Study On High Performance Realization Of Power System ILL-conditioned Power Flow Calculation

As the fundamental of power system analysis,power flow calculation can help dispatchers understand the running status of power system.The result of power flow calculation is the basis for other stability calculations.Factors like heavy loads and unreasonable network parameters can cause an ill-conditioned Jacobian matrix,which leads to regular algorithms like Newton-Raphson and fast decoupled method not converging.Thus,algorithms need to be improved to solve the ill-conditioned power flow problems.In addition,the high performance realization of large-scale ill-conditioned power flow calculation is also an important aspect in application.This thesis studies the high performance realization of power system ill-conditioned power flow calculation concentrating on computational efficiency.Firstly,several mainstream ill-conditioned power flow algorithms are studied and compared,and the self-adaptive LM method is chosen to improve the convergence of power flow calculation.Secondly,two sparse solution methods in LM iterations are compared in terms of fill-ins and computational efficiency.Moreover,sparse Cholesky factorization is introduced to improve the speed of ill-conditioned power flow calculation.The widely-used supernodal Cholesky factorization is studied.It is found that the supernodal algorithm cannot exploit its advantages because of the extremely sparsity of the matrices,when the algorithm is applied to ill-condition power flow calculation.Thus,up-looking Cholesky is chosen,and the algorithm is improved according to the features of LM method.An implicit Cholesky factorization method for J~TJ+?I is proposed to improve matrix factorization speed in LM iterations.With the proposed method,redundant computation in regular methods can be reduced.Finally,the scheme of the high performance realization of power system ill-conditioned power flow calculation is proposed.And the ill-conditioned power flow calculation platform is developed to provide a useful tool for ill-conditioned power flow problem.The correctness and efficiency of the proposed scheme is tested by large-scale cases.Test results show that self-adaptive LM method can obtain the correct solutions for well-conditioned power flow and feasible ill-conditioned power flow problems.And the method can obtain the least square solutions for infeasible ill-conditioned power flow problems.Test results also show that the implicit Cholesky factorization method can accelerate the speed of factorization by 20%,compared with explicit method.And the developed platform can reduce nearly half of the overall running time compared with PSD-BPA.