A Hybrid Data-Driven Method with High-Order Taylor Polynomials to Forecast DER Penetration

Forecasting the penetration of distributed energy resources (DERs) is a challenging problem because DER historic data is scarce which causes overfitting in forecast models. To address this problem, this paper proposes a hybrid DER forecast method that uses a relatively small number of parameters and that blends the strengths of: (i) exponential smoothing to decompose dataset into level and seasonal series, high-order Taylor polynomial to predict level trends, and long short-term memory to forecast seasonal behavior. The forecast rules account for the impact of external stressors such as electricity prices and technology costs. We contrast the performance of the proposed approach against other machine-techniques by using publicly available datasets.