Usage of battery energy storage system (BESS) to facilitate demand response (DR) programs such as time of use (TOU) pricing can reduce utility bills for residential customers. However, using such a system for minimizing electric bills under these rate structures has the potential to cause an increase in emissions from the grid system. The increased emissions were majorly due to bulk energy storage of electricity produced by off-peak generators with higher emission rates and excess energy consumption due to battery inefficiency. BESS operating to optimize competing objectives to minimize utility cost and minimize CO2 emissions requires complex models that require an accurate forecast of future energy demand. These models get less effective as errors in demand forecasts increase. Demand forecasts for residential consumers are challenging due to high demand variability. Moreover, these models require computationally expensive mixed-integer linear programming (MILP) models in the day-to-day operation of BESS. In this work, a machine learning model (ML) was developed that attempts to predict an optimal battery schedule for an upcoming day based on easy to obtain information such as day of the week, month, previous day’s demand, average temperature, and relative humidity. The ML model’s utility bill and CO2 emission results were then compared to a no BESS scenario as well as a multi-objective optimization model based on perfect (OPT model) and forecasted (FORECAST model) demand data. The models were tested on two customers each from California and Arizona. The paired t-test comparison showed that the ML model results were not statistically different from the FORECAST model. The ML model was able to capture 65% of potential cost savings that could be generated from the OPT model. The model was also efficient in balancing the reduction of utility costs as well as CO2 emissions. Moreover, it requires less time and effort as is required for building and maintaining the FORECAST model.
Library of Congress Subject Headings
Energy storage--Economic aspects; Energy storage--Environmental aspects; Machine learning; Electric utilities--Rates--Time-of-use pricing
Industrial and Systems Engineering (MS)
Department, Program, or Center
Industrial and Systems Engineering (KGCOE)
George, Kevin, "Design and Evaluation of a Machine Learning Based Model for Optimization of Residential Battery Energy Storage System Scheduling for Cost and Emissions Reductions" (2022). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus