Zhijun Zhang

Increased adoption of electric vehicles demands reliable battery health and lifespan predictions for safety and performance. However, current machine learning methods often struggle with limited labeled data and poor generalization across battery types and operating conditions, resulting in training inefficiencies and reduced prediction accuracy. To fill this gap, this study proposes a generative pre-training (GPT) method for battery degradation, named batteryGPT, to enhance the training efficiency and prediction accuracy of tasks for state of health (SOH) and remaining useful life (RUL) estimations. The proposed batteryGPT integrates attention-based multi-modal fusion, dynamic masking, and a Transformer with windowed multi-scale mechanisms during pre-training to capture robust temporal relationships across voltage, current, and temperature data. By leveraging these methods, batteryGPT can learn generalized features from heterogeneous data sources, improving its ability to adapt to diverse operational environments. Fine-tuning adapts the model to specific tasks through task-specific heads, ensuring accurate battery health and lifecycle predictions. Experimental results show that, when fine-tuned with only a few labeled cells, batteryGPT achieves SOH estimation errors below 0.005 and early-cycle RUL prediction errors of about 0.03 across multiple public datasets. These results are obtained under a few-sample fine-tuning setting, while the pre-training stage relies exclusively on unlabeled degradation data.

Reliable power line monitoring is critical for consumer energy systems that increasingly depend on distributed sensors, smart terminals, and IoT-enabled edge devices. In practical deployments, fault detection performance is often affected by changing operating conditions, noisy measurements, and unreliable prediction confidence. This paper proposes a risk-aware AIoT-based fault detection framework that integrates multi-domain signal representation, uncertainty-aware inference, and confidence-adaptive decision making. High-frequency waveform signals are analyzed across time, frequency, and wavelet domains to capture complementary fault characteristics, and are adaptively fused to preserve localized and nonstationary patterns. Instead of producing deterministic outputs, a Dirichlet-based Prior Network is employed to jointly estimate fault probability and predictive uncertainty. These quantities are combined into a scalar risk score that enables a three-outcome decision strategy, allowing the system to issue alarms, defer uncertain cases, or identify normal operation. Experiments on a real-world power line dataset show that the proposed approach maintains stable detection performance and well-calibrated confidence under distribution shift. The results demonstrate that incorporating uncertainty-aware inference and risk-adaptive decision logic improves the reliability of AI-enabled monitoring systems, making the framework suitable for IoT-driven consumer energy applications.

While Electric Vehicles (EVs) have the potential to contribute to low-carbon transportation, the burgeoning adoption of EVs and uncoordinated charging can place a significant burden on the power grid. This paper addresses the challenges of the integration of massive EVs into smart grids by introducing a comprehensive framework with heterogeneous flexibility aggregation. First, a Spatio-Temporal Heterogeneous Graph Neural Network (STH-GNN) is proposed to accurately predict EV charging demands by analyzing complex spatio-temporal relationships. Furthermore, a bi-level coordinated charging scheduling framework optimizes grid operations and individual EV charging schedules, relieving grid burden and enhancing efficient energy use based on the STH-GNN prediction. It solves the gaps between prediction errors and real-time actual charging demand. Additionally, the EV flexibility set is modelled to effectively aggregate heterogeneous EV resources, allowing for optimized deployment of large-scale EV assets by constructing the inner approximation of the Minkowski sum of the individual flexibility sets. The proposed method is verified in simulation, utilizing data from a realistic urban setting with diverse EV charging behaviours and grid conditions. Simulations demonstrate the effectiveness of the STH-GNN in accurately forecasting EV charging demands across different times and locations. The bi-level charging scheduling framework successfully manages grid burden while showcasing significant improvements in operational efficiency and cost reduction. Results validate the proposed model’s robustness and scalability, proving its potential applicability in real-world smart grid environments.

The large-scale integration of distributed photovoltaics (PVs) in the new power system increases the proportion of dispatchable resources, enabling their participation in fast power-frequency regulation (FPFR) of the system. However, the large number and wide distribution of distributed PVs limit their ability to directly participate in regulation. To address this issue, this paper proposes a hierarchical fast power-frequency regulation method based on aggregated distributed PVs. Firstly, to tackle the aggregation difficulty posed by the diverse locations, capacities, and adjustment capabilities of distributed PVs, a clustering-based aggregation method for distributed PVs is proposed, which utilizes the multi-dimensional characteristics of distributed PVs to precisely quantify the node relationships and then optimize the modularity function, thereby achieving the accurate aggregation of distributed PVs. Secondly, to mitigate the impact of power fluctuations and nonlinearity in aggregated distributed PVs on their participation in second-timescale FPFR, a combined prediction method is proposed, which integrates the strengths of both Autoregressive Integrated Moving Average (ARIMA) model and grey system model to enable faster and more accurate power prediction of the aggregated distributed PVs in the next few seconds. Furthermore, considering the effect of information and electricity flow among aggregations, a consensus-based power allocation strategy is designed from a cyber-physical fusion perspective. This strategy dynamically adjusts communication paths to mitigate the impact of communication delays while simultaneously minimizing the line power losses among aggregations. Finally, simulation results validate the effectiveness of the proposed method.

