop season was stored and was not released on time to
provide with enough storing capacity for incoming floods in Spring. It is of interest to test the proposed real time
model and see if it can improve the situation in a real life operation context.
Table 2. presents the results obtained by the adaptive real-time operation model proposed. We can see
that water demands for agricultural, domestic, industrial and environmental are completely met. Also the sum of
spilled water is zero showing that it has successfully prevented unwanted spillage by releasing water for LU in
January, February and March with the amounts of 156.4, 171.5 and 171.5 MCM, respectively. Therefore, the
resulting end of month (EoM) storage of March is determined to be 422.5 MCM, which is nearly half of the
capacity of Bukan reservoir. This allowed having enough space left for the incoming 517 MCM inflow in April
and managing the flood occurred in this year with a volume of 300 MCM. This has been made possible mainly
by the updating capacity of the proposed framework.

575
 Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

Table 2- Demands for various sectors downstream of Bukan Dam (all values are in
MCM).
Oct Nov Dec Jan Feb Mar Apr May June Jul Aug Sep sum
EoM storage 245.2 279.4 345.1 302 329 422.5 740.7 785.7 671.8 551.9 417.8 308.3
Agricultural
1.9 18.3 0 0 0 0 0 51.2 90.9 81.6 99.5 72.8 416.2
Release
Domestic
11.5 10.2 9.8 9.8 10.3 10.5 10.5 14.1 12.5 13.5 14 15.1 141.7
Release
Industrial
0.5 0.5 0.5 0.5 0.5 0.5 0 0 0 0 0 0 3
Release
Environmental
18.1 10 9 5 13.5 13.5 20 13.5 13.5 13.5 13.5 13.5 156.6
Release
LU Release 0 0 0 156.4 171.5 171.5 165 76.2 15.7 0 0 0 756.2
Spillage 0 0 0 0 0 0 0 0 0 0 0 0 0
Env+LU Release 18.1 10 9 161.4 185 185 185 89.7 29.2 13.5 13.5 13.5 912.8

Figure 4. Monthly inflows to Bukan reservoir for the tested water year (2015-2016).

5. CONCLUSIONS
This paper presented an adaptive real-time operation framework proposed for forecast-based optimal
operation of Bukan reservoir in Lake Urmia. It consists of a streamflow forecast module, a medium-term reservoir
operation optimization module and an updating procedure to adjust the forecasted inflows and the reservoir
releases as time moves forward and the planning horizon becomes shorter. Testing the proposed methodology, an
adaptive network-based fuzzy inference system (ANFIS) developed for inflow forecasting coupled with a linear
operation optimization model performed well enough for the Bukan system operation in year 2015-2016 in terms
of supplying water for agricultural, domestic, industrial and environmental uses as well as controlling a severe
flood experienced in this year. It provided LU with 756.2 MCM of water released in months prior to the crop
season, which is the desired time period to make releases for LU.

576
 Long-Term Behaviour and Environmentally Friendly Rehabilitation Technologies of Dams (LTBD 2017) DOI:10.3217/978-3-85125-564-5-076

6. REFERENCES
1. Nohara, D., & Hori, T. (2017). Integrated Reservoir Operation Considering Real-Time Hydrological
Prediction for Adaptive Water Resources Management. In Sustainable Water Resources Planning and
Management Under Climate Change (pp. 101-132). Springer Singapore.
2. Chen, L., Singh, V. P., Lu, W., Zhang, J., Zhou, J., & Guo, S. (2016). Streamflow forecast uncertainty
evolution and its effect on real-time reservoir operation. Journal of Hydrology, 540, 712-726.
3. Gavahi, K., Mousavi, S. J., Ponnambalam, K. (2019). Adaptive forecast-based real-time optimal reservoir
operations: Application to Lake Urmia. Journal of Hydroinformatics, 21(5), 908–924.
4. Nohara, D., Nishioka, Y., Hori, T., & Sato, Y. (2016). Real-Time Reservoir Operation for Flood Management
Considering Ensemble Streamflow Prediction and Its Uncertainty. In Advances in Hydroinformatics (pp. 333-
347). Springer Singapore.
5. Mujumdar, P. P., & Ramesh, T. S. V. (1997). Real‐time reservoir operation for irrigation. Water Resources
Research, 33(5), 1157-1164.
6. Nikoo, M. R., Gavahi, K., and Khoramshokooh, N. (2019). A Multi-objective Simulation–Optimization Approach
for Design of Cutoff Walls and Apron of Diversion Dams. Iranian Journal of Science and Technology -
Transactions of Civil Engineering, Springer International Publishing, 43(2), 241–252.
7. Unver, O. I., & Mays, L. W. (1990). Model for real-time optimal flood control operation of a reservoir
system. Water resources management, 4(1), 21-46.
8. Wasimi, S. A., & Kitanidis, P. K. (1983). Real‐time forecasting and daily operation of a multireservoir system
during floods by linear quadratic Gaussian control. Water Resources Research, 19(6), 1511-1522.
9. Hsu, N. S., & Wei, C. C. (2007). A multipurpose reservoir real-time operation model for flood control during
typhoon invasion. Journal of Hydrology, 336(3), 282-293.
10. Huang, W. C., & Hsieh, C. L. (2010). Real‐time reservoir flood operation during typhoon attacks. Water
Resources Research, 46(7).
11. Gavahi, K., Mousavi, S. J., and Ponnambalam, K. (2018). Comparison of Two Data-Driven Streamflow
Forecast Approaches in an Adaptive Optimal Reservoir Operation Model. EPiC Series in Engineering, (3), 755–
763
12. Jang, J. S. (1993). ANFIS: adaptive-network-based fuzzy inference system. IEEE transactions on systems,
man, and cybernetics, 23(3), 665-685.
13. Jang, J. S. R., Sun, C. T., & Mizutani, E. (1997). Neuro-fuzzy and soft computing: a computational approach to
learning and machine intelligence.
14. Firat, M., Turan, M. E., & Yurdusev, M. A. (2009). Comparative analysis of fuzzy inference systems for water
consumption time series prediction. Journal of hydrology, 374(3), 235-241.
15. Nikoo, M. R., Karimi, A., Kerachian, R., and Gavahi, K. (2018). Optimum Operation of Reservoirs in the
Karkheh Basin in Iran Considering Impacts of Non-integrated Development and Climate Change. Iranian
Journal of Science and Technology, Transactions of Civil Engineering, 1–11.
16. Keskin, M. E., Taylan, D., & Terzi, O. (2006). Adaptive neural-based fuzzy inference system (ANFIS)
approach for modelling hydrological time series. Hydrological sciences journal, 51(4), 588-598.
17. Zounemat-Kermani, M., & Teshnehlab, M. (2008). Using adaptive neuro-fuzzy inference system for
hydrological time series prediction. Applied Soft Computing, 8(2), 928-936.
18. Nayak, P. C., Sudheer, K. P., Rangan, D. M., & Ramasastri, K. S. (2004). A neuro-fuzzy computing technique
for modeling hydrological time series. Journal of Hydrology, 291(1), 52-66.
19. Ilich, N. (2011). Improving real-time reservoir operation based on combining demand hedging and simple
storage management rules. Journal of hydroinformatics, 13(3), 533-544.
20. Yasi, M., & Ashori, M. (2017). Environmental Flow Contributions from In-Basin Rivers and Dams for Saving
Urmia Lake. Iranian Journal of Science and Technology, Transactions of Civil Engineering, 41(1), 55-64.

View publication stats