A5071362658- Multilevel Inverters and Converters
- Advanced DC-DC Converters
- Microgrid Control and Optimization
- Topic Modeling
- Silicon Carbide Semiconductor Technologies
- Natural Language Processing Techniques
- HVDC Systems and Fault Protection
- Smart Grid Energy Management
- Multimodal Machine Learning Applications
- Optimal Power Flow Distribution
- Power Systems and Renewable Energy
- Islanding Detection in Power Systems
- Electric Power System Optimization
- Reinforcement Learning in Robotics
- Advanced Graph Neural Networks
- Magnetic Bearings and Levitation Dynamics
- Electric Motor Design and Analysis
- Advanced Battery Technologies Research
- Geoscience and Mining Technology
- High-Voltage Power Transmission Systems
- Power Line Communications and Noise
- Advanced Decision-Making Techniques
- Geomechanics and Mining Engineering
- Complex Network Analysis Techniques
- Concrete and Cement Materials Research
Jilin University
2005-2025
Union Hospital
2024-2025
Ministry of Transport
2024-2025
Shanghai Ninth People's Hospital
2025
Shanghai Jiao Tong University
2020-2025
Haier Group (China)
2025
Beijing University of Technology
2015-2024
Shandong University
2013-2024
Massachusetts General Hospital
2024
Harvard University
2024
We present a novel language representation model enhanced by knowledge called ERNIE (Enhanced Representation through kNowledge IntEgration). Inspired the masking strategy of BERT, is designed to learn strategies, which includes entity-level and phrase-level masking. Entity-level masks entities are usually composed multiple words.Phrase-level whole phrase several words standing together as conceptual unit.Experimental results show that outperforms other baseline methods, achieving new...
Recently pre-trained models have achieved state-of-the-art results in various language understanding tasks. Current pre-training procedures usually focus on training the model with several simple tasks to grasp co-occurrence of words or sentences. However, besides co-occurring information, there exists other valuable lexical, syntactic and semantic information corpora, such as named entities, closeness discourse relations. In order extract from we propose a continual framework ERNIE 2.0...
Today, conventional power systems are evolving to smart grids, which encompass clusters of AC/DC microgrids, interfaced through electronics converters. In such systems, increasing penetration the electronics-based distributed generations, energy storages, and modern loads provide a great opportunity for quality control. this paper, an overview control hybrid microgrids is presented. Different types issues studied first, with consideration real-world microgrid examples, including data...
Recently, sentiment analysis has seen remarkable advance with the help of pre-training approaches. However, knowledge, such as words and aspect-sentiment pairs, is ignored in process pre-training, despite fact that they are widely used traditional In this paper, we introduce Sentiment Knowledge Enhanced Pre-training (SKEP) order to learn a unified representation for multiple tasks. With automatically-mined SKEP conducts masking constructs three knowledge prediction objectives, so embed...
We propose a knowledge-enhanced approach, ERNIE-ViL, which incorporates structured knowledge obtained from scene graphs to learn joint representations of vision-language. ERNIE-ViL tries build the detailed semantic connections (objects, attributes objects and relationships between objects) across vision language, are essential vision-language cross-modal tasks. Utilizing visual scenes, constructs Scene Graph Prediction tasks, i.e., Object Prediction, Attribute Relationship tasks in...
Pre-trained models have achieved state-of-the-art results in various Natural Language Processing (NLP) tasks. Recent works such as T5 and GPT-3 shown that scaling up pre-trained language can improve their generalization abilities. Particularly, the model with 175 billion parameters shows its strong task-agnostic zero-shot/few-shot learning capabilities. Despite success, these large-scale are trained on plain texts without introducing knowledge linguistic world knowledge. In addition, most an...
This paper proposes an adaptive dc-link voltage control method for the two-stage photovoltaic inverter during low ride-through (LVRT) operation period. The will be controlled to follow change of grid LVRT maintain high modulation ratio so that frequency harmonics injected into can attenuated significantly. Besides, when suffering asymmetrical faults, proposed could some extent attenuate double-line-frequency ripple keep in safe operational range by shifting power front-end dc input source,...
We propose a knowledge-enhanced approach, ERNIE-ViL, which incorporates structured knowledge obtained from scene graphs to learn joint representations of vision-language. ERNIE-ViL tries build the detailed semantic connections (objects, attributes objects and relationships between objects) across vision language, are essential vision-language cross-modal tasks. Utilizing visual scenes, constructs Scene Graph Prediction tasks, i.e., Object Prediction, Attribute Relationship tasks in...
Current pre-training works in natural language generation pay little attention to the problem of exposure bias on downstream tasks. To address this issue, we propose an enhanced multi-flow sequence and fine-tuning framework named ERNIE-GEN, which bridges discrepancy between training inference with infilling mechanism a noise-aware method. make closer human writing patterns, introduces span-by-span flow that trains model predict semantically-complete spans consecutively rather than predicting...
Recent studies have demonstrated that pre-trained cross-lingual models achieve impressive performance in downstream tasks. This improvement benefits from learning a large amount of monolingual and parallel corpora. Although it is generally acknowledged corpora are critical for improving the model performance, existing methods often constrained by size corpora, especially low-resource languages. In this paper, we propose Ernie-M, new training method encourages to align representation multiple...
Recently reinforcement learning (RL) has emerged as a promising approach for quadrupedal locomotion, which can save the manual effort in conventional approaches such designing skill-specific controllers. However, due to complex nonlinear dynamics robots and reward sparsity, it is still difficult RL learn effective gaits from scratch, especially challenging tasks walking over balance beam. To alleviate difficulty, we propose novel RL-based that contains an evolutionary foot trajectory...
As highway tunnel operations continue over time, structural defects, particularly cracks, have been observed to increase annually. Coupled with the rapid expansion of networks, traditional manual inspection methods proven inadequate meet current demands. In recent years, machine vision and deep learning technologies gained significant attention in civil engineering for detection analysis defects. However, accurate defect identification tunnels presents challenges due complex background...
Supercapacitors (SCs) are increasingly used in energy storage system to tackle the fast power transients. They can be individually or together with other units like batteries. In a microgrid, multiple SCs could installed at different locations without communications among them. This paper focuses on management strategy (EMS) of dc where distributed generation and controlled plug-and-play feature. A novel EMS is proposed, which uses state-of-charge-based adaptive virtual impedance facilitate...
This letter proposes a novel seven-level (7L) hybrid-clamped converter, which has competitive performance and lower device count compared to existing topologies. It can easily balance the floating capacitor voltage at switching frequency, therefore ensure low ripples even under very fundamental frequencies. Both multipulse diode front-end active structures be used with proposed 7L where front end ability dc-link capacitors. The converter is suitable for medium drives (4.16 6.6 kV) that...