The topology of low-voltage distribution networks is complex and variable, and smart meters are mainly deployed on the customer side, resulting in insufficient transparency of the network topology and difficulty in accurately obtaining impedance parameters. Therefore, this paper proposes a topology identification method for low-voltage distribution networks with latent nodes considering impedance parameter constraints. Firstly, an equivalent electrical distance matrix generation method considering impedance parameter constraints is proposed based on the branch linear current model. Then, the impedance information is used to set the connection relationship criterion for judging node pairs, and the topology is generated step by step from the user node to the root node by an iterative recursive method, while obtaining more accurate distribution line impedance parameters. Finally, the effectiveness of the proposed method is verified based on actual data and simulation design of low-voltage distribution network.

Non-Intrusive Load Monitoring (NILM) provides a solution for smart energy management in consumer electronics, enabling the monitoring of internal appliance operating statuses by analyzing the aggregated load data. Recently, various machine learning algorithms have been employed for NILM, however, their effectiveness largely depends on the availability of substantial high-quality labeled data. In practice, the manual and automatic labeling based on event waveform extracted from master meter data inevitably introduce the noisy labels, which impacts the performance of load identification neural network model. To this end, we propose a label noise robust learning approach for non-intrusive load monitoring based on meta-learning. On the basis of traditional teacher-student framework, multiple meta networks are generated by adjusting student network parameters using pseudo labels, which are derived from the noisy labeled data according to neighbor feature similarity. These meta networks further generate predictions, which are compared with the teacher network predictions to constitute the meta loss. The NILM model trained with meta loss can extract valuable information from noisy labeled data and avoid overfitting to incorrect labels. Comparison tests on two public datasets and one private dataset demonstrate that our method provides better performance under different noise rates.

Non-intrusive load monitoring (NILM) technology enables the estimation of both the power consumption and real-time operational states of individual (or target) appliances by analyzing voltage and aggregate current data at the power inlet. The notable performance of modern deep neural networks in energy disaggregation and appliance identification largely stems from the availability of large-scale, appliance-level consumption data with fully accurate annotations. For the widely adopted event-based appliance identification methods, as the construction of customized label data relying on intrusive sub-metering is cost-prohibitive, the aggregate power profiles are often used to acquire this dataset, which tends to result in inaccurate labels. In this work, we propose a contrastive partial label learning method for appliance identification, where each training instance is associated with a coarse candidate label set, which consists of several possible labels including the ground truth. Firstly, a momentum contrastive learning module is employed, which enriches the feature learning space by maintaining a dynamic memory queue of positive and negative samples. This mechanism enables the model to learn more discriminative and consistent representations for the same appliance class, thereby enhancing intra-class compactness and inter-class separability in the embedding space. Secondly, a specific label disambiguation cross-entropy loss function is designed. By dynamically restricting the predicted results only to the candidate label set, it enables the model to progressively increase the predicted probability of the true label during training, thereby effectively facilitating label disambiguation. By integrating the momentum contrastive learning with the label disambiguation cross-entropy loss, the proposed method effectively addresses ambiguously labeled data and achieves comparable appliance identification accuracy in real-world NILM scenarios. Extensive experiments on publicly available NILM datasets validate the effectiveness of the proposed method under the coarse label data.

A critical bottleneck in the design and practical deployment of Organic Rankine Cycle (ORC) systems lies in the poorly understood interplay between working fluid composition, thermophysical properties, heat transfer characteristics, and overall cycle performance. This study addresses this gap by developing a hybrid simulation framework for a zeotropic ORC utilizing R245fa/R134a as the working fluid, in which the pump and heat exchangers are described by semi-empirical and component-level mechanistic models, while artificial neural network (ANN)-based modules are employed for pressure prediction and expander modeling. The framework is rigorously validated against experimental data, with deviations in key performance indicators consistently below 5%, confirming its fidelity and predictive capability. Leveraging this validated framework, the coupled effects of composition, mass flow rate, and heat source/heat sink boundary conditions on cycle performance are systematically investigated. A key mechanistic insight reveals that reducing working fluid viscosity significantly enhances heat transfer capability, directly contributing to improved system efficiency. Building on this understanding, a multivariable performance optimization is conducted, yielding predictive correlations for optimal shaft power output and thermal efficiency under specified operating constraints. Subsequent parametric analysis identifies distinct optima: the maximum net power output (NPO) occurs at an R245fa mass fraction of 0.56, while peak thermal efficiency is achieved at 0.63. These findings not only quantify the trade-offs inherent in mixture selection but also provide actionable, optimization-oriented guidance for tailoring working fluid composition to specific performance objectives.

