2026 ASEE Annual Conference & Exposition

Microgrids are small, self-sufficient power systems that can operate independently using their own distributed energy sources and can provide services to clients even during main grid outages. Smart residential microgrids can provide a practical and educational platform for testing and teaching renewable energy systems, data analytics, and artificial intelligence based control strategies. This paper presents the development of Layer 2 of a two stage Artificial Neural Network (ANN) framework designed for intelligent operational control of a smart residential microgrid. Layer 1 of the system focused on predicting the next hour of PJM’s Locational Marginal Price (LMP) for the Citizens Electric node (PNODE 3387777) at ________ University using real operational market data. PJM’s Data Miner 2 API provides both Day Ahead (DA) and Real Time (RT) LMP data; however, RT prices are only published the following business day. To enable proactive decision making, Layer 1 was developed to predict the RT price at least one hour ahead of operation time. Layer 2 uses these predicted prices to guide real time control actions within the residential microgrid. The AcuRev 2000 smart meter communicates with a Raspberry Pi controller through an assigned Ethernet port, automatically saving daily CSV logs every 5 minutes via a scheduled crontab task. Under normal operation, each daily AcuRev CSV (e.g., 07152025.csv) contains 288 entries corresponding to 5-minute intervals throughout the 24-hour day with small deviations (±1 row) only during reboots or midnight overlaps. Across July to August 2025, the system achieved 99.2% AcuRev logging uptime and 93.9% Dropbox upload success, confirming high data completeness and reliability. Each CSV records live electrical parameters for multiple residential loads; voltage (V), frequency (Hz), current (I), real power (W), reactive power (VAR), apparent power (VA), and power factor (cosθ), which, combined with hourly weather data (temperature, humidity, pressure, solar irradiance, and wind speed) and the predicted next hour LMP from Layer 1, form the training inputs for the intelligent controller. The training dataset spans July 1 to August 31, 2025, and the validation dataset covers September 1st to September 30th, 2025, a three-month period of continuous, stable operation. The trained controller determines optimal load operation modes, Normal, Load Management, Pre Heating / Pre Cooling every hour and Islanded (Isolation from the grid) every 5 minutes based on cost, grid frequency, and voltage stability constraints. Assessment will compare simulated cost savings, frequency and voltage stability, and ANN accuracy (MAE, MAPE) against a baseline rule-based control done during the grid’s inception in 2015. Preliminary results demonstrate reliable data acquisition and the foundation for real time predictive control. This work fits within the ASEE ECE Division topic area “Current ECE Issues and Integration into the Curriculum , AI/ML Applications in Smart Grids and Real Time Analytics.” The project has a direct impact on engineering student education, as it provides a hands-on, research driven environment where students apply artificial intelligence, data acquisition, and control theory to a functioning residential microgrid, bridging the gap between classroom learning and real-world distributed and renewable energy systems.