Research Brief: Electric Load Forecasting
Project Brief
Type of Research: Electric Load Forecasting
Motivation for Research: The constant evolution of the electric grid with the integration of generation from renewable sources and “smart” components makes peak electric load management an essential aspect to ensure the grid’s reliability and safety. In order to pass the financial burden of managing these loads on to the consumers, utilities around the world have established peak load charges that can amount to up to 70% of electricity costs in the case of the United States of America. These pricing schemes have created a need for efficient electric load management strategies that consumers can implement in order to reduce the financial and environmental impact of peak electric loads.
Research Approach
During the fall semester of 2017 and under the advice of Professor Katie T. McConky, Ph.D., a RIT Master of Science in Industrial Engineering student presented a thesis demonstrating how machine learning techniques could be used to forecast upcoming days when peak electric loads would occur. This work demonstrated how consumers could achieve significant financial savings by incorporating the intelligence provided by these forecasting efforts into their demand response strategies. Taking into consideration that this work was performed using load data from a circuit without behind the meter renewable electricity generation (BTMREG), Professor McConky and Ph.D. student Omar Aponte decided to investigate if similar results could be obtained for a circuit with BTMREG and document the effects of BTMREG adoption on forecasting methodologies.
Aponte and McConky have enhanced the peak electric load day (PELD) forecasting methodology presented in 2017 improving its performance and making it applicable to consumers regardless of the presence of BTMREG. They completed a data-driven analysis of a circuit’s yearlong electric load and solar production that identified and described how the load profile is impacted by the adoption of BTMREG. Following this analysis, they tested their PELD forecasting methodology incorporating autoregressive integrated moving average (ARIMA), classification and regression trees (CART), random forest (RF), artificial neural network (ANN), and ensemble based forecasting models in scenarios with and without BTMREG. Aponte and McConky have produced the first of their kind side-by-side performance comparisons of electric load and PELD forecasting models for scenarios with and without BTMREG. They have also produced the first of its kind PELD forecasting model savings comparison for scenarios with and without BTMREG.
Key Findings
- Counterintuitively, there can be more financial savings to be achieved by consumers using PELD forecasting methodologies after adopting BTMREG.
- The adoption of BTMREG has negatively affected the performance of the regression-based models evaluated.
- The classification-based models evaluated have not been affected by the adoption of BTMREG.
- The peak loads observed when BTMREG is present, tend to occur during the hours when BTMREG is either low or inactive and normal operations are still ongoing.
- Existing demand response strategies need to be revised as soon as a consumer adopts BTMREG in order to ensure maximum reduction of peak load charges.
Looking Ahead
This work is the basis for Aponte's Ph.D. dissertation, which he expects to defend at the end of the Fall 2021 semester. While many questions remain, two aspects that Aponte will be focusing on are: 1. Developing and testing a robust threshold prediction method for threshold based peak electric load day models; 2. Developing a machine learning based ensemble method that reduces the number of false positive alerts observed.