Assignment # 8

Article Reference:
T.R Neelakantan, N.V. Pundarikanthan, Neural Network-Based Simulation-Optimization Model for Reservoir Operation, Mar-Apr 2008, Journal of Water Resources Planning and Management

Summary:
The paper focuses on planning a model for reservoir operation which uses simulation-optimization approach. Author takes Chennai water supply as the study area. Reservoir planning was improved by demand management using the hedging rule which distributes the deficits for longer time by rationing water supply. For defining and optimizing the decision variables for hedging rule, the neural network based simulation model was used as a sub model to the Hookes and Jeeves programming model. Simulation analysis requires a operation policy, a standard policy is the optimal policy where the objective is to minimize the total deficit over the time. Hedging and rationing rule helps in distributing the deficit over a longer time. The total storage in the reservoir is divided into different zones and based on the falling level in each zone the release target can be fixed. For the study the reservoir is divided into four zones and the storage levels are S1, S2 and S3. This simulation model is sent as sub model to the Hookes and jeeves nonlinear programming model which passed management decision vectors to the simulation model to find the objective function as an output.

The whole study has been carried out in various stages. In the first stage the back propagation neural network is used to simulate the reservoir operation. Further in the second stage this neural network was fed as sub model to Hookes and Jeeves, this combined simulation-optimization model is used to screen the operation policies. In the third stage the best solution of the optimization is fine tuned using the conventional simulation-optimization model.

Authors have considered Chennai water supply as the study area. Chennai’s water supply is catered through three reservoirs- Poondi, Cholavaram and Red hills and ground water sources. The reservoirs are fed only by the north east monsoon which lasts only for three months. Due to increased extraction of ground water, the water table fell rapidly and there are traces of sea water intrusion. They have planned for the augmentation of the system by Krishna water. With the new excess water they have proposed to use Chembarambakkam reservoir as one of the terminal reservoirs. As Chennai is facing severe drought conditions the criteria would be to reduce the shortfalls as much as possible. During the drought condition the water supply managers prefer smaller shortfalls than dealing with larger ones and therefore they are following the method of hedging and rationing. Distribution of deficits should be done in such a way that they are spread for a longer period and with lesser magnitude. To do the above deficit index (sum of squared deficits) is followed which should be minimum among the other options. With the augmentation of Krishna water supply, equity of reservoir levels for two parallel reservoirs (Red Hills and Chembarambakkam) needs to be maintained. So the demand ratio needs to be 1:x and the deficit ratio to 1:x. As deficit index is the second order parameter the ratio between the two reservoirs needs to be 1:x^2.

The simulation model uses the mass balance principle and follows the constraints on canal capacity, minimum and maximum storage capacity, and evaporation loss and so on. The objective function is to minimize the deficit index by changing the decision management variables S1, S2, and S3 (Storage levels). The study was carried out using different scenarios considering different source and terminal reservoirs, to estimate which combination proves to improve the objective function. Inflow data into the reservoirs are studied to see the performance of the reservoir. Authors have considered catchment inflow, percolation loss, and transmission losses to be negligible. The study was carried out with different supply levels – full demand, 80% demand 75% demand, 67% demand and 50% demand.

Discussion:
This paper was interesting as the study deals with how we can optimize the operating policy based on the hedging rule. As this model is flexible to include various reservoirs and more complex system, I think it is a good model to carry out such a study. It has the flexibility of changing the nonlinear programming model instead of Hookes and Jeeves, which would be interesting to explore.