Existing power anomaly detection is mainly based on a pattern matching algorithm. However, this method requires a lot of manual work, is time-consuming, and cannot detect unknown anomalies. Moreover, a large amount of labeled anomaly data is required in machine learning-based anomaly detection. Therefore, this paper proposes the application of a generative adversarial network (GAN) to massive data stream anomaly identification, diagnosis, and prediction in power dispatching automation systems. Firstly, to address the problem of the small amount of anomaly data, a GAN is used to obtain reliable labeled datasets for fault diagnosis model training based on a few labeled data points. Then, a two-step detection process is designed for the characteristics of grid anomalies, where the generated samples are first input to the XGBoost recognition system to identify the large class of anomalies in the first step. Thereafter, the data processed in the first step are input to the joint model of Convolutional Neural Networks (CNN) and Long short-term memory (LSTM) for fine-grained analysis to detect the small class of anomalies in the second step. Extensive experiments show that our work can reduce a lot of manual work and outperform the state-of-art anomalies classification algorithms for power dispatching data network.

Household loads are becoming dominant in virtual power plants (VPP). However, their dispatch potential has not yet been explored due to the lack of detailed user power management. To solve this issue, a novel two-layer user energy management strategy based on HG-multi-agent reinforcement learning has been proposed in this paper. Firstly, a novel two-layer optimization framework is established, where the upper layer is applied to coordinate the scheduling and benefit allocation among various stakeholders and the lower layer is applied to execute intelligent decision-making for users. Secondly, the mathematical model for the framework is established, where a detailed household power management model is proposed in the lower layer, and the generated predicted power demands are used to replace the conventional aggregate model in the upper layer. As a result, the energy consumption behaviors of household users can be precisely described in the scheduling scheme. Furthermore, an HG-multi-agent reinforcement-based method is applied to accelerate the game-solving process. Case study results indicate that the proposed method leads to a reduction in user costs and an increase in VPP profit.

Constrained by current market mechanisms, small-scale virtual power plants (SS-VPPs) on the distribution network side struggle to exert their market characteristics. To address this, this paper proposes a trading framework and operational strategy for distribution-side SS-VPPs to participate in the day-ahead energy market. First, an operation and trading framework for distribution networks involving SS-VPPs is proposed. This framework comprehensively considers the clearing process of the electricity energy market, the operation mechanism of the distribution network, and the cost structures of various stakeholders, while clarifying the day-ahead market clearing mechanism at the distribution network level. Next, accounting for energy balance constraints and distribution network congestion constraints, this paper establishes a collaborative optimization model between SS-VPPs and active distribution networks. After obtaining the energy optimization results for all stakeholders, distribution locational marginal pricing (DLMP) is determined based on the dual problem solution to achieve multi-stakeholder market clearing. Finally, simulations using a modified IEEE 33-node test system demonstrate the rationality and feasibility of the proposed strategy. The framework fully exploits the market characteristics and dispatch potential of SS-VPPs, significantly reduces overall system operating costs, and ensures the economic benefits of all participants.