Predictive Analysis in Electric Distribution Grid: Case Studies on Solar and Energy Price Forecast

Abstract

The electricity distribution system is undergoing profound changes as the society moves towards more sustainable utilization of energy resources. A common challenge in both supply and demand sides is how to provide accurate near term (within a day) forecast of the uncertainties to enable the distribution grid operation to modernize their decision making and deliver clean, affordable, and reliable electricity services.

This thesis focuses on the common challenge mentioned above, namely, how to improve the predictive capability for distribution system operators and load serving entities (LSEs). In particular, this thesis focuses on two of the major uncertain variables in future distribution grid: solar and electricity price forecast. Series of data-driven analysis are applied to develop efficient prediction models of these two variables. For the solar power generation prediction, the spatial temporal autoregressive model (ST ARX) is applied to the distribution system by including the neighboring data at nearby locations. Comparing to the benchmark models, the proposed model results in a better prediction accuracy and indicates the strong correlation between optimal neighboring distance and prediction time scale. As for the electricity price prediction, a comprehensive classification model based on decision tree algorithm is developed for the EnergyCoupon system. This algorithm is tested in Houston area with 10 customers and results in a good accuracy.