Font Size: a A A

Parallel Algorithms Of Power Flow And Unreliability Tracing For Large-scale Complex Power Systems

Posted on:2011-07-26Degree:DoctorType:Dissertation
Country:ChinaCandidate:B HuFull Text:PDF
GTID:1102360308957818Subject:Electrical engineering
Abstract/Summary:PDF Full Text Request
Calculation and analysis techniques of large scale power systems (PS) are facing many serious challenges due to continuously increasing in the size of modern PS and the computational complexity. Even for the most basic power flow calculation, the traditional serial calculation methods cannot satisfy the requirements of real-time analysis and control of large-scale PSs. However, the parallel computation technique provides a feasible and useful approach for solving these problems. As the low cost parallel Beowulf clusters become available, they provide an extremely attractive opportunity to solve the online simulation and calculation problems of PSs. Supported in part by the National Key Basic Research Program of China (973 Project No. 2004CB217908) and National Natural Science Foundation of China (No. 50577072), the main works of this research are focused on the parallel models and algorithms of power flow, reliability evaluation, and unreliability tracing (UT) for large-scale PSs.A pre-conditional Generalized Minimal Residual (GMRES) method is presented based on the characteristics of high dimensional sparse matrix and short vectors of Newton solution for power flow correction equations of large scale PSs. In order to search the best pre-conditional operator for the power flow calculation of large-scale PSs, the effect and performance of a few pre-conditional methods, including incomplete LU (ILU) decomposition, incomplete Cholesky decomposition, Jacobi, successive over-relaxation (SOR) and diagonal operator, were compared. The relevant GMRES methods associated with different pre-conditional operators above are also presented. The results of case studies indicate that the ILU pre-conditional operator requires less iterations and floating point operations than other pre-conditional operators, particularly for large scale PS's. When the number of buses in a PS is more than 3000, the number of floating-point operations using the ILU can be reduced to about 50% of that used in the traditional LU methods.Because there is no back/forward sweep-based dependence in the pre-conditional GMRES method and matrix-vector operations have the potential parallelism, a pre-conditional parallel GMRES method is proposed based on the structure features of Jacobi matrix. In this method, a set of quasi-diagonal parallel pre-conditional block- matrices are constructed to facilitate the parallelism of matrix updating. The number of its block-Jacobi pre-conditional matrices is determined by the number of parallel processors. A parallel computing method for the iterative updating of Jacobi matrix was also proposed using the parallel matrix-vector operational method and vectorization process of the sparse Jacobi matrix. Case studies indicate that the proposed parallel power flow method has excellent superiorities in parallel speedup and efficiency compared with the traditional LU parallel methods when the number of buses in PS is more than 3000.A parallel reliability evaluation model based on a dynamic task allocation technique is proposed based on the variation pattern of square difference of reliability indices along with the sampling times in Monte-Carlo methods. This model can overcome the excessive communications and over calculations in the bulk PS reliability evaluation based on the static task allocation technique. The proposed heuristic dynamic task allocation method is designed based on the De Moivre-Laplace central limit theorem and the curve fitting theory. Using the simulating information in the parallel task allocation process, a conservative evaluation technique for total sampling times has been used to maximize allocated processes in a single task and minimize the total number of communications. Case studies show that the proposed dynamic task allocation method can improve the efficiency of parallel computation.A large scale PS has a complex structure with a large number of components, and each component has a different impact on the system reliability. In general the unreliability tracing (UT) for a large scale PS requires a huge number of calculations because of the complexity of reliability evaluation process. Therefore, it is difficult to perform the UT analysis for a large scale PS using a serial computation technique due to its limited space-time processing capability. In order to solve the problems above, a parallel UT model for large-scale PSs is presented based on the proportional sharing principle (PSP) and the dynamic task allocation method. Only the failed components have the contributions for the system unreliability in a system state during the UT process. In other words, the components in normal operating condition do not participate in the unreliability tracing. The reliability indices are proportionally distributed to the failed system components based on the PSP. The case studies indicate that the developed technique can rapidly distribute the system unreliability indices to components impartially and reasonably, and recognize the weak parts of a large scale PS effectively.With the continuous development of high voltage direct current (HVDC) transmission techniques and increasing practical HVDC engineering applications, modern PSs are becoming trans-provincial and cross-regional AC/DC hybrid PSs. In order to recognize the weak parts of an AC/DC hybrid PS quickly and accurately, a parallel UT model and algorithm for large-scale AC/DC hybrid PS are proposed using the UT theory and the parallel computation technique. Based on the features of AC/DC hybrid PS reliability assessment process, the proposed UT model is formulated as a multi-layer chained model. In other words, the reliability indices are layer-wise distributed in the reliability calculation process, and the contribution of each component to the AC/DC hybrid system unreliability can be obtained by merging the intermediate contribution results. Case studies show that the proposed chained parallel UT method for AC/DC hybrid PS is high-efficient and reasonable.
Keywords/Search Tags:large-scale power systems, parallel computation, power flow, reliability evaluation, unreliability tracing
PDF Full Text Request
Related items