Machine learning-based diagnosis methods have achieved remarkable success under the assumption that the training and test data are identically distributed. However, a critical requirement of these methods is the generalization capability to unseen domains when deployed to actual diagnosis scenarios. We introduce the challenging problem of domain generalization, i.e., learning from multiple source domains to produce a model that can directly generalize to unseen domains without target information. We adopt a model-agnostic learning produce that maximizes the dot product of gradients between the source domains. Such a gradient alignment objective encourages finding a common optimization path for all source domains, which helps to focus on invariant representations. Furthermore, we propose two feature regularizations that explicitly regularize the feature space. Global feature regularization aligns class relationships between different domains to preserve the domain-invariant knowledge. Local feature regularization encourages the model to learn domain-agnostic class-specific representations with intraclass compactness and interclass separability. The effectiveness of the proposed method is demonstrated with generalization experiments on two benchmarks.

To meet voltage and capability needs, batteries are grouped into packs as power sources. Abnormal ones in a pack will lead to partial heating and reduced available life, so removing anomalies out during manufacturing is of great significance. The conventional methods to detect abnormal batteries mainly rely on grading systems and manual operations. Current data-driven methods use statistical, machine learning and neural network approaches, building models, then applying them on the unlabeled. However, both cannot make full use of multiple source data and expert knowledge. Therefore, how to use these multi-source data and knowledge to improve the effect of battery anomaly detection process has become a research focus. We put forward a data-driven multi-source data feature fusion and expert knowledge integration (FFEKI) network architecture that follows encoder-decoder structure with multiple integration units and a corresponding joint loss function. First, we collect multi-source data and obtain fusion features. Then, we refine filters from expert knowledge and transform them into neural network layers as components of integration units. By this way, supervisory knowledge is integrated into our network. We evaluate our scheme by sets of experiments comparing with most widely used approaches on real manufacturing data. Results show that FFEKI obtains a maximum 100% anomaly detection rate (ADR). Meanwhile, when the number of detection T is greater than the actual number of anomalies in the testing set, our method can achieve full ADR faster. It is concluded that the proposed FFEKI achieves effective performance on power lithium-ion battery anomaly detection.

Accurate identification of critical nodes and regions in a power grid is a precondition and guarantee for safety assessment and situational awareness. Existing methods have achieved effective static identification based on the inherent topological and electrical characteristics of the grid. However, they ignore the variations of these critical nodes and regions over time and are not appropriate for online monitoring. To solve this problem, a novel data-driven dynamic identification scheme is proposed in this paper. Three temporal and three spatial attributes are extracted from their corresponding voltage phasor sequences and integrated via Gini-coefficient and Spearman correlation coefficient to form node importance and relevance assessment indices. Critical nodes and regions can be identified dynamically through importance ranking and clustering on the basis of these two indices. The validity and applicability of the proposed method pass the test on various situations of the IEEE-39 benchmark system, showing that this method can identify the critical nodes and regions, locate the potential disturbance source accurately, and depict the variation of node/region criticality dynamically.

Objective Online estimation of voltage stability margin (VSM) is critical to the long-term safety and stable operation of power system. In this paper, a deep neural network (DNN)-based model that incorporates the intrinsic grid topology information is proposed to achieve accurate VSM estimation. Methods The node embedding information together with the electrical parameter measurements are combined together into an array of local structures as the input feature vector. A local-global deep neural network (LGDNN) model is proposed to extract and integrate the local information into high-level global representation for VSM estimation. The Node2vec algorithm is employed to obtain embedded node vectors and a NodeRank algorithm is designed to form local structures, which are composed of the measurements of a fixed number of similar nodes according to the similarity among their embedded vectors. A sequential DNN with cascade-connected depthwise separable 1D convolutional layer and fully connected layer is trained to estimate the current VSM. Results The model performance is validated on the IEEE-39 and IEEE-118 benchmark systems under normal and post-contingency situations. Experiment results indicate that the proposed model can achieve VSM estimation with high precision that surpasses seven mainstream approaches in multiple estimation error indices. Conclusion Through the integration of local structures, the proposed LGDNN increases the estimation accuracy with relatively few parameters than various DNN-based models. Thus, it is an practical model for online power grid voltage stability monitoring.

Consistence of lithium-ion power battery significantly affects the life and safety of battery modules and packs. To improve the consistence, battery grouping is employed, assembling batteries with similar electrochemical characteristics to make up modules and packs. Therefore, grouping process boils down to unsupervised clustering problem. Current used grouping approaches include two aspects, static characteristics based and dynamic based. However, there are three problems. First, the common problem is under utilization of multi-source data. Second, for the static characteristics based, there is grouping failure over time. Third, for the dynamic characteristics based, there is high computational complexity. To solve these problems, we propose a distributed multisource data fusion based battery grouping approach. The proposed approach designs an effective network structure for multisource data fusing and feature extracting from both static and dynamic multisource data. We apply our approach on real battery modules and record state of health (SOH) during charging-discharging cycles. Experiments indicate that the proposed approach can increase SOH of modules by 3.89% and reduce the inconsistence by 68.4%. Meanwhile, with the distributed deployment the time cost is reduced by 87.9% than the centralized scheme.