Data Analytic-based Adaptive Forecasting Framework for Smart Energy Management in Electrical Microgrids
This research establishes a framework to quantify and integrate the effects of renewable energy intermittency, energy market, policy decisions, and environmental constraints into the grid. Solutions are proposed for integration of solar energy into the electric grid addressing its variability in real-time operations and its uncertainty in the day-ahead energy market. Novel methodologies are developed for autonomous energy management in electrical microgrids and for solar energy forecasting. Standard performance measures are considered based on the latest benchmarking references and performances of the proposed approaches are illustrated and compared to those of the current state-of-the-art and the common reference cases in literature.
Potential benefit that the proposed microgrid energy flow control scheme brings to the local microgrid owner and the solar forecasting's added value to the utility companies and load serving entities (LSE) are presented. Microgrid energy management is implemented using an evolutionary qualitative decision-making scheme based on Genetic-Fuzzy System (GFS) which combines the Fuzzy logic with Genetic Algorithms (GA) to find solutions to the multiple-objective optimization problem. Results are presented for different microgrid conditions and control scenarios and it is demonstrated how control of storage energy flow in electrical microgrids will improve management of demand-supply profiles, reduce electricity pay-rates, and reduce quantifiable amounts of air pollution. Microgrid owners looking into adopting a smart decision-making tool for battery management systems (BMS) can use the analytics developed in this work to see a return on investment (ROI) of greater than 1.
In the future, distribution and transmission grids across the world will incorporate vast amounts of solar energy (or other renewable sources) generation to minimize the amount of net load to be provided from conventional sources such as coal, petroleum and natural gas. As a result, a data analytics-based adaptive forecasting framework is developed combining data mining, statistical analysis and artificial intelligence. Application of the proposed system for day-ahead prediction of solar energy is considered. Models are developed, tested and verified utilizing a large dataset from the National Renewable Energy Laboratory (NREL) archive, the Automated Surface Observing System (ASOS), and the NREL solar position and intensity calculator (i.e. NREL-SOLPOS) sampled at 1-minute intervals during 8 years (2005 - 2012) at NREL site in Golden, Colorado, USA. A uniqueness of the developed framework is that an integrated serial time-domain analysis coupled with multivariate analysis was used for pre-processing to enhance the recorded data. The resulting enhanced dataset is used for adaptive training of the neural networks forecast engine. Standard performance measures are obtained. The forecast results are compared to those of the standard persistence approach and the state-of-the-art on solar energy forecasting methodologies. The proposed day-ahead solar energy forecasting framework brings an added value of 9.5% to 14.5% to the utility companies compared to a common reference case of persistence forecasting. This translates to more than $1 M annual revenue elevation for a 300 MW solar PV plant. The methodology is now ready to be deployed in San Antonio, Texas with data collected in the 41 MW Alamo 1 PV plant, the largest in the state, which is the first solar plant in Texas that is connected to the transmission grid allowing solar energy bidding into the market.