HEC Montréal, Canada, 2  4 mai 2011
Journées de l'optimisation 2011
HEC Montréal, Canada, 2 — 4 mai 2011
MA4 Programmation dynamique approchée : théorie et applications / Approximate Dynamic Programming: Theory and Applications
2 mai 2011 10h30 – 12h10
Salle: Cogeco
Présidée par Michel Denault
4 présentations

10h30  10h55
A Bayesian Approach to the Exploration/Exploitation Dilemma
We use a correlated Bayesian belief model to represent our uncertainty about the value function in ADP. Correlations between the values of different states allow us to learn about multiple states from a single decision. We apply the knowledge gradient concept from optimal learning to create a new exploration strategy.

10h55  11h20
Optimal Control of a Two Dam Hydroelectric Facility Under Time Varying Prices
We investigate optimal control strategies for the variable price electricity market operations of a hydroelectric facility comprising two dams linked in series by a common reservoir which receives the outflow from the first dam and supplies the inflow to the second dam. Using a dynamic programming approach, we are able to obtain some interesting insights about the behaviour of these plants and compare the similarities and differences of the pumpwaitrelease cycle of this two dam system to a similar one dam system. In particular, we show that the electricity price plays a more important role in a two dam system than in a one dam system.

11h20  11h45
Extending the Simulations and Regressions A.D.P. Approach to Value and Control a Hydropower System
We solve a control problem by using simulations and regressions, a common approach in approximate dynamic programming. Specifically, in a hydropower system problem, we have one exogeneous stochastic state variable, power price, and one endogeneous state variable, the water level in the dam. Neither is discretized.

11h45  12h10
The Optimization Strategy of the ModelBased Analysis of Cell Divisions in Arabidopsis Leaf
We studied the temporal control of epidermal cell divisions in the Arabidopsis leaf. We built a mathematical model with several parameters to study the rules whereby different cell types divide and expand. To optimize the parameters, the experimental data were fitted into a computational model by the downhill simplex method or NelderMead method.