Model Parameter Identification for Synchronous Generator Based on MeDA and Parameter Classification Approaches

• Main Authors: Li-YaWu, 吳利亞
• Other Authors: Hong-Tzer Yang
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/gd4bfj

碩士 === 國立成功大學 === 電機工程學系 === 105 === Nowadays, power system simulation is widely used for power system operation management and planning, and the accuracy of the synchronous generator model plays a critical role in power system simulation. Models with huge deficiencies may fail to reflect the system's behavior, especially the system's dynamic responses. It might lead to wrong predictions, causing incorrect decisions and operation, and end up with damage to the system or even a regional blackout that causes economical loss. The accuracy of a model comprises 1) proper structure and 2) correct setting of the model parameters. Since many power system simulation tools have been developed, equipped with well-designed templates for model construction, we assume that there is no problem with the model structure in this thesis. However, due to device aging, changes in the operating conditions, or the inaccessibility of some parameters, there is a veritable need for a parameter identification method to set or calibrate the model parameters. In this thesis, we propose a new parameter identification method based on a modified electron drifting algorithm (MeDA), which can obtain the best-fit parameter values by using only the comparison of the measurement data with the model outputs instead of the information of the model's inner equations. To enhance performance, parameter selection and classification are applied. The parameter selection analyzes the parameter-output relationship and selects parameters that are the most influential and identifiable. The parameter classification categorizes the selected parameters into groups according to their effects on individual outputs. Then, it applies the four-stage algorithm to effectively solve the optimization problem. In addition, MeDA has a unique feature of using a database that stores the search historical data. It can efficiently utilize the data collected from the parameter classification, and thus perfectly integrates the parameter classification with the optimization search. The feasibility and performance of the proposed method are verified through three tests. They confirm that the proposed method performs better than other methods in both simulations and experimental tests. Last but not least, the proposed method is easy to implement, showing its great applicability to various areas.