分布式电源和电动汽车大量接入配电网，打破了配电网的三相平衡状态，致使传统的配电网保护方案无法适应时代的发展需求。为此，在现有平衡系统故障恢复策略的基础上，提出了一种基于电动汽车充电站的不平衡配电网灾后两阶段故障恢复策略。第一阶段以故障恢复成本最小为目标确定故障后的孤岛运行方案。第二阶段综合考虑电动汽车充电站及各分布式电源机组的折旧、运维、弃负荷惩罚等成本，以配电网运行收益最大为目标建立协同优化模型，以确定故障时各充电站出力情况。最后通过改进的IEEE 33节点系统验证所提两阶段故障自愈模型及优化运行策略的有效性。结果表明，所提方法较传统的孤岛划分法弃负荷率更低，为解决不平衡配电网的故障恢复问题提供了新思路。

 The three-phase equilibrium state of a distribution network is broken owing to the large number of distributed power sources and electric vehicles connected to the distribution network, resulting in the traditional distribution network protection scheme illegally adapting to current development needs. Therefore, this study presents a two-stage fault recovery method for imbalanced distribution networks based on electric vehicle charging stations using an existing balancing system fault recovery technique. The island operation strategy after the fault is determined in the first stage to reduce the fault-recovery cost. In the second stage, a collaborative optimization model is established to determine the output of each charging station when a fault occurs, with the goal of maximizing the operating income of the distribution network, considering the cost of depreciation, operation and maintenance, and the load abandonment penalty of electric vehicle charging stations and distributed generation units. Finally, the enhanced IEEE 33 node system validates the effectiveness of the two-stage fault self-healing model and the optimal operating method suggested in this study. The results reveal that the proposed technique has a lower load rejection rate than the classic island division method, thereby providing a novel solution to the fault-recovery problem of an imbalanced distribution network.

考虑分时电价和充电利用率特征对电动汽车充电站负荷的影响，提出了融合长短记忆网络和支持向量回归（long short-term memory-support vector regression， LSTM-SVR）的大型电动汽车充电站负荷短期预测方法。首先，建立了分时电价、充电利用率、气象信息等影响充电负荷的因素以及历史充电负荷功率数据作为输入的特征矩阵。其次，运用自适应噪声完备经验模态分解（complete ensemble empirical mode decomposition with adaptive noise，CEEMDAN）方法将包含分时电价和充电利用率的特征矩阵序列进行分解，扩充了数据多样性，并采用组合相关系数方法实现了数据降维和特征选择。然后采用北方苍鹰优化（northern goshawk optimization， NGO）算法分别优化LSTM和SVR的超参数，求解权重系数并构建融合LSTM-SVR模型。最后采用某城市一座大型充电站数据进行验证，对比实验表明，考虑分时电价和充电利用率特征可有效提高电动汽车充电站负荷预测精度8%以上，同时采用所提出的融合LSTM-SVR预测方法能使预测精度进一步提高。

Considering the impact of time-of-use electricity pricing and charging utilization characteristics on the load of electric vehicle charging stations, a short-term load forecasting method is proposed for large scale electric vehicle charging stations that integrates long short-term memory and support vector regression (LSTM-SVR). Firstly, a feature matrix is constructed that includes factors influencing the charging load such as time-of-use electricity pricing, charging utilization rates, meteorological information, and historical data on charging load power. Secondly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method is applied to decompose sequences of the feature matrix containing time-of-use electricity pricing and charging utilization rates, which enhances data diversity. This process also employs a combined correlation coefficient method for data reduction and feature selection. Subsequently, the northern goshawk optimization (NGO) algorithm is used to optimize the hyperparameters of both LSTM and SVR, solving for weight coefficients and constructing the integrated LSTM-SVR model. Finally, the model is validated using data from a large scale charging station in a specific city. Comparative experiments demonstrate that considering time-of-use electricity pricing and charging utilization features can effectively improve the accuracy of electric vehicle charging station load forecasting by over 8%. Moreover, the proposed LSTM-SVR forecasting method further enhances prediction accuracy.

针对大规模电动汽车无序接入电网引发的用户充电开销大和电网负荷波动加剧等问题，提出了基于深度强化学习（deep reinforcement learning，DRL）的电动汽车充电行为优化方法。首先，以最小化电网负荷波动和用户充电费用为目标，建立了电动汽车有序充电优化调度模型。其次，将电动汽车的充电行为构建为马尔科夫决策过程（Markov decision process，MDP），根据电网负荷预测信息和分时电价对充电时段进行优先级评定，并根据优先级控制电动汽车充电行为。通过双延迟深度确定性策略梯度（twin delayed deep deterministic policy gradient，TD3）算法对电动汽车有序充电策略进行快速优化。最后，通过算例验证了所提方法在减少用户的充电开销和配电网的负荷波动方面的有效性。

The large-scale disordered integration of electric vehicles （EVs） into power grid poses many problems, such as enlarged power fluctuation for grid and increased charging cost for users. To solve these problems, this paper proposes an optimized EV charging method based on deep reinforcement learning （DRL）. Firstly, an EV ordered charging scheduling model is established aiming at minimizing the power fluctuation and user charging cost. Secondly, the EV charging behaviour is formulated as a Markov decision process （MDP） which evaluates the priority for each charging period based on load prediction information and time-of-use electricity tariff. The charging behaviour of EVs is controlled by the priority. The twin delayed deep deterministic policy gradient （TD3） algorithm is adopted to solve the MDP problem, which quickly optimizes EV ordered charging strategy. Finally, the effectivenesses of the proposed method in reducing charging cost and load fluctuation of distribution network are verified by case studies based on numerical examples.

This paper develops a model-free volt-VAR optimization (VVO) algorithm via multi-agent deep reinforcement learning (DRL) in unbalanced distribution systems. This method is novel since we cast the VVO problem in distribution networks to an intelligent deep Q-network (DQN) framework, which avoids solving a specific optimization model directly when facing time-varying operating conditions in the systems. We consider statuses/ratios of switchable capacitors, voltage regulators, and smart inverters installed at distributed generators as the action variables of the agents. A delicately designed reward function guides these agents to interact with the distribution system, in the direction of reinforcing voltage regulation and power loss reduction simultaneously. The forward-backward sweep method for radial three-phase distribution systems provides accurate power flow results within a few iterations to the DRL environment. The proposed method realizes the dual goals for VVO. We test this algorithm on the unbalanced IEEE 13-bus and 123-bus systems. Numerical simulations validate the excellent performance of this method in voltage regulation and power loss reduction.