Stochastic Dynamic Programming for Long Term Hydrothermal Scheduling Considering Different Streamflow Models

This paper is concerned with the performance of stochastic dynamic programming for long term hydrothermal scheduling. Different streamflow models progressively more complex have been considered in order to identify the benefits of increasing sophistication of streamflow modeling on the performance of stochastic dynamic programming. The first and simplest model considers the inflows given by their average values; the second model represents the inflows by independent probability distribution functions; and the third model adopts a Markov chain based on a lag-one periodical auto-regressive model. The effects of using different probability distribution functions have been also addressed. Numerical results for a hydrothermal test system composed by a single hydro plant have been obtained by simulation with Brazilian inflow records.