Short-Term Probabilistic Forecasting of Renewable Energy Generation with Direct Multistep-Ahead Strategy and Recurrent Neural Network

Abstract

With the rapid development of renewable energy generation, probabilistic forecasting has attracted more attention compared to deterministic forecasting. This paper focuses on generating short-term probabilistic forecasting of renewable energy power with quantiles chosen as uncertainty representation. First, in order to avoid the crossing-quantile problem, some constraints associated with quantile-increment series, which are obtained by reformulating the quantile series, are proposed. Then, recurrent neural network is adopted to depict the complex nonlinear relationship between predictors and quantiles, and a reasonable decoder structure is designed to obtain multistep-ahead quantiles prediction directly. Numerical results on a real-world solar power dataset verify the effectiveness of our proposed model, which is capable of providing the high-quality quantiles with less time compared with some advanced benchmarks.