基于深度自适应动态规划的孤岛主动配电网发电控制与优化一体化算法
Deep adaptive dynamic programming based integration algorithm for generation control and optimization of islanded active distribution network

随着多种分布式新能源的并网, 如风电与光伏发电、生物质能发电、储能与电动汽车等, 传统情况下孤岛 配电网的发电控制方法已很难达到高品质频率稳定控制的要求. 为解决此问题, 本文提出了一种新颖的深度自适 应动态规划算法. 该算法将自适应动态规划算法中的神经网络替换为机器学习领域中的深度神经网络, 并在其中 添加深度模型预测网络. 所提算法能一次性完成传统模式下“发电控制算法+指令优化分配算法”共同完成的工 作. 最后, 为验证深度自适应动态规划算法的鲁棒性, 设计了多种配电网的仿真实验, 即正常情况、“即插即用”启 停机情况、通讯故障情况、全天扰动仿真情况、变拓扑结构的孤岛配网情况和变参数模型的仿真, 设置的总仿真时 长达25年. 仿真结果验证了所设计的深度自适应动态规划算法有效性、可行性与强鲁棒性.
With the development of many distributed generations (DGs) (e.g. wind power and photovoltaic power, biomass energy, energy storage device and plug in hybrid electric vehicle (PHEV)), the traditional methods for generation control of isolated island power grids cannot meet the requirements of frequency stability. This paper proposes deep adaptive dynamic programming (DADP) algorithm to solve this problem. Replacing neural network (NN) in adaptive dynamic programming (ADP) algorithm by the deep neural network (DNN) in the field of machine learning (ML), and adding deep model forecast neural network in, the proposed DADP algorithm is designed. Generation commands, which are achieved by the algorithms of both generation control and generation command dispatch in the traditional way, are obtained by the proposed DADP algorithm. Finally, to verify the robustness of the proposed DADP algorithm, many simulations for isolated island micro-gird are simulated, for instance, the normal isolated island situation, plug and play situation, communication failure situation, all-day disturbance situation, time-varying topology situation and systemic internal parameters varying situation, and the duration of all these simulations is configured as long as 25 years. The effectiveness, feasibility and strong robustness of the proposed DADP algorithm are verified by the simulation results.