Application Of Nonlinear Predictor-Corrector Interior Point Method On Reactive Power Optimization

Reactive power optimization is one of the essential problems in the power system. Reactive power optimization is to make the power loss minimum by adjusting the terminal voltage of generators, the tap position of transformers and the reactive compensations such as capacitors or statcom, with precondition of all the appointed constraints under the conditions of the configuration parameters and loads given.Reactive power optimization can be intrinsically formulated using a large-scale mixed integer nonlinear programming model involving both discrete and continuous variables. How to enhance the computational speed and ensure the correctness of the computational result are main problems of reactive power optimization. At the same time, lots of discrete elements exist in the actual power system. Whether the disposal of discrete control variables is right will affect the rationality and correctness of the result directly.Aiming at the above problems, this paper carries the following researching work: Firstly, nonlinear predictor-corrector interior point method for reactive power optimization is proposed, and the process is carried out with normative C++ language. The proposed algorithm is divided into two phases of predictor and corrector. Firstly in the phase of predictor, the affine direction is calculated; then in the phase of corrector, the Newton direction is obtained with correction. Thus a longer iteration step can be obtained by use of the improved nonlinear predictor-corrector interior point method than pure primal-dual interior point method, so the convergence can be speeded up. In the realization of the process, the sparse matrix method of orthogonal linked list is used and the computational efficiency of the algorithm can be improved consumedly.Secondly, aiming at the several familiar discrete elements, this paper ensures its static model and the method by a penalty function. By selecting the timing of introducing the penalty function and the values of penalty factors, the discretization of discrete variables can merge well with the nonlinear predictor-corrector interior point method and does not cause some visible fluctuations in the optimization process.Thirdly, this paper validates the proposed algorithm through two samples of IEEE 14 and 118 systems, and compares it with the primal-dual interior point method which does not use predictor-corrector technology. Through the validation, the proposed algorithm is accurate, reliable and effective.