Neelakantan TR, Pundarikanthan NV (2000) “Neural network-based simulation-optimization model for reservoir operation,” Journal of Water Resources Planning and Management, 126(2) pg. 57-64.
Summary
This study was to act as an effort to develop the planning model for a reservoir operation. The reservoir operation model was based on a simulation-optimization approach, which was chosen for time consumption reasons. Simulation modeling was practical for an operation schedule because of the way that the modeling could accurately signify the reservoir’s qualities and characteristics that may be too complex or difficult to model. The model was also used to portion water use in the reservoir for reasons of relieving future drought conditions. The location that the model was tested on was in the Chennai water supply system in India.
The nonlinear programming model that the authors chose to use was the Hooke and Jeeves unconstrained linear programming model. This model included a “neural-network-based simulation sub-model.” This was introduced as a model that will closely mimic brain neurology.
The research took multiple steps to identify the optimal reservoir schedule. Firstly, the network was adjusted to simulate the accurate operation of the reservoir system. Secondly, the neural network model was built and linked as the sub-model, which was used together to “screen the operation policies.” Lastly, the optimization stage of the model was conducted. The operation policy that will yield the better objective function value is chosen from the dual simulation-optimization results. These results will further be filtered through using the “conventional simulation model”, versus the neural network simulation model.
Discussion
This article was interesting to me, because I think reservoir operation policy techniques are intriguing. The results of the model were found to be satisfactory, compared to the conventional simulation-optimization model. The authors said that a certain amount of exemplars is necessary for the network to be trained accurately.
This method was also found to be quite flexible and can easily adjust to complex operations. The authors seemed to have no superiority towards a Hooke and Jeeves nonlinear optimization model. They said that other optimization models can be used in place of Hooke and Jeeves. So what I would like to know is how much the results would be modified by using a different nonlinear optimization model. We all saw in our homework that results can vary a significant amount when comparing 5 distinctive nonlinear optimization models.
Subscribe to:
Post Comments (Atom)
Kate, I think that your question might have been partially answered after listening on Monday to an explaination of each of the different optimization models we used in the most recent homework. The range in results helped to demonstrate some of the shortcomings found with the different models. As was demonstrated in class, the initial guess might regulate if the solution will be the global or local minima of the objective function.
ReplyDeleteWhen looking at a problem like the one presented in this week's article. The difference in results between the different optimization model would be due to the local and global minima present in the system.