This thesis provides a novel method to improve distribution system state estimation by an effective approach to processing bad data in... Show moreThis thesis provides a novel method to improve distribution system state estimation by an effective approach to processing bad data in measurements. The first part of this research is focused on modeling distribution system state estimation with bad data rejection capability. We apply transmission level model to the distribution level system with specific properties, such as fewer real measurement data for state estimation in the distribution level system, three phase unbalance power flow and so on. For building a robust state estimation model, we optimize the system in the following ways: First, we optimize objective function. We use forecasted load as pseudo measurements. Then we apply different weights to distinguish the forecasted data and actual measurements in the state estimation. Second, we apply three phase power equation in the analysis. We add real power, reactive power, active line power, reactive line flow, voltage magnitude, phase angle and others as nonlinear constraints in the three phase model of state estimation. Third, we flexibly change objective function and constraints in the state estimation model. We can change objective function when state estimation method changes. Meanwhile, we can add power flow and bus limitations in the optimization to avoid state estimation results exceeding power system limitations. Finally, we conduct hybrid calculation. In the first optimization, we filter the bad data. Then, we add another weight to reduce the bad measurement weight and enlarge the good measurement weight. After this process, we get optimized state estimation results. The second part focuses on the implementation of the model. We explain how to preprocess testing case data in this part. The third part is case study. We use IEEE 34 node feeder to test this model. There are four test cases. One test case has no bad data. Other cases have bad data in different types of measurements. We compare these cases with conventional WLS approach. The results obtained from simulation indicate our model has better performance when there is bad data in measurements. M.S. in Electrical Engineering, May 2015 Show less