A Distributed Dynamic State Estimator Using Cellular Computational Network

The proper operation of smart grid largely depends on the proper monitoring of the system. State estimation is a core computation process of the monitoring unit. To keep the privacy of the data and to avoid the unexpected events of the system, it needs to be made fast, distributed, and dynamic. The traditional Weighted Least Squares (WLS) estimator is neither scalable, nor distributed. Increase in the size of the system increases the computation time signi cantly. The estimator can be made faster in di erent ways. One of the major solutions can be its parallel implementation. As the WLS estimator is not completely parallelizable, the dishonest Gauss Newton method is analyzed in this dissertation. It is shown that the method is fully parallelizable that yields a very fast result. However, the convergence of the dishonest method is not analyzed in the literature. Therefore, the nature of convergence is analyzed geometrically for a single variable problem and it is found that the method can converge on a higher range with higher slopes. The e ects of the slopes on multi-variable cases are demonstrated through simulation. On the other hand, a Cellular Computational Network (CCN) based framework is analyzed for making the system distributed and scalable. Through analysis, it is shown that the framework creates an independent method for state estimation. To increase the accuracy, some heuristic methods are tested and a Genetic Algorithm (GA) based solution is incorporated with the CCN based solution to build a ii hybrid estimator. However, the heuristic methods are time-consuming and they do not exploit the advantage of the dynamic nature of the states. With the high data-rate of phasor measurement units, it is possible to extract the dynamic natures of the states. As a result, it is also possible to make e cient predictions about them. Under this situation, a predictor can be incorporated with the estimation process to detect any unwanted changes in the system. Though it is not a part of the power system to date, it can be a tool that can enhance the reliability of the grid. To implement the predictors, a special type of neural network named Elman Recurrent Neural Network (ERNN) is used. In this dissertation, a distributed dynamic estimator is developed by integrating an ERNN based predictor with a dishonest method based estimator. The ERNN based predictor and the dishonest method based estimator are each implemented at the cell level of a CCN framework. The estimation is a weighted combination of the dishonest module and the predictor module. With this three-stage distributed computation system, it creates an e cient dynamic state estimator. The proposed distributed method keeps the privacy and speed of the estimation process and enhances the reliability of the system. It ful lls the requirements of the deregulated energy market. It is also expected to meet the future needs of the smart grid.