The increasing ratio of renewable energy sources and the need for fast demand response of controllable power equipment have led to increasing complexity in smart distribution network control. In this paper, we propose a robust control scheme for smart distribution networks based on a multi-agent deep reinforcement learning algorithm for the smart grid voltage stability control problem. This design is deployed in two aspects. In terms of interaction environment design, considering that capacitors, regulators and distributed energy connected to a smart distribution grid have both discrete and continuous actions, the action space for interacting with agents is set to be a multidimensional discrete action space. In terms of algorithm design an intelligent active and reactive power compensation method based on multi-discrete soft actor critic (MDSAC) algorithm is proposed to realize voltage stability control. The designed MDSAC-agent is trained offline and tested online to achieve real-time smart dynamic voltage stability for smart distribution network. The proposed voltage control method is validated using the modified IEEE-13 and IEEE-34 bus systems containing controllable capacitor reactive power compensator, on-load tap changer(OLTC) and renewable energy sources. The simulation results show that the proposed method has better higher reward values and lesser voltage deviation than the other method.

More and more distributed generation (DG) and energy storage (ES) devices are being connected to the distribution network (DN). They have the potential of maintaining a stable supply load during failure periods when using islanding operations. Therefore, DG and ES have capacity value, i.e., improving the power supply capability of the system. However, there are strong fluctuations in DG outputs, and the operations of ES devices have sequential characteristics. The same capacity of DG has different load-bearing capabilities compared to conventional thermal or hydroelectric units. This paper proposes a method for evaluation of power supply capability improvement in DNs. First, the temporal fluctuation in both power source and load demand during fault periods is considered. A DN island partition model considering the secondary power outage constraint is established. Then, a modified genetic algorithm is designed. The complex island partition model is solved to achieve accurate power supply reliability evaluation. And the incremental power supply capability associated to DG and ES devices is calculated. Finally, a case study is conducted on the PG & E 69-bus system to verify the effectiveness of the proposed method. It is found that with a 20% configuration ratio of ES devices, the power supply capability improvement brought about by 6 MW DG can reach about 773 kW.

This paper investigates a distributed control scheme for the coordination of an islanded microgrid considering renewable energy sources (RES). A distributed finite-time secondary frequency controller is designed to compensate the frequency deviation in a finite-time manner. The model of an islanded microgrid considering RES is presented with the connection of graph theory. Comparing with the traditional dynamic model, this paper gives a specific model of dc power source considering RES. Further, the design process of the distributed secondary frequency controller is described in detail, including the upper bounds of the convergence time. Finally, a simulation is provided to verify the effectiveness of our proposed method.

The problem of optimal dispatching of a power system containing a high proportion of renewable energy is of great significance for the realisation of new energy consumption and the economic and reliable operation of the power system. For the solution of non-linear, non-convex, multi-objective problems for the optimal operation design of a power system with wind and photovoltaic access, traditional methods have difficulties in terms of computational real-time and iterative convergence. To address this issue, a deep reinforcement learning-based optimal scheduling method for the hybrid power system is proposed, which enables continuous action control to obtain an optimal scheduling strategy through the interaction between the agent and the hybrid power system. Firstly, a mathematical description of the optimal scheduling problem containing wind power and photovoltaic power system is presented, and the state space, action space, and reward function of the agent are designed. Secondly, the basic framework of the deep reinforcement learning optimal scheduling model is constructed, and the basic principles of the twin delayed deep deterministic policy gradient algorithm are introduced. Finally, the effectiveness of the deep reinforcement learning model for day-ahead optimal scheduling of the hybrid power system is verified by means of an arithmetic analysis of the modified New England 39-bus system.

With the rapid development of modern power systems, the penetration rates of power electronic equipment and renewable energy are increasing, which bring challenges to stable operation. Due to the strong randomness and uncertainty, the conventional method based on mathematical model can not control modern power systems. So in this paper, a deep reinforcement learning (DRL) based method is proposed to realize the stable autonomous control of power systems. Specific, soft actor-critic (SAC) is used to reroute power flow in transmission lines via autonomous topology optimization control. Besides, to solve huge action space in topology switching and the vulnerability of DRL agents in power systems, a pre-trained scheme based on imitation learning (IL) is presented to use in SAC. Simulations on the IEEE 118-bus system for topology optimization are carried out under random perturbations and common adversarial perturbations to validate the effectiveness and robustness of the proposed methods. The results show that our methods have outstanding performance.

In order to improve the prediction accuracy of rooftop photovoltaic (PV) power, a PV power prediction method based on data reconstruction and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is proposed, taking into account the lack of rooftop PV power. Firstly, the power relationship between rooftop PV and adjacent centralized PV is established based on daily power generation to reconstruct rooftop PV power. Subsequently, considering the difference in power generation performance between rooftop PV and adjacent centralized PV, the relative generation efficiency is defined to correct reconstructed value. Finally, the rooftop PV power prediction model is established based on CEEMDAN and long short-term memory network (LSTM). The operation data acquired from PV power station in northern China are considered as an example to verify the effectiveness of the proposed method. The experimental results show that the proposed method has a high prediction accuracy under different weather types.