Integrating Machine Learning and Weather Analytics for Sizing Variable Generation with Utility-Scale Energy Storage

Date

2021-05

Authors

Sun, Fei

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

To reverse climate change, both manufacturing and power sectors are undergoing a paradigm change by integrating renewable energy for sustainability operations. The goal of this study is to model and design a cost-effective, eco-friendly microgrid system to meet the uncertain load of large industrial and commercial users. The microgrid consists of wind turbines, solar photovoltaics, utility-scale hybrid energy storage system (HESS), and feed-in tariff program. HESS comprises the battery and supercapacitor made by lithium-ion and graphene materials, respectively. First, hybrid forecasting models combining multi-layer neural network and statistical inference are developed to predict the wind speed and weather states. The proposed models are implemented in six US cities with diverse climate profiles. The results show that proposed models outperform time series models in 3-to-24 hours ahead of wind speed forecasting by reducing 20 percent error. The weather state model shows yearly forecasting outperforms season-based prediction. A stochastic optimization program is further proposed to minimize the levelized cost of energy based on estimated power capacity factor. Finally, virtual power plant system accommodating both electricity and thermal generation is proposed to minimize the operation cost of a three-tier supply chain network. Various uncertainties are considered, including random power demand, time-of-use rate, government incentives, and the loss of load probability. Through sensitivity analysis, it is found the optimal sizing of renewable generators and lithium-ion battery is not only correlated with the climate conditions, but also depends on time-of-use rate and reliability criteria. Design of experiments is applied to investigate the capacity fade degree with three levels of factors: state of charge, the porosity of positive electrode, and the particle radius size of positive active electrode material. Simulation results show lithium-ion battery yields a better performance when the positive active electrode material has a smaller size, while the porosity of positive electrode and state of charge are at a high level. In conclusion, Lithium-ion battery and graphene-based supercapacitor are the promising technology due to their declining cost and improving performance. The study shows that the distributed generation with utility-scale HESS is cost-effective in the long term and enhances power resilience against extreme weather.

Description

Keywords

lithium-ion battery, graphene-based supercapacitor, hybrid energy storage system, multi-layer neural network, levelized cost of energy, virtual power plant

Citation

Sun, F. (2021). Integrating machine learning and weather analytics for sizing variable generation with utility-scale energy storage (Unpublished dissertation). Texas State University, San Marcos, Texas.

Rights

Rights Holder

Rights License

Rights URI