Adaptive deep dynamic programming for integrated frequency control of multi-area multi-microgrid systems

Abstract

To reduce the frequency deviation of a multi-area multi-microgrid system, a framework of integrated frequency control is designed in this paper, which can replace load frequency control (LFC) and generation command dispatch (GCD).Then an adaptive deep dynamic programming (ADDP) scheme is proposed for the integrated frequency control. The ADDP contains three deep neural networks, i.e., deep prediction neural network, deep critic neural network and deep action neural network. Deep prediction neural network is applied to predict the next state of the multi-area multi-microgrid system from the previous states and the previous actions. Deep critic neural network is employed in the evaluations of the performance of the deep action neural network. Deep action neural network is introduced to simultaneously provide generation commands for all the LFC units in the multi-area multi-microgrid system. The ADDP is compared with other 157 algorithms under six case studies, i.e., basic situation, plug-and-play, communication failure, all-day long disturbance, time-varying topology and parameters varying. The other 157 algorithms consist of adaptive dynamic programming and 156 combined algorithms, which combined with 12 control algorithms for the controller of LFC and 13 optimization algorithms for GCD. Simulation results verify the effectiveness and superiority of the ADDP for integrated frequency control of a multi-area multi-microgrid system.