The data-driven method with strong approximation capabilities and high computational efficiency provides a promising tool for optimal power flow (OPF) calculation with stochastic renewable energy. However, the topology change dramatically increases the learning difficulties and the demand for learning samples. In this work, we propose a physics-guided graph convolution neural network (GCNN) for OPF calculation with consideration of varying topologies, including the physics-guided graph convolution kernel, feature construction, and loss function formulation. Specifically, a physics-embedded graph convolution kernel is derived by aggregating the features from local neighborhoods utilizing the nodal OPF model formulation. An iterative feature construction method is also developed that encodes both the physical feature and practical constraints into the node vector. Finally, a correlative learning loss function to optimize the unbalanced power injection is developed. Extensive numerical results carried out on various IEEE test systems show that the prediction accuracy of OPF using the proposed method under varying topology changes can be improved by an average of 13.30% and up to 32.63% compared with state-of-the-art methods.

Probabilistic analysis tool is important to quantify the impacts of the uncertainties on power system operations. However, the repetitive calculations of power flow are time-consuming. To address this issue, data-driven approaches are proposed but they are not robust to the uncertain injections and varying topology. This article proposes a model-driven graph convolution neural network (MD-GCN) for power flow calculation with high-computational efficiency and good robustness to topology changes. Compared with the basic graph convolution neural network (GCN), the construction of MD-GCN considers the physical connection relationships among different nodes. This is achieved by embedding the linearized power flow model into the layer-wise propagation. Such a structure enhances the interpretability of the network forward propagation. To ensure that enough features are extracted in MD-GCN, a new input feature construction method with multiple neighborhood aggregations and a global pooling layer are developed. This allows us to integrate both global features and neighborhood features, yielding the complete features representation of the system-wide impacts on every single node. Numerical results on the IEEE 30-bus, 57-bus, 118-bus, and 1354-bus systems demonstrate that the proposed method achieves much better performance as compared to other approaches in the presence of uncertain power injections and system topology.

Data-driven optimal power flow (OPF) approach has been a research focus in recent years. However, the current data-driven OPF approaches face the following difficulties: 1) the data-driven solutions may have large deviations and are not trustable, facing out-of-distribution (OOD) samples and 2) it is hard to judge whether the solutions of the data-driven approach can be trusted. To handle these problems, this article first improves the generalization ability of the data-driven OPF method by embedding the inherent pattern of the OPF solution into the data-driven learning process. As an optimization problem, the OPF solution has certain fixed patterns that are not influenced by the distribution of samples. For example, the load balance constraints should always be satisfied. This leads to an inherent requirement of output vectors, which can be utilized to guide the learning process of the data-driven OPF method. Second, an adaptability judging method based on the decoder neural network is proposed to determine whether the data-driven OPF approach can produce trustable solutions. By measuring the decoding error from latent features to input features, the adaptability of neural networks for input samples could be accurately judged. According to extensive results on various systems, the proposed method can improve the calculation accuracy of OOD data by an average of 30.19% compared with state-of-the-art methods. With the adaptability judgment method, the accuracy of the data-driven approach can achieve higher than 98% for OOD data, whereas the accuracy of other methods ranges from 34.08% to 94.50% on the same set of OOD test data.

Ensuring cybersecurity in power monitoring systems is of paramount importance to maintain the operational safety and stability of modern power grids. With the rapid expansion of grid infrastructure and increasing sophistication of cyber threats, existing manual alarm-tracing methods face significant challenges in handling the massive volume of security alerts, leading to delayed responses and potential system vulnerabilities. Current approaches often lack the capability to effectively model complex relationships among alerts and are hindered by imbalanced data distributions, which degrade tracing accuracy. To this end, this paper proposes a power monitor system cybersecurity alarm-tracing method based on the knowledge graph (KG) and graph convolutional neural networks (GCNN). Specifically, a cybersecurity KG is constituted based on the historical alert, accurately representing the entities and relationships in massive alerts. Then, a GCNN with attention mechanisms is applied to sufficiently extract the topological features along alarms in KG so that it can precisely and effectively trace the massive alarms. Most importantly, to mitigate the influence of imbalanced alarms for tracing, a specialized data process and model ensemble strategy by adaptively weighted imbalance sample is proposed. Finally, based on 70,000 alarm information from a regional power grid, by applying the method proposed in this paper, an alarm traceability accuracy rate of 96.59% was achieved. Moreover, compared with the traditional manual method, the traceability efficiency was improved by more than 80%.

Data-driven optimal power flow (OPF) leverages neural networks (NNs) to learn the OPF maps from samples. However, the inherent intricate nonlinearity in the OPF problem poses significant challenges to learning different samples. It often results in unacceptable errors in highly nonlinear regions, such as localized areas with identical active constraints and global boundaries where active constraints vary. Therefore, this paper proposes a nonlinearity-focused sampling method to enrich the OPF samples with high nonlinearity. The proposed method forces the NNs to pay more attention to those nonlinear regions and further improves their robustness. Firstly, the local nonlinearity-focused sampling technique is established based on physics-informed adversarial searching. By merging the gradient of physics models and NNs, it can precisely search the local samples with intolerable errors caused by high nonlinearity. Additionally, a global nonlinearity-focused sampling algorithm is proposed to precisely locate the boundary where active constraints vary, representing an entirely different OPF solution pattern. An iterative strategy is presented to acquire the global samples exhibiting high nonlinearity and learning difficulty. Consequently, the constructed dataset could accurately capture the inherent intricate nonlinearity in OPF and emphasize what NN needs to learn. Extensive simulations on various systems demonstrate that the NNs trained using the OPF datasets constructed by the proposed method can achieve more robust performance.