未受限網路最佳化問題及分佈式狀態估計的解法

Abstract

首先, 我們提出一個快速的方法來求解非線性大型實際系統的未受限 最佳化問題, 它是一個結合近似比例梯度法及具有線搜尋的方塊高斯－塞 德法。我們證明了我們的方法是完全收斂的, 並且由測試的結果得知, 我 們所提出的方法, 其效率遠較以稀疏矩陣為基礎的方法優越。 再來, 我們提出一個可以應用在電力系統分佈式狀態估計上的部份非同步 方塊賈可比法。我們將系統分成p 個區域, 且每個區域是由至少擁有一部 電腦系統的局部控制中心所支配, 這些非集總式的電腦系統是由一套通訊 網路所連接而成。分佈式狀態估計和分佈式錯誤資料處理架構是被設計成 由這些電腦網路所完成。我們所提出的分佈式狀態估計因具有下列的性 質, 故可保證它的實用性: 1)它僅使用在非集總局部控制中心的電腦, 2)它在一個小規模的電腦通訊網路上執行, 3)當通訊鏈上偶然的錯誤發生 時, 它亦能收斂, 4)非集總式的電腦系統可不必做同步運算。 而且, 我們的分佈式錯誤資料處理架構能在每個區域所相對應的區域電腦 系統上, 各自獨立的處理錯誤資料。 最後, 我們提出一個實驗性質的個人電腦網路來製成電力系統分佈式狀態 估計器,我們的測試狀況考慮有計算非同步、通訊誤差及通訊連接失敗等 情形。我們使用 NetBIOS介面來執行在分佈式狀態估計中所有的通訊需求 。我們在所設置的個人電腦網路上, 以 IEEE 118 匯流排系統測試了幾個 分佈式狀態估計的例子, 而最後被估計出的狀態皆符合要求。 First of all, we present a fast descent method that combines an approximatescaled gradient method with a block Gauss-Seidel with line search method tosolve unconstrained optimization problems for nonlinear large practicalsystems. We prove that the method is globally convergent and demonstrate itssuperior efficiency compared to a sparse matrix technique based method byseveral weighted least square problems of power system state estimation. Secondly, we present a partially asynchronous block Jacobi method basedpower system distributed state estimator. The system under consideration ispartitioned into p areas, each of which is governed by a local control centerthat has at least one computer system. These decentralized computer systemsare linked by a communication network. The distributed state estimator anddistributed bad data processing schemes are designed to be carried out in thiscomputer network. The proposed distributed state estimation possesses thefollowing properties, which ensure its implementability: 1) it uses only thecomputers available in the decentralized local control centers, 2) it isexecuted in a small-scale computer communication network, 3) it can convergeeven if errors occasionally occur in the communication links, and 4) thedecentralized computers need not be synchronous. Furthermore, our distributedbad data processing schemes can process the bad data in each areaindependently using the corresponding local computer system. Finally, we show an experimental PC-network implementation of thedistributed state estimator. Our experiments consider the computationalasynchronism, communication errors, and failures of communication links in acomputer network. We use NetBIOS interface to carry out all the communicationneeds in our distributed state estimator. We have tested several cases ofdistributed state estimation on the IEEE 118-bus system in this PC-network,and the final estimated states in all cases are satisfactory.