Hybrid Deep Learning Gaussian Process for Deterministic and Probabilistic Load Forecasting

Various hybrid load forecasting models have been proposed in recent years, but they generally assign weights to individual forecasting models for optimal combinations and without taking full advantage of the strengths of each model. In this paper, a hybrid Deep Learning Gaussian Process (HDLGP) model for short-term deterministic and probabilistic load forecasting (DLF and PLF) is proposed. This model merges the predictive power of artificial neural networks (ANN) and the ability to handle uncertainty of Gaussian Process (GP) by a composite kernel. Firstly, we design a multi-layer perception (MLP) neural network to learn high fluctuating load data. Then a GP with a composite kernel is incorporated to capture the residuals based on MLP so that further boost accuracy of DLF, meanwhile performing high-quality probability density estimation. Our model guarantees both reliability and sharpness of the PLF. Verifying our proposed model based on the realistically available data, it indicates that our model outperforms the other list approaches both in DLF and PLF.