Stochastic electric power system operations planning

The main objective of an electric power system operations plan is to be able to meet demand at the least possible cost, and with some specified level of reliability and adequate quality. This dissertation presents two new techniques for electric power system operations planning under conditions of uncertainties in both the load demand and unit availability. Electric power operations planning is generally formulated as a nonlinear mixed integer dynamic programming problem. The system hourly load demand is modelled as a Gaussian normal random variable. Unit outages are modelled as Markov processes with appropriate characteristics. The unit commitment-status variables are (0,1) integers while the unit loading/dispatch levels take on decimal values; thus making the problem a mixed integer one.;In both methods generalized stochastic Hopfield and Chua-Kennedy neural networks for the non-linear optimization problem are formulated. The first method formulates a Hopfield artificial neural network for the deterministic equivalent of the stochastic optimization problem. The neural network when started from an arbitrary state settles at a local minimum of the deterministic equivalent energy function. The problem solutions are the expected values of the unit commitment-status variables, and unit loading levels. In the second method, the fact that the system load demand and unit availability are random processes is utilized. Hence the unit commitment-status variables and the unit loading levels are modelled as sample path solutions of appropriately derived stochastic differential equations. Both methods are applied to power system data from a utility in the USA.;The major advantages of the new methods over those currently in use for the stochastic problem are the ability to commit and dispatch units simultaneously, and to account for the effect of forced outages on unit availability in a chronological manner. The operations plans obtained by using the methods developed in this dissertation are better than those obtained by conventional techniques due to the more realistic representation of the stochastic phenomena in the power system.