Optimal generation unit commitment in thermal electric power systems

This thesis is devoted to the optimal commitment of generation units in an all-thermal, single or multiple area, power system. The problem, known as unit commitment, is a nonlinear mixed program typically with thousands of 0-1 integer variables and diverse constraints. An exact optimal solution to the problem is only possible via (explicit or implicit) enumeration, which requires a prohibitively long computation time for large problem instances.;Lagrangian relaxation combines the solution of the dual of the unit commitment problem with feasibility search to obtain primal feasible solutions. The feasibility search is necessary because a solution to the dual seldom solves the primal, and because little theory is available to bridge the optimal dual and primal solutions. In this thesis, several new feasibility search procedures to find a near-optimal primal feasible solution from the dual solution are developed and tested. These procedures are independent of the data constituting different problem instances, and are more rigorous and systematic than the existing ones. With these procedures, Lagrangian relaxation is successfully and efficiently applied to both single and multiple area unit commitment.;Simulated annealing exploits the resemblance between a minimization process and the cooling of a molten metal. The method generates feasible solution points randomly and moves among these points following a strategy which leads to a global minimum in a statistical sense. Simulated annealing is very flexible for handling diverse and complicated constraints, such as those typical of the unit commitment problem. Simulated annealing is analyzed, evaluated and implemented for unit commitment in this thesis.;Five major algorithms, proposed in this thesis for unit commitment and reserve-constrained economic dispatch, are extensively tested and compared by numerical simulation on sample power systems of 10 to 100 units. The simulation results show the efficiency of the tested algorithms for large-scale unit commitment and demonstrate the general applicability of simulated annealing. A comparison with the priority list method and a study of the convergence rates of the subgradient type algorithms are also included in the simulation.;Two optimization approaches, Lagrangian relaxation and simulated annealing, are explored in this thesis for efficient and near-optimal unit commitment.