A Two-Stage Deep Learning Approach for Solving Microgrid Economic Dispatch

The intermittency of renewable generation and uncertainties of electricity demand motivates the real-time economic dispatch (ED) of assets in microgrids. However, numerical optimization problems are extremely hard to solve in real time. This article proposes a data-driven neural network (NN) approach for solving the ED problem of microgrids. To deal with intermittency of renewable generation, a two-stage training approach is proposed to better learn the spatio-temporal characteristics of renewable generation and conventional generation. In addition, to improve the learning process and increase the accuracy of the proposed NN framework, a short-time Fourier transform is utilized as a preprocessor and denoiser. Detailed comparison with conventional numerical optimization approaches validate the effectiveness of the proposed data-driven approach for optimally allocating microgrid resources in real time.