Control System Optimization for a Smart Residential Microgrid
Author:
Anthony Nyoyoko (Graduate Student)Co-Authors:
Dr. Peter Mark JanssonFaculty Mentor(s):
Dr. Peter Mark Jansson, Electrical and Computer EngineeringFunding Source:
Open Discourse Coalition (ODC)Abstract
This thesis began with a simple objective: to restore and optimize the control system of a smart residential microgrid so that it can respond intelligently to electricity prices while remaining safe and reliable on low-cost embedded hardware. The central research question is whether artificial intelligence can improve economic dispatch decisions without compromising system stability.
The work required restoring a legacy microgrid commissioned in 2015. Data acquisition was rebuilt using a Raspberry Pi 4 and an AcuRev smart meter to log voltage, current, power, power factor, and frequency at five-minute intervals. A stable data pipeline was achieved, with over 99 percent local logging uptime and 93 percent cloud upload reliability. This phase transformed the project from a control study into a grounded cyber-physical systems investigation.
A two-layer artificial neural network framework was developed. Layer 1 predicts next-hour PJM Real-Time Locational Marginal Prices using historical and time-based features, explaining about 92 percent of price variation. Layer 2 integrates predicted prices with real-time electrical measurements to guide operational modes such as load management and islanded operation. Although the controller achieved measurable economic improvement over a rule-based baseline, its performance was limited by constrained historical and seasonal data. Nevertheless, the implementation validated the complete data-to-decision pipeline and established a practical foundation for refinement.
The key finding is that economic optimization alone is insufficient. Safety must be explicitly enforced through hybrid control, combining AI prediction with rule-based protections.