Non-intrusive load monitoring (NILM) disaggregates household aggregate power into per-appliance profiles without dedicated sub-meters. Conventional supervised approaches require dense frame-level power annotations obtainable only through per-device sub-metering, which is expensive and impractical at scale. In contrast, two cheaper forms of weak supervision are widely accessible: inexact supervision in the form of window-level activity labels, and incomplete supervision in the form of limited binary on/off state labels. We proposed the State-Anchored Hard-Decoupled Architecture (SA-HDA), a three-stage framework that explicitly separates appliance state detection from power amplitude estimation, assigning each sub-task to its appropriate supervision signal. A CRNN trained under the multiple-instance learning framework handles state detection from inexact and incomplete labels; a frozen large language model (LLM) with a compact adapter handles amplitude calibration from scarce power labels; a physics-constrained hard gate anchors LLM outputs to CRNN state decisions at inference. Experiments on UK-DALE across five appliances demonstrated consistent reductions in MAE and SAE over a state-only baseline, with F1 preserved by construction.

Distribution Network Reconfiguration (DNR) exploits active grid regulation potential and minimizes active power losses by altering topologies under time-varying conditions. However, lacking an underlying safety masking mechanism, existing deep reinforcement learning (DRL) methods often conduct blind exploration in invalid topological spaces, severely limiting the agent's learning efficiency and action feasibility. To this end, a DRL framework integrating bridge-aware safe masking and Soft Actor-Critic (SAC) is proposed to autonomously optimize DDNR strategies. First, a bridge-aware safe masking mechanism derived from graph theory is introduced to pre-filter illegal switching actions, overcoming the inefficiency of soft penalties and strictly guaranteeing the radiality and connectivity constraints. Then, the graph attention network is designed as an encoder to dynamically extract spatiotemporal electrical features, improving the agent's perception capability of the distribution network. Based on this, a pointer attention scorer is introduced as the policy decoding layer to establish the dynamic mappings between global graph representations and discrete candidate switches, determining the optimal reconfiguration action sequence. Experimental results demonstrate that the proposed method achieves lower active power losses with fewer switching operations while ensuring topological feasibility, proving its effectiveness and practicality.

Well-annotated load datasets are a crucial prerequisite for construction of non-intrusive load monitoring (NILM) models, especially those based on deep learning. However, it is challenging to create large-scale labeled datasets in the real world due to the labor and time cost, as well as user privacy concern, which limits the development of well generalizing deep learning-based NILM models. To address this issue, we propose a hybrid mechanism-data driven load profile generator for NILM, called LP-generator. It generates load profiles by combining a knowledge library of appliance usage behavior with a model library of appliance power consumption. The knowledge library of appliance usage behavior is represented by the temporal distribution of the on/off states of appliances, including various statistical variables such as the frequency of use, starting time, activation probability of appliance usage correlations. The model library of appliance power consumption is built for generating the load profiles of appliance operation. It generates transient and steady-state power segments based on simulation strategies that are defined by the characteristics of appliance power consumption. Furthermore, it simulates the transition processes between different power segments according to the hidden semi-Markov model (HSMM). Both qualitative and quantitative evaluations demonstrate that LP-generator can generate high-quality synthetic labeled load profiles that imitate real-world scenarios, thus it can serve as a data augmentation tool in supporting the development of NILM model.

Power distribution networks, directly connected to end-users, have stringent reliability requirements. Singlephase grounding faults are the most common in distribution systems, disrupting power supply and potentially damaging equipment, leading to safety incidents or large-scale accidents. Timely fault detection is therefore crucial. In this paper, the improved Prony analysis is employed to decompose the negative sequence current of faults on the low-voltage side, fitting equidistant sampled current data with a linear combination of exponential functions. This allows for the derivation of amplitude, phase, and other relevant signal information, facilitating the analysis of fault characteristic patterns and feature comparisons. By utilising electrical data from multiple measurement points on the low-voltage side, accurate fault line selection on the medium-voltage side is achieved based on the amplitude features of the fault's negative sequence current. Simulation experiments on various scenarios confirmed the effectiveness of our method, which enhances fault characteristic salience, enabling precise fault line identification on the medium-voltage side.

