2026 ASEE Annual Conference & Exposition

A microgrid is a small, independent power system that can use renewable energy and keep running even when the main grid goes down, and it is becoming an increasingly important platform for the grid of the future based on its ability to test renewable energy integration and intelligent energy management at the small scale. This work-in-progress paper presents the development of layer 1 of a two-stage Artificial Neural Network (ANN) designed to predict the next-hour of PJM’s Locational Marginal Price (LMP) for a smart residential microgrid at ________ University, using Citizens Electric node (PNODE 3387777) real operational data as the market reference price and forecasted hourly weather parameters, which will be discussed in greater detail by this paper. The system runs on a Raspberry Pi 4 controller that communicates with PJM Data Miner 2 through its public API to automatically retrieve market data. Further, by using a Davis Vantage Pro2 weather station via a TP-Link TL-MR3020 router in access-point mode it runs Meteobridge firmware, which is linked to the OpenWeatherMap API to provide 24 hour local weather forecasts. The Pi 4 is configured through crontab ( a scheduler on Linux) to download Day-ahead electric grid pricing (posted daily between 12:00p.m. and 1:30 p.m. ET) at 5:58 pm each day. These data serve as the forecasting basis for the PJM Real-Time (RT) market prices. The RT market prices (posted on business days between 11:00 a.m. and 12:00 p.m. ET and representing the previous day’s actual market performance) are also scheduled in crontab for automatic download at 1:00 p.m. each day, and these data serve as validation inputs for the ANN training process .For example, on Mondays, the RT postings include data for the entire weekend. Because PJM’s RT prices are only available after the fact, the ANN is designed to perform hour-ahead real-time prediction of upcoming RT market prices using historical Day-Ahead, RT and weather forecast data. Analysis shows that the PJM RT prices at the Citizens Electric node differ by about 5–8 percent from day-ahead averages, providing meaningful training patterns for the ANN. Also, Weather forecast data (irradiance (W/m²), temperature (°F), humidity, wind speed (mph), and pressure (inHg)) are automatically saved as CSV files at 5:58 p.m. daily for analysis. A future layer 2 phase would use these predictions with actual AcuRev 2000 power meter logs of voltage (V), current (A), power (W), and power-quality parameters to enable intelligent control optimization of the microgrid. The ANN-layer one model has been developed in MATLAB/Simulink to learn from forecasted weather data and PJM’s LMP data and predict the next-hour PJM RT LMP in operational real-time. The goal is to create a simple, low-cost predictive tool that enhances the economic intelligence of residential microgrids, enabling proactive decision-making and improved cost efficiency. This work helps students understand how energy policies and market economics influence real engineering decisions, giving them practical insight into managing and planning sustainable microgrids.