Forecasting Of Wind, PV Generation, And Market Price For The Optimal Operations Of The Regional PEV Charging Stations

Abstract

The transition from the conventional spark ignition engine vehicles to the electric vehicular technologies helps reduce greenhouse gas (GHG) emissions as well as improve the energy efficiency in the transportation sector. In the transformation of the electric vehicle, the hybrid electric vehicle (HEV) has evolved into the plug-in electric vehicle (PEV) due to the advancement in battery technologies that extend the electric driving distance of vehicles; however, this trend also creates concern among PEV users about how long or how far they might travel per battery charge. A well-planned charging infrastructure with a fast (level 3) charging station is critical to overcome the range anxiety of PEV users, which can then promote the deployment and public acceptance of PEV. In addition, the PEV charging station must be considered from a regional point of view, especially in terms of operation optimization and support for the high penetration of PEVs in metro areas. Integrating renewable energy sources such as wind and solar PV power generation with electricity from the grid into PEV charging stations is critical for sustainable future development. A PEV charging station with a distributed energy storage system will be able to participate in the deregulated market to support the power system and optimize its operational cost. However, sufficient accuracy in the forecasting of energy sources and market prices are prerequisite to achieving the above mentioned benefits and goals.Using the Dallas/Fort Worth (DFW) as an example, this dissertation develops novel approaches for the wind/PV generation and market price predictions. These predictions are calculated every 15 minutes (15-minute ahead prediction) for the following 15-minute settlement interval set by the Electric Reliability Council of Texas (ERCOT) market. Support Vector Classification (SVC) and Support Vector Regression (SVR) of Support Vector Machines (SVMs) are adopted for the prediction of categorical and continuous values, respectively. SVR is used to predict the wind/PV generation because they are considered continuous functions. The validations of the estimation performance for these two predictions are illustrated using the wind power data from a wind farm in Oklahoma (a virtual wind farm for this study) and the PV generation from Dallas Redbird airport, respectively. The proposed method improves the forecasting performance of both predictions compared to the persistence model. In addition to attaining accurate market price predictions in the deregulated market, a hybrid market price forecasting method (HMPFM) including SVC and SVR with data clustering techniques is proposed. SVC is adopted to predict spike price occurrence, and SVR is used for market price magnitude prediction of both non-spike and spike prices. Additionally, three clustering techniques including Classification and Regression Trees, K-means, and Stratification methods are introduced to mitigate the higher error of spike magnitude estimation. The performance of the proposed hybrid method is validated with the ERCOT wholesale market price. The results from the proposed method show significant improvement over typical approaches. In order to fulfill the comprehensive study, the characteristics of the forecast uncertainty have to be investigated to understand their stochastic nature for optimizing the benefits of operating PEV charging stations. In this dissertation, the Martingale Model Forecast Evolution (MMFE) is used for the investigation, since it explores the multivariate random vector of the forecast change, which can apply to the multivariate case in this problem. Finally, the results show the effectiveness of the MMFE to generate the stochastic nature of the proposed predictions.