Research On The Parallel Algorithm For Unit Commitment Based On Lagrange Relaxation

Traditional economic dispatch implements the self-balanced mode by region. With the increasing of the installed unit capacity and the development of the interconnected power system, the dispatch pattern changes into integration and coordination, which requires the unified optimal allocation of generation resources. Unit commitment is the core of economic dispatch in power system, and it’s the key problem of safe and economic operation of power system. In the view of mathematics, unit commitment is a high-dimension, non-convex, non-linear, mixed-integer programming problem. However, the unified computing of large power system unit commitment problem can’t be solved by existing optimal methods. In this paper, researches are conducted with establishment of the model of large power system parallel unit commitment and improvement of the parallel algorithm which will improve the computing speed and accuracy as the central point.This paper first introduces the model of traditional unit commitment and economic dispatch based on which the model of Lagrange relaxation dual parallel unit commitment is established. On the basis of analyzing the relations of Lagrange multipliers, sub-gradient and the spare capacity, the improved iteration strategy of the multipliers is adopted, in which historical sub-gradients are considered. Compared with the generation scheduling results of unified serial computing, parallel unit commitment algorithm improved the computing speed greatly while keeping a sufficient accuracy.This paper gives heuristic methods to further improve the computing speed and accuracy of the Lagrange relaxation parallel unit commitment algorithm introduced before. Aiming at the feature that the parallel algorithm has difficulty in getting feasible solution, quasi priority list method is adopted to construct feasible solution by fine-turning the iterated solution. Through analysis of the reasons that more units open than expected in the approximate optimal solution, the method of unit decommitment is proposed to close some unnecessary units. The example results show that quasi priority list method can accelerate the convergence speed to reduce the computing speed while unit decommitment method can improve the accuracy greatly.Based on the research of the theory of parallel computing and the introduction of Message Passing Interface (MPI),the programming strategy is designed to implement the parallel unit commitment algorithm. In the process of programming, the common problems of parallel algorithms are solved, such as task dispatch, communication and synchronization. The slave processes will wait for a long time while the master process is constructing feasible solution and determining convergence, and the master parallelization strategy is adopted to solve this problem.