Time Series Generation Adversarial Network Controller Based on Optimal Control of Micro-Grids

The traditional micro-grid co-generation control framework contains two time scales, which may lead to the incoherence problem. In order to avoid the incoherence problem, this paper proposes a smart generation control framework with a unified time scale, based on which a timeseries generative adversarial network controller for optimal control of micro-grids is proposed. The proposed time-series generative adversarial network controller contains reinforcement learning, generator deep neural network and discriminator deep neural network. The generator deep neural network generates predicted states based on multiple historical states, multiple historical actions, and multiple longterm actions. The discriminator deep neural network, on the other hand, determines whether the data from the generator deep neural network or real-life data. Numerical simulation results under an IEEE 300-bus power system validate that the proposed time-series generative adversarial network controller can simultaneously achieve higher control performance and smaller long-term economic cost than conventional control algorithms and optimization algorithms.