Razi University

State estimation is a fundamental task in the monitoring and operation of electric power distribution networks, where accuracy and speed play a crucial role in maintaining real-time situational awareness of the grid. This thesis presents the development and evaluation of a multi-area, data-driven framework for distribution network state estimation. In the proposed framework, the network is partitioned into five areas; following feature engineering based on mutual information, various models—including Linear Regression, Lasso, ElasticNet, Multilayer Perceptron (MLP), and an Ensemble model—were employed to estimate voltage magnitude and phase angle. The method's performance was evaluated using standard 33-bus and 69-bus networks, utilizing metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Maximum Error (MaxError). The results demonstrated that the regression models achieved voltage estimation accuracy comparable to the MLP, while their training times remained under 2.5 seconds for both networks; in contrast, training the MLP required approximately 45.2 seconds for the 69-bus network and 47.1 seconds for the 33-bus network. Under conditions of full observability, the ensemble model achieved a maximum error of 0.309% for voltage and 0.1337 degrees for phase angle in the 69-bus network. Corresponding values ??for the 33-bus network were 0.406% and 0.2294 degrees, respectively. Compared to the reference study [53], which reported a maximum voltage error of 0.30% and a phase angle error of 0.34°, the proposed method performed comparably in terms of voltage error and superiorly in terms of phase angle estimation for the 69-bus network; for the 33-bus network, although the voltage error was higher, the phase angle error remained lower than the reference value. Furthermore, the multi-area structure in the 69-bus network reduced the total training time from 153.94 to 117.71 seconds—a reduction of 23.5%. The study's primary innovation lies in presenting and evaluating an integrated framework comprising multi-area estimation, mutual information-based feature engineering, simultaneous comparison of regression and neural network models, and their combination into an ensemble model; this approach enables the simultaneous assessment of accuracy, computational cost, and robustness against measurement loss. Keywords: State estimation, distribution network, artificial intelligence, ensemble model.