Optimal reservoir operation policies using novel nested algorithms

Delipetrev, Blagoj and Jonoski, Andreja and Solomatine, Dimitri (2015) Optimal reservoir operation policies using novel nested algorithms. In: European Geosciences Union General Assembly 2015, 12–17 April 2015, Vienna, Austria.

[thumbnail of 1. Blagoj Delipetrev -Optimal reservoir operation policies using novel nested algorithms.pdf]
1. Blagoj Delipetrev -Optimal reservoir operation policies using novel nested algorithms.pdf

Download (1MB) | Preview
[thumbnail of EGU2015-0.pdf]

Download (37kB) | Preview
[thumbnail of egu.pdf]

Download (217kB) | Preview


Historically, the two most widely practiced methods for optimal reservoir operation have been dynamic programming
(DP) and stochastic dynamic programming (SDP). These two methods suffer from the so called “dual
curse” which prevents them to be used in reasonably complex water systems. The first one is the “curse of
dimensionality” that denotes an exponential growth of the computational complexity with the state – decision
space dimension. The second one is the “curse of modelling” that requires an explicit model of each component
of the water systemtoanticipate theeffect ofeach system’s transition.
We address the problem of optimal reservoir operation concerning multiple objectives that are related to 1)
reservoir releases to satisfy several downstream users competing for water with dynamically varying demands, 2)
deviations from the target minimum and maximum reservoir water levels and 3) hydropower production that is a
combination of the reservoir water level and the reservoir releases.
Addressing such a problem with classical methods(DPandSDP)requires a reasonably high level of discretization
of the reservoir storage volume, which in combination with the required releases discretization for meeting
the demands of downstream users leads to computationally expensive formulations and causes the curse of
We present a novel approach, named “nested” that is implemented in DP, SDP and reinforcement learning (RL)
and correspondingly three new algorithms are developed named nested DP (nDP), nested SDP (nSDP) and
nested RL (nRL). The nested algorithms are composed from two algorithms: 1) DP, SDP or RL and 2) nested
optimization algorithm. Depending on the way we formulate the objective function related to deficits in the
allocation problem in the nested optimization, two methods are implemented: 1) Simplex for linear allocation
problems, and 2) quadratic Knapsack method in the case of nonlinear problems. The novel idea is to include the
nested optimization algorithm into the state transition that lowers the starting problem dimension and alleviates
thecurseofdimensionality.The algorithms can solve multi-objective optimization problems,without significantly
increasing the complexity and the computational expenses.The algorithms can handle dense and irregular variable
discretization, andarecoded in Java asprototype applications.
The three algorithms were tested at the multipurpose reservoir Knezevo of the Zletovica hydro-system located in
the Republic of Macedonia, with eight objectives, including urban water supply, agriculture, ensuring ecological
flow, and generation of hydropower. Because the Zletovica hydro-system is relatively complex, the novel
algorithms were pushed to their limits, demonstrating their capabilities and limitations. The nSDP and nRL
derived/learned the optimal reservoir policy using 45 (1951-1995) years historical data. The nSDP and nRL
optimal reservoir policy was tested on 10 (1995-2005) years historical data, and compared with nDP optimal
reservoir operation in the same period. The nested algorithms and optimal reservoir operation results are analysed
and explained.

Item Type: Conference or Workshop Item (Speech)
Subjects: Natural sciences > Computer and information sciences
Natural sciences > Earth and related environmental sciences
Divisions: Faculty of Computer Science
Depositing User: Blagoj Delipetrev
Date Deposited: 30 Sep 2015 12:47
Last Modified: 30 Sep 2015 12:47
URI: https://eprints.ugd.edu.mk/id/eprint/13898

Actions (login required)

View Item View Item