Data Mining Of Electrical Power System Dynamic Response

Electric power systems are large-scale, complex non-linear systems disturbed continuously.The new characteristics as environment restriction,network interconnection and market-driven generation make the operation environment of power system more rigorous and impossible to be forecast,so it is more and more important to know operation dynamics of interest.The calculation of transient stability by the regular time domain simulation method is based on power system models,parameters and given fault cases.But it is very difficult to accurately simulate the real operation dynamics because the variant operating states,complex disturbance scenarios and inaccurate continuous or discontinuous time-variant parameters.In contrast,the disturbed power system trajectories recorded by data measuring instruments as Phasor Measurement Units(PMU) can reflect the real dynamics of power systems.Extracting deep-seated information from measured dynamic response is of great significance to master power system operation and stability characteristics.The work reported in this thesis is about data mining of measured power system response.The existent problems in application of phasor measurement is discussed as well as the state of art. Two algorithms are presented to process trajectories acquired by PMU,one for assessing stability margin based on Extended Equal-Area Criterion(EEAC),one for extracting time-variant oscillation performance based on wavelet ridge algorithm.Similar to numerical integration simulation,PMU can give accurate trajectories for a disturbed power system,and then a stable-or-not type conclusion might be concluded as a matter of experience.However,no quantitative stability information could be deduced from the trajectories.If power system operation point is close enough to stability bound,no matter how small a parameter changes,the system may not remain stable.So not only the stability should be judged but also the stability margin should be calculated.The algorithm presented in this thesis assesses stability margin of generator-swing trajectories acquired by phasor measurement units (PMU) based on EEAC,which gives necessary and sufficient condition and rigorous stability margin of transient stability.EEAC calculates stability margin based on trajectories so it is not only adequate to the analysis of simulated trajectories but also to that measured by PMU,just avoiding using inaccurate model parameters. Although EEAC has been widely used to quantitatively assess power system stability with digital simulations, new problems should be solved before applying EEAC to the real-time measured trajectories, where there is lack of both system models and parameters. Simulations show that the algorithm can accurately evaluate stability margin of the swing curves without any knowleage on either system models or parameters.During the process order-reduced network admittance matrix is derived by dynamic network equivalence.In the analysis of large systems,coherence equivalence is performed to accommodate to its characteristic of large number of nodes,great scale and complex transient performance.The application of PMU promotes the research of extracting dyanmic performance from measured trajectories,which is of great significance to discover mechanism of low frequency oscillation and design controllers.Power system is a complex non autonomous system,so its time response is non-stationarity,which is especially important to measured trajectories.When the power system is close to stability bound,disturbance may excitate strong time-varying non-linear kinetic characteristics.During the transient state process,different oscillation mode may coexist and vary.The eigenanalysis can't reflect this time-varying characteristic,nor can it be used in strong nonautonomous nonlinear systems.Nonlinear analysis can give nonlinear characteristic of oscillation,but it can't provide oscillation parameters and enforceable control measures. Based on a wavelet ridge algorithm, a novel approach is presented for extracting time-variant oscillation performance f...