(492a) Industrial Battery Storage Dispatch and Optimization Using Gaussian Process Regression and Bayesian Decision Theory

Electrical load forecasting has been a large focus for research over the past decade with the intention to improve grid stability, efficiency, and expansion. This is typically done at the grid-scale but has also been done for individual industrial facilities [1]. At the facility level, these load forecasts are used to inform facility personnel of opportunities to reduce overall energy use and utility expenses. Typical load forecasting utilizes regression and machine learning techniques, and some grid-scale forecasting has used gaussian process regression [2] [3] . The practice of peak shaving in industrial facilities, made possible by load forecasting, has been shown as a viable way to reduce energy expenditures and improve industrial load factors [4] [5]. These methods utilize time-of-use (TOU) utility rates to cut utility usage during on-peak hours. Traditional methods use equipment shutoff, thermal energy storage, potential energy storage, and chemical energy storage to reduce on-peak usage or overall electrical energy.

In this study, a gaussian process regression is used to predict electrical industrial load profiles. These load profiles are input to a dynamic battery model that utilizes Bayesian decision theory to determine dispatch and control of the industrial battery installation which aims to charge and discharge the battery to minimize facility electrical costs. A combination of gaussian process regression and Bayesian decision theory has not been applied to industrial battery control. The methods used in this study utilize only knowledge of the electrical load itself and not specific nuances of the process to improve the battery capacity usage and economic viability for industrial users. A year’s worth of electrical load data from an actual industrial facility is utilized to show the control method with an actual facility. The study shows that the utilization of Bayesian decision theory improves the usage of installed battery capacity, helping to facilitate the size optimization of the installed battery.

[1] A. Bracale, G. Carpinelli, P. De Falco, and T. Hong, “Short-term Industrial Load Forecasting : A Case Study in an Italian Factory,” in IEEE PES Innovative Smart Grid Technologies Conference Europe (ISGT Europe), 2017, doi: 10.1109/ISGTEurope.2017.8260176.

[2] S. Ungureanu and A. Cziker, “Industrial load forecasting using machine learning in the context of smart grid,” in International Universities Power Engineering Conference, 2019, doi: 10.1109/UPEC.2019.8893540.

[3] D. J. Leith, M. Heidl, J. V Ringwood, and S. Member, “Gaussian Process Prior Models for Electrical Load Forecasting,” in International Conference on Probabilistic Methods Applied to Power Systems, 2004, no. October 2004, doi: 10.1109/PMAPS.2004.242921.

[4] D. Machalek and K. Powell, “Automated electrical demand peak leveling in a manufacturing facility with short term energy storage for smart grid participation,” J. Manuf. Syst., vol. 52, no. June 2018, pp. 100–109, 2019, doi: 10.1016/j.jmsy.2019.06.001.

[5] G. Karmiris and T. Tengnér, “PEAK SHAVING CONTROL METHOD FOR ENERGY STORAGE,” in Electrical Energy Storage Applications and Technologies (EESAT) Conference, 2013.