The low-voltage (LV) distribution network is directly connected to power users, thus its operational reliability significantly influences user satisfaction. In distribution network operation, issues such as line aging and equipment anomalies may arise. Due to the extensive coverage of distribution lines, it is challenging to locate these faults timely relying on manual inspections. The widespread adoption of advanced metering infrastructure technologies has enabled data-driven approaches to identify such issues. Considering that the loop impedance can help assess the health status of power lines, this paper proposes a method for evaluating loop impedance based on smart meter data to monitor the condition of LV power lines. It establishes an equivalent model of system loop impedance using voltage and current data collected by smart meters and data concentrators. By selecting data from periods with significant current variations, the method reduces the deviation range in impedance calculation results. Furthermore, impedance results from multiple time periods are processed using a normal distribution fitting-based estimation approach, enabling effective loop impedance assessment.

Nowadays, distributed photovoltaic (PV) systems are globally and rapidly developed. However, the existing netload metering policies cannot identify the distributed PV systems installed behind the meter or unregistered PV capacity increment. In light of the drawbacks of existing PV generation estimation approaches, we propose a PV generation disaggregation method in this paper. Firstly, load ON/OFF states are identified by a hybrid of convolutional neural networks and bi-directional long short-term memory. After capturing the loads’ rate power from the period of data where the PV system is on standby, the load power signals are generated. Finally, the PV generation is disaggregated by removing all estimated loads from the netload, with post-processing steps for result refinement. Through validation on a dataset collected from a real-world mineral processing enterprise, our proposed method outperforms a direct PV generation disaggregation benchmark. Moreover, a case study is conducted to investigate the influence of the ratio between PV capacity and load volume.

Non-intrusive load monitoring (NILM) is a technique that disaggregates household total power consumption into the usage of individual appliances, which can promote the application of home energy management through detailed monitoring of electricity usage behavior. The rapid development of deep learning has driven significant advancements in the theory and algorithms of NILM. However, most deep learning models are fed with data from multiple households during the training process, and they suffer from performance decay due to the distribution discrepancy between the source domains and target domain. To address this issue, this paper proposes a novel data distillation-based multi-source domain adaptation mechanism that not only considers the different distribution distances between multiple source domains and the target domain but also investigates the varying similarities between source domain samples and target domain samples. Concretely, the proposed mechanism consists of four stages: a) pre-train the feature extractor and energy regressor using labeled data from the source domain; b) adversarially train the feature extractor that adaptive to the target domain by minimizing the empirical Wasserstein distance between the source and target domain; c) select the source domain samples that are closer to the target domain to fine-tune the energy regressor; d) aggregate power estimations from each source domain to the target domain based on each domain weight. Using data from five households in the public REFIT dataset, multiple comparative experiments validate the effectiveness of the proposed algorithm. Extensive experiments, including feature visualization and ablation studies, demonstrate that the proposed algorithm can significantly enhance the domain adaptation capabilities of the model.

Nonintrusive load monitoring (NILM) is an economical technology for promoting demand-side management (DSM) by offering electricity usage details. Driven by the fact that most existing NILM works do not support either real-time response or plug-and-play, this article proposes a real-time unsupervised NILM method based on novel graph signal processing (GSP) concepts. Instead of traditional features such as power variation and steady-state sequence, more informative state transition sequences (STSs) are featured. STSs of various types of loads are adaptively extracted based on specifically defined anchor points. After each falling STS (FSTS) referring to load switching-off event is paired with the optimal rising STS (RSTS) in milliseconds, such STS pairs are labeled as identified loads based on an undirected graph with introducing a penalty term in graph total variation regularization. In addition, the scaling factor selection and negative samples are designed to enhance NILM performance. Validation on three publicly available datasets shows the proposed method generally outperforms seven state-of-the-art benchmarks in three NILM metrics and real-time performance.

Power equipment malfunctions pose a threat to the safe operation of the entire power grid, making the rapid and accurate identification of power equipment faults a key focus of research in the electric power field. Existing methods based on artificial intelligence and expert systems face issues such as strong dependence on sample data and difficulties in acquiring expert knowledge. In response, this paper proposes a device fault detection method based on TDA, leveraging TDA's capability in feature recognition. First, the time-series data is mapped to a high-dimensional space, allowing the application of topological methods for analysis. Betti curves are used to analyze the topological structure of the time series, revealing its topological features at different scales. Then, in addition to the topological features, statistical features of the time series are incorporated to enhance the expressive power of the features. Finally, these topological and statistical features are integrated into a feature space, and a machine learning classifier is used for fault type recognition and classification. The method has shown promising results in experiments.

