A Data-Driven Method for Probabilistic Prediction of Power Grid Critical Section Power from Probability Distributions of Renewable Energy

Abstract

The probabilistic prediction of power grid critical section power is essential to the operational analysis and control of high penetration renewable energy power system. However, the heavy computational burden of Monte Carlo Simulation (MCS) methods hardly meets the demand of practical application, the linearization assumption of the analytical methods bring inherent error in scenarios with high-dimensional input random variables. This article proposes a data-driven method for the complex nonlinear mapping from probability distributions of renewable energy to those of critical section power (called distributions-in-distributions-out) with higher computational efficiency and accuracy. Compared with the existing data-driven probabilistic power flow methods where machine learning is used for power flow solution, a larger number of probability distributions are considered in the proposed method by constructing probability distributions as features. The proposed graph convolutional network (GCN) framework is proved better than long short-term memory (LSTM) network, deep belief network (DBN) and multi-layer perceptron (MLP) in terms of adaptability to topology changes. Comparison results on the IEEE 39-bus and 118-bus systems indicate that the proposed method achieves better accuracy and robustness under different scenarios.

Keywords: critical section power, graph convolution network (GCN), probabilistic power flow, probabilistic prediction