Microgrid Optimization Through Integration of A Real Time Controller
Abstract
Microgrid energy generation that includes non-renewable energy sources (e.g., fossil fuels) relies on long-term historical power demand data to forecast future power requirements and fuel use. Previous models seeking to optimize microgrid performance under time-varying loads introduced penalty functions to smooth the energy demand between generators and energy storage systems (ESS). However, these models requirecomplete demand profiles prior to generating feasible solutions, making them useful for analysis but impractical for operational use. To develop a near-real-time microgrid controller, we use a sample of omniscient model outputs, a history of demand data, and time-varying demand profile variables to train several machine learning models to estimate future demand, which we then use to prescribe energy generation output based on current conditions. The models create short-term forecasts for energy generation requirements regarding future demand profiles. The models also penalize changes to the generators' energy generation loads, which reduces fuel consumption, supports energy supply resiliency, and decreases microgrid maintenance costs. We compare the results obtained using our predictive models to a previous omniscient model that requires future demand profiles; our model is shown to have an average fuel consumption rate only 0.0016 gal/hr greater than the previous omniscient model over a75-day rolling test horizon, less than 0.09% from omniscient performance.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2023
- Accession Number
- AD1213146
Entities
People
- Drew E Chapman
Organizations
- Naval Postgraduate School