Non-intrusive load monitoring (NILM) is a technique that involves analyzing changes in voltage and current flowing through the main feeder to determine which appliances are in operation and their energy consumption. With the increasing amount and diversity of electric loads nowadays, it is becoming increasingly important to extract unique load signatures and build robust classification models for NILM. Topological data analysis (TDA) studies the properties of space that are preserved under continuous deformations, which can reveal the structure and relationships within complex datasets, such as networks, graphs, and manifolds. In this paper, we use TDA as feature extractor to mine vast of non-linear shape features from Voltage-Current (V-I) trajectory for appliance identification. Then, an adaptive feature selection method based on mutual information (MI) is proposed, as a result, the selected subset of features is more discriminative for the adopted dataset. Further, using selected features, the strategy of Frienemy Indecision Region Dynamic Ensemble Selection (FIRE-DES), which adaptively selects combination of classifiers according to data-samples by considering the improvement by indecision region, is employed for non-intrusive appliance classification. By such fusion of TDA and technique of ensemble learning, the experimental results on two public datasets prove efficacy of proposed method in both identification accuracy and computation time.

In smart meters, loose contact at screw terminals can lead to prolonged overheating and arcing, posing significant fire hazards. To mitigate these risks through early fault detection, this study proposes a data-driven framework integrating the Local Outlier Factor (LOF) and Multiple Linear Regression (MLR) algorithms. Voltage differentials, extracted from operational data collected via a simulated multi-meter metering enclosure, are leveraged to diagnose terminal contact degradation. Specifically, LOF identifies arc faults, characterized by abrupt and transient voltage deviations, by detecting outliers in voltage differentials, while MLR quantifies contact resistance through regression analysis, enabling precise loose contact detection, a condition associated with gradual and persistent voltage changes due to increased resistance. Extensive validation demonstrates the framework’s robustness, outperforming conventional centralized methods in diagnostic accuracy and adaptability to diverse load conditions.

Energy consumption in the industrial consumers accounts for a significant proportion of total usage in society. Employing non-intrusive load monitoring (NILM) for estimating device-level industrial energy consumption is of great economic and environmental significance. However, existing NILM methods are primarily designed for residential scenarios, and these methods generally suffer from performance degradation after migrating to industrial scenarios which are with more complex load characteristics and serious equipment overlapping operation. In this regard, this paper proposes a temporal attention and dilated convolution based multi task model, named TADCMT. Firstly, a temporal attention module is introduced to strengthen the capacity of the model in capturing load features at key time steps. Furthermore, by incorporating dilated convolution structures, the receptive field is effectively enlarged, which enables a more accurate representation of complex spatial features. In addition, a multi-task learning framework is adopted, through which load disaggregation and state identification are jointly addressed, and the synergy between the two tasks is exploited to further improve overall model performance. Comparison experiments conducted on the public HIPE dataset demonstrate that our TADCMT model performs better on multiple appliances than the baseline, which validates its effectiveness and robustness in load disaggregation for industrial NILM.

Non-intrusive load monitoring (NILM) offers a practical solution for intelligent energy management in consumer electronics by estimating fine-grained energy consumption information for individual appliances from aggregated power signals measured at limited locations. With the rapid increase in appliance brands and types, constructing unique load signatures and establishing reliable appliance identification methods become progressively crucial for NILM. Deep neural networks have proven effective feature extraction capabilities and excellent identification performance in different fields. The primary approach in the current literature to extend this trend involves expanding the size of networks. However, this leads to rapidly increasing computational costs, with only marginal improvements in performance. In this paper, we propose and compare three different strategies for the fusion of the V-I trajectory and statistical features in an end-to-end trainable manner, evaluating feature fusion at various stages within the model (Feature-Level Fusion and Decision-Level Fusion), which automatically learn jointly from V-I features and a set of manually defined statistical features. As a result, the proposed approach improves cost efficiency while enhancing appliance identification performance. Experiments on diverse benchmark datasets demonstrate that the proposed feature fusion models cost-effectively achieve cutting-edge appliance identification performance.

Non-Intrusive Load Monitoring (NILM) involves identifying and analyzing electrical loads through monitoring fluctuations in physical parameters like voltage and current. The escalating diversity and quantity of modern electrical devices have underscored the critical need for effective feature extraction and robust classification modeling in NILM. While deep learning has advanced feature extraction, these methods often produce features with ambiguous mathematical meanings and high computational costs. This study introduces h-vectors from combinatorial commutative algebra as a novel feature extraction tool for NILM. By leveraging mathematical principles, this approach rapidly extracts nonlinear, shape-based features from active power signals, offering clear interpretability. Experimental validation on the PLAID public dataset reveals that our method achieves faster load identification while significantly improving accuracy for complex appliances compared to state-of-the-art techniques. Thus, the proposed framework represents a balanced NILM solution, harmonizing computational efficiency, accuracy, and interpretability—a key advancement for practical deployment.