Optimization of the unit commitment problem by a coupled gradient network and by a genetic algorithm. Final report

This report presents our work on the unit commitment problem. We propose a coupled gradient-type neural network for its solution as well as a customized genetic algorithm-based solution. The unit commitment problem is an interesting and difficult constrained mixed-integer mathematical programming problem, where total operational cost is to be minimized subject to some unit and generation constraints. In this report, we present computer simulations of a coupled gradient network and genetic algorithm for both the static and temporal unit commitment problems. We compare solutions found by the gradient network and the genetic algorithm with solutions found by Lagrangian relaxation. For the static unit commitment problem, the coupled gradient network and genetic algorithm were able to obtain high quality solutions. Although we were able to obtain good local solutions for the temporal unit commitment problem, the coupled gradient net had a high probability of converging to solutions which are not feasible, while the genetic algorithm required excessive computation time. Several interesting issues relating to improving the quality of the proposed algorithms for the unit commitment problem and potentially exceeding those of the Lagrangian relaxation algorithm are also raised in our report.