Year: 2020
Type: dissertation
Abstract: A power grid comprises of generation, transmission and distribution components along various stakeholders. This dissertation is a study of artificial intelligence application throughout the power system. Forecasting of generation and demand play a key role in planning and operation of a power system and highly accurate prediction can result in large savings in large scale systems. In this dissertation, deep learning models have been explored in improving the accuracy on the two ends of the power system spectrum – generation and distribution. Renewable energy sources are gaining stronger foothold with every passing moment because of large supply and being environment friendly. Solar energy is the most popular among the renewable energy resources. However, solar energy, like most other renewable energy sources, are highly variable in nature as it is controlled by environment. So accurate solar energy forecasting is a challenging task as it can potentially change in an effervescent manner. Traditionally solar energy generation is forecasted using complex numeric weather prediction (NWP) models which depend on a lot of weather variables. In this work, solar generation has been forecasted using simple time series analysis and deep learning model RNN and LSTM outperform the other conventional methods and machine learning-based methods. Similarly, on the demand side, deep learning models trump traditional statistical time series model. These enhanced predictions are integrated to design a heuristic controller for a microgrid, and they successfully optimize the resource allocation in the microgrid. \nToday’s modern grid with improved communication infrastructure has opened the window of vulnerability to cyber-attacks. The reliability of a power grid now is susceptible to natural faults and new paradigm of cyber-attacks. Also, resiliency of the grid depends on successfully detecting and mitigating all types of anomalous events. This work establishes a benchmark for accomplishing this task using state-of-the-art machine learning techniques utilizing big data collected from PMU measurements. Sophisticated techniques like feature engineering, hyperparameter tuning are applied in improving performance, saving costs and tackling the curse of dimensionality. Overall, the studies conducted for this dissertation tackles the issues related to the power system from a modern perspective and can be used as a strong foundation to further improve the resiliency of the power grid.
Author: Taufique Atique
Language: English
Topic: Smart Grid and Power Systems
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3194794326