New Methods Of Identifying Synchronous Machine Parameters Based On Load Rejection Test And Online Time-Domain Test

Synchronous machine is one of the key devices in power systems and synchronous machine models play an extremely important role in power system analysis. Power system stability analysis and designs of control strategies for control equipment are fully dependent on reasonable synchronous machine modeling and accurate parameters in models.Parameter identification using measured data is the most important method for determining synchronous machine parameters. It is still a challenging topic although many researches have been conducted in this area in the past decades. Considerable efforts have been paid to load rejection tests and online time-domain tests for parameter identification of synchronous machines so far because of their advantages superior to other methods. However, there are deficiencies and shortcomings in existing researches of these two measurement-based methods. As well known, noises in measurements are unavoidable in any test. Unfortunately, the response curve method which is a popular technique under load rejection test cannot handle noisy measurements properly, whereas the optimization modeling method under load rejection test not only requires selection of initial parameter values but also may result in large estimation errors caused by measurement noises. Under load rejection test, it is often needed to implement parameter conversion between a time-constant-based operational inductance model and an equivalent-circuit based model due to requirements in calculations. Although a conversion formula under some condition is provided in the IEEE std 1110-2002, it failed to address errors of parameters which will be caused by error transfer in the conversion. In fact, this problem has never been presented in all existing researches. In online time-domain test, nonlinear dynamic saturation effects in a disturbance process have been missed in majority of previous work on online estimation of generator parameters. A third-order model for synchronous machine parameter online estimation has been used in a few published papers. However, the existing identification models need to be further improved and more effective estimation methods are required to deal with the non-linear feature of the third-order model.In this thesis, new identification models and parameter estimation techniques are investigated systematically and deeply on the basis of load rejection test and online time-domain test. With an objective in improving accuracy of estimated parameters and increasing capacity of handling measurement-noises, the thesis focuses on effects of measurement noises on estimation results, errors in parameter conversion and solution to avoiding error transfers, nonlinear effects of magnetic saturation in a disturbance process and the nonlinear feature of third-order model.A novel method for estimating d-axis parameters of synchronous generators under d-axis load rejection test is presented. The essence of the method is to simultaneously identify the coefficients of two single-input single-output transfer functions. From a viewpoint of modeling, it is a time-constant-based operational inductance model. Firstly, linear regression equations are constructed using the q-axis voltage, excitation current and their numerical integrations. Secondly, the regression coefficients with measurement noises are estimated using an instrumental variable approach. Finally, the d-axis parameters are calculated from the estimated regression coefficients. Simulation results indicate that the proposed method has a better performance than the traditional response curve method or optimization modeling method.The method of single-input single-output transfer function is directly used to estimate q-axis synchronous inductance and time constants under load rejection test. Essentially, this method is to identify q-axis operational inductance. Simulation results demonstrate the effectiveness of the method.A method to accurately determine d-axis equivalent circuit parameters from an operational inductance model is presented. In existing researches, either a time-constant based operational inductance model or an equivalent-circuit based model is obtained under load rejection test. However, the parameter conversion between the two models was not discussed. It is necessary to perform a parameter conversion when estimated parameters are different from those required by stability analysis software. In the case where the d-axis is represented by one damper winding and the parameters in operational inductance model have been estimated, the IEEE Std 1110-2002 has recommended that equivalent-circuit parameters be calculated using analytical formulas. However, the direct computation method using the analytical formulas may result in unreasonable parameter values which deviate far away from their true values because of the errors that always exist in the measurement-based operational inductance model. Literature searches indicate that this issue has not been addressed so far. To obtain accurate equivalent circuit parameters, the thesis presents a two-step method. Initial values of equivalent circuit parameters are computed first using the analytical formulas; and then a prediction error method is used to estimate the parameters using a hybrid state model of generator. Simulation results demonstrate that the direct computation method using the analytical formulas does cause large parameter errors whereas the presented method is very effective under any designated noise level. The presented method can be used as a complement to the analytical formula method given in the IEEE Std 1110-2002.Majority of previous work on online estimation of generator parameters ignored the nonlinear dynamic saturation effect in a disturbance process. In the thesis, a new estimator is presented to incorporate the saturation effect in a disturbance process. The estimator focuses on recognizing field winding and d-axis damper winding parameters of both salient-pole and round-rotor generators. Simulation results demonstrate that neglecting the saturation effect in an identification procedure will create a large estimation error. Simulation results from a lot of noisy samples also demonstrate that the proposed estimator not only can accurately capture the non-linear dynamic saturation effect in a disturbance process but also has a good capacity of handling measurement noises.A new method to estimate the parameters of a synchronous generator using the square-root unscented Kalman filter (SRUKF) is also presented in the thesis. A new third-order model for the parameter estimation of both round rotor and salient generators is developed first and then the SRUKF method is applied to the third-order model to perform the joint estimation of state variables and unknown generator parameters. The essential feature of the presented estimation method is incorporation of nonlinearity of the model. Simulation results demonstrate the effectiveness of the proposed method in parameter estimation of synchronous generator. The estimation processes of generator parameters steadily converge to the estimated values, whereas the estimation processes of state variables are consistent with the dynamic responses in the numerical simulations. Simulations indicate that the presented method can obtain more accurate results than the traditional extended Kalman filter (EKF) method.