Neelakantan TR, Pundarikanthan NV (2000) “Neural network-based simulation-optimization model for reservoir operation,” J. of Water Resource Planning and Management, 126(2) pp. 57-64.
Chennai (formerly known as Madras) is a major city in the Tamil Nadu region of southern India. Although it receives 51 in/yr of rainfall, this mostly comes during the three-month monsoon, so the city is prone to droughts. The city's water is supplied by aa series of reservoirs, which have been typically regulated through standard operating policy. This paper attempts to study the reservoir management through the implementation of a neural network model.
The optimisation approach used was to minimize the overall deficit index, where the ODI is the sum of the squares of the deficits of all the reservoirs, after the deficits have been normalized based on the size of demand from the different reservoirs. Then, the two scenarios ( one with the current system and one with proposed reservoirs included) were optimised to test their performance.
Optimisation was done by a neural network based optimisation, where a neural network is an algorith based on how a brain works. First, the neural network must be trained, and then optimisation is done. The authors concluded that neural-networks can optimizee the workings of large water resource systems.
This paper was interesting in that it uses a new technique to try to solve an existing problem. Personally, I didn't have nearly enough knowledge of neural netwoks to understand most of what this paper is talking about.
My problems with (what I could understand of) this paper are:
I'm not really sure why the authors decided to force the system to maintain equity among the reservoirs. It seems like leaving this out might allow a more optimal solution. Also, I feel like the process of training sounded way too complicated to ever come into widespread use.