With increasing integration of intermittent renewable energy into power systems, effective frequency regulation is more urgent. Virtual power plants (VPPs), aggregating demand-side distributed resources, have great potential for frequency regulation. Nevertheless, deep cyber-physical interactions render VPPs susceptible to diversified cyber attacks, such as denial of service (DoS), false data injection (FDI), and deception attacks, which degrade frequency performance. Notably, existing research on VPPs’ frequency regulation largely ignores comprehensive impacts of such attacks. To address this gap, this paper presents a multi-agent system based cooperative power control method for distributed VPPs to ensure frequency regulation under diversified cyber attacks. First, we introduce a novel VPPs cooperative power control strategy to obtain optimal power references for VPPs and mitigate the combined effects of DoS and FDI hybrid attacks. It combines acknowledgment technique, event-triggered mechanism, and trust agents-based state screening and compensation method, enhancing reliability and economic efficiency of VPPs’ frequency regulation. Moreover, we propose a unique inverters dynamic power control scheme for accurate reference tracking of each internal unit within VPPs under deception attack. Based on a new event-triggered dynamic power control model with deception attack, it employs a specially designed dynamic event-triggered communication-based control strategy, ensuring robust tracking control and reducing communication burdens. Finally, simulation results demonstrate the validity and superiority of our presented method.

The state-of-the-art research in non-intrusive load monitoring (NILM) has emphasized the difficulty of obtaining a considerable volume of labeled data and proposed weak learning algorithms for NILM. However, the availability of aggregate load measurement can be compromised due to temporary sensor failures, leading to missing data and incomplete labels. Most existing methods fail to address these issues simultaneously or extract comprehensive features. To overcome this, we propose GPT-iMVML, a generative pre-trained transformer (GPT)-assisted multi-view multi-label learning framework designed for NILM under incomplete data and label conditions. The model incorporates attention-based bi-directional long short-term memory (LSTM) networks and GPT to extract short-term and long-term features from diverse measurements, enhancing representation learning. A view availability matrix and attention mechanism are applied to fuse multi-view features effectively, while an LSTM-based decoder is used to validate and correct extracted features. Furthermore, an improved weighted binary cross-entropy loss is introduced to mitigate the adverse effects of incomplete labels on model training. Experimental results on public datasets demonstrate that GPT-iMVML significantly improves classification performance and robustness, outperforming typical baseline methods. The proposed framework effectively alleviates the impact of simultaneous data and label loss, offering enhanced feature extraction, generalization, and adaptability for NILM applications under realistic constraints.

The virtual power plants (VPPs) formed by aggregating distributed energy resources on the demand side have high potential in frequency regulation of power systems. However, due to the complexity of real communication network, diverse communication issues, such as random packets loss and malicious cyber-attacks, may cause undesirable power dispatch and inaccurate dynamic control. This brings great challenges to the participation of VPPs in frequency regulation. To bypass these hurdles, we propose a cyber-physical integration design to facilitate VPPs frequency regulation in a real-time and reliable manner. First, we suggest a joint design scheme of VPPs power dispatch and communication resources allocation for packets loss under cloud-edge collaboration. That is, in the physical-layer, we propose the VPPs power dispatch under packets loss and in the cyber-layer, we develop the communication resources allocation strategy. This scheme aims to eliminate the packets loss impacts and thus determine accurate power dispatch. Second, we present a high reliable information transmission approach for preventing cyber-attacks under edge-end collaboration. Specifically, in the physical-layer, we implement dynamic control for responding to received regulation power and in the cyber-layer, we execute digital signcryption algorithm. Its objective is to improve system information transmission against cyber-attacks and thus ensure the accurate dynamic control. Finally, simulation results validate the feasibility and superiority of our proposed integration design.

To alleviate the pressure of traditional fixed charging methods, mobile charging solutions have emerged in recent years. This article presents a two-stage coordination approach for mobile charging robots (MCRs) and active distribution networks (ADN). In the first stage, we maximize the total revenue for the MCRs from providing charging services for electric vehicles (EVs) and interacting with the ADN within a frame-based time scale, where the duration of each frame is determined by the charging decisions of MCRs. Furthermore, we consider the long-term objectives of meeting the battery energy deficit constraint for each MCR, thereby improving their battery life. Lyapunov optimization is utilized to transform frame-based scheduling into an optimization problem within a specified time slot, simplifying the process of solving the optimization problem. To avoid affecting charging efficiency when MCRs interact with the power grid, we use the reactive power of free MCRs to provide voltage regulation support for the ADN, improving power quality and minimizing total network loss simultaneously. Decoupling active and reactive power in two stages ensures both the profitability of the MCRs cluster and the voltage security of the ADN, resulting in a win-win situation. The simulation results, based on the 18-bus and 51-bus distribution networks, confirm the effectiveness and superiority of the proposed approach.

Virtual power plants (VPPs) possess significant potential to support power systems’ frequency regulation through the aggregation of demand-side distributed energy resources. However, due to their heavy reliance on information technology, VPPs inevitably face diverse communication issues, such as cyber-attacks, privacy disclosure, and packet losses, which poses a serious challenge to ensure efficient power control for frequency regulation. To this end, this paper proposes a multiagent system-based VPPs cooperative power control strategy to achieve frequency regulation in a resilient, optimal, and scalable manner. First, a joint design scheme of VPPs cooperative power control and information interaction mechanism is presented to determine the optimal regulation power of VPPs for frequency regulation, where a consensus-based collaborative strategy is designed for cyber-attacks and a homomorphic encryption technology is introduced for privacy protection. Also, an event-triggered mechanism is adopted for communication burden. Second, a dynamic power control scheme for inverter is presented to enable the internal units of VPPs to track the power reference from the joint design scheme, in which an improved model predictive control method is suggested for packet losses to enhance control performance. Finally, simulation results reveal that, when confronting diverse communication issues, the proposed strategy exhibits significant effects in defending against cyber-attacks, ensuring privacy security, alleviating the communication burden, and achieving fast response, thus effectively facilitating VPPs frequency regulation.