Date of Award

6-10-2024

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2024

Department

Computer Science and Engineering

First Advisor

Salem Al-Agtash

Abstract

Microgrids have made a revolutionary change in the realm of energy distribution due to the features that they offer, including localized, resilient, and sustainable energy solutions. Operating renewable resources in a microgrid while maintaining generation-load balance and acceptable voltage-frequency limits has been an open research problem. This thesis presents smart python agents for microgrid systems to automate the operations and control of microgrid renewable resources in an effort to provide resilient solutions to the intermittence issues that could potentially arise within the microgrid energy system. The smart agents operate the microgrids by not only integrating the use of renewable energy resources but also optimizing energy consumption while maintaining a balance between power generation and load and keeping voltage and frequency within limits. We used SPADE (Smart Python Agent Development Environment) for the design and implementation of the microgrid multi-agent system. We assign each agent a different set of functionalities and responsibilities. These include reinforcement learning and forecasting of end-user energy demand (load) and the solar and wind power generation. We used four types of forecasting models, namely linear regression (as a baseline), random forest regression, gradient boosting regression, and long short-term memory (LSTM). The agent design embeds these forecasting models for load power consumption, solar power, and wind power. We used SCU’s 2021-2023 load dataset for training and testing, which covers the energy usage of several residential and academic buildings across campus. We also used Kaggle datasets for the solar and wind forecasting. The forecasting results show a high degree of accuracy for the load models when compared to solar and wind. This is because, for the load, we used a large temporal dataset that lends itself to sequence models and specifically LSTMs. In the case of solar and wind, though, the variance and quality of the dataset degraded the forecasting models’ predictive power. Agents also embed reinforcement learning to achieve optimal power distribution and energy market participation. This minimizes the cost of the system (selling unused power to the grid), while maximizing the use of available renewable energy. We use XMPP and ACL protocols for agent communication and authentication. Therefore, the SPADE multi-agent framework presented in this thesis proves to be a viable technology solution for microgrid systems. It can further be enhanced to accommodate real-time operation and control in real-world scenarios.

Download

Share

COinS