Research On Feature Extraction And Assessment Method Of Power System Transient Stability Based On WAMS

The power system is a nonlinear, nonautonomous, and complex system which is in the turbulence all the time. It plays a significant role in guaranteeing the safe and stable operation of power system that how to release an accurate, fast, timely and reliable transient stability assessment (TSA) report to the power grid operators. The.TSA of power system are focusing on the rapidity of the assessment methods and the reliability of the TSA results. The wide area measurement system (WAMS) provides a lot of real-time operating data for the power system. However there is a lot of redundancy in the WAMS data, it is meaningful to extract information from the mass to dynamic information which characterizes the transient stability of power system and to mine the internal relations between various measurements in order to release an on-line TSA. It is also an important researching direction in the TSA domain to reveal the physical nature of the fault mechanism.Pattern recognition algorithm is one of the common methods of TSA, which employs the transient stability features to assess the power system transient stability by establishing their relationship. There are two key steps for this method: one is how to obtain the correct feature set which can characterize the power system transient stability, the other is how to utilize these features to obtain an on-line TSA. In view of a large number of domestic and foreign literatures and summarized, review and investigation and basic research and discussing with power system operators, this paper proposes the theoretical research and prototype application using pattern recognition method for TSA based on the WAMS data and the main research contents are as follows:(1)A spatial-temporal panel data modeling method was proposed in this study based on the real WAMS data, which could organize the different WAMS data from different PMUs in various places and the WAMS data in the same range of time together. This kind of modeling method could visually display the dynamic process of the key features in the first several seconds after the failure occurrence, which could help to pretreat the WAMS data more effectively.(2)In order to validate the correctness of the panel data modeling method, a data pretreatment and several test steps were proposed for the WAMS panel data. Firstly, the unit root test was employed to validate the stationary of the features. Secondly, the cointegration test was utilized to check out whether there was some balance relationship among the features. Finally, a regression equation was set which set the power angle as the dependent variable and set the other features as the independent variables in order to obtain the influence coefficients. In the IEEE8-generator36-bus test system, the simulation results proved the validity and universal property of the panel data modeling method.(3)A two-stage feature extraction algorithm was employed so as to pick up the key feature subset which could describe the dynamic process of the power system TSA. In first stage, the sensitive feature subset was selected from the original feature set based on the relative sensitivity principle in order to pick up the most active feature subset.In the second stage, the improved grey relation and cluster algorithm was utilized to analyze the subset from the output of the first stage carefully with the influence coefficients from (2) so as to find the important feature which could effect the power angle most in each grey cluster. In the IEEE8-generator36-bus test system, the simulation results proved that this kind of feature extraction method could maintain the dynamic process of the TSA mostly without destroying the real physical meanings.(4)According to the dynamic evolutionary process of the power system TSA, a novel TSA method based on the hidden Markov model (HMM) was proposed in this thesis. Moreover continuous HMM (CHMM) and discrete HMM (DHMM) were realized for different kinds of features input. In the IEEE8-generator36-bus test system, in the New England10-generator39-bus test system and in a real large power system, the simulation results proved the validity and effectiveness. Also the HMM method needed less training samples and converged more quickly to obtain the same accurate rate. (5) A prototype application on-line system was established based on the real WAMS data combining the "863" project and the National Grid Key project. This prototype application system was based on the cloud computing platform:Hadoop with some optimization in order to meet the real-time requirement of TSA. Moreover the mass real WAMS data could be pretreated and the TSA could be parallel calculated in this prototype system, which verified the feasibility and the application values.