|
|
| 說明 | 96 p |
| 附註 | Source: Dissertation Abstracts International, Volume: 61-05, Section: B, page: 2681 |
| Adviser: Ali Keyhani |
| Thesis (Ph.D.)--The Ohio State University, 2000 |
| A novel technique to estimate and model synchronous generator parameters from operating data is presented. First, a methodology is developed to investigate the feasibility of estimating armature circuit and field winding parameters of synchronous generators using the synthetic data obtained by the machine natural <italic>abc</italic> frame of reference simulation. A proper data set required for estimation is collected by perturbing the field side of the machine in small amounts. The recursive maximum likelihood (RML) estimation and Output Error Estimation (OEM) techniques are used for parameter estimation. For each estimation case, the estimation performance is also validated with noise corrupted measurements |
| In the second stage, the methodology developed previously is implemented to estimate armature and field winding parameters of an actual large steam turbine-generator from real time operating data. An artificial neural network (ANN) based estimator is used to model estimated saturated mutual inductances based on the generator operating conditions. It has been observed that <italic> L<sub>aq</sub></italic> estimates are very sensitive to the accuracy of δ measurements for small angles |
| In the third stage, the rotor-body parameters of the same large steam turbine-generator are estimated and modeled from large excitation disturbance and fault event data. For each set of disturbance data collected at different operating conditions, the rotor body parameters are estimated using OEM technique. ANN based estimators are then used to model the non-linearities in the estimated parameters based on the generator operating conditions |
| Finally, another large steam turbo-generator is identified and modeled from its operating data to verify the generality and validity of developed techniques in previous stages |
| The developed models are validated with measurements not used in the training of ANN and with large disturbance responses. Validation studies show that ANN models can correctly interpolate between patterns not used in training. It is expected that richer data set collected at different loading and excitation levels would improve the performance of such ANN models |
| It has also been shown through extensive simulations that estimated machine parameters clearly outperform the manufacturer parameters and through which the internal variables of the machine can be correctly predicted during large transient events |
| School code: 0168 |
| 主題 | Engineering, Electronics and Electrical |
| Computer Science |
| Statistics |
| 0544 |
| 0984 |
| 0463 |
| ISBN/ISSN | 0599766646 |
