A5106484176- Advanced Neural Network Applications
- Smart Grid Energy Management
- Power Systems and Renewable Energy
- CCD and CMOS Imaging Sensors
- Optimal Power Flow Distribution
- Advanced Vision and Imaging
- Innovation Diffusion and Forecasting
- Advanced Image Processing Techniques
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Energy Load and Power Forecasting
- Electric Power System Optimization
- Autonomous Vehicle Technology and Safety
- Advanced Manufacturing and Logistics Optimization
- Generative Adversarial Networks and Image Synthesis
- Brain Tumor Detection and Classification
- Smart Grid and Power Systems
- Neural Networks and Applications
- Real-time simulation and control systems
- Power System Optimization and Stability
- Energy Efficiency and Management
- Energy, Environment, and Transportation Policies
- Microgrid Control and Optimization
- Electric Vehicles and Infrastructure
- Occupational Health and Safety Research
NARI Group (China)
2020-2024
Xilinx (United States)
2018-2019
Beijing University of Posts and Telecommunications
2018
Electric Power Research Institute
2014
Convolution neural networks (CNNs) have been widely applied in the fields of computer vision tasks. However, it is hard to deploy those standard into embedded devices because their large amount operations and parameters. MobileNet, state-of-the-art CNN which adopts depthwise separable convolution replace has significantly reduced parameters with only limited loss accuracy. A high-performance processor based on FPGA proposed this paper. To improve efficiency, two dedicated computing engines...
State-of-the-art techniques in Generative Adversarial Networks (GANs) have shown remarkable success image-to-image translation from peer domain X to Y using paired image data. However, obtaining abundant data is a non-trivial and expensive process the majority of applications. When there need translate images across n domains, if training performed between every two complexity will increase quadratically. Moreover, with domains only at time cannot benefit other which prevents extraction more...
Continuous increase of wind power penetration brings high randomness to system, and also leads serious shortage primary frequency regulation (PFR) reserve for system whose capacity is typically provided by conventional units. Considering large-scale participating in PFR, this paper proposes a unit commitment optimization model with respect coordination steady state transient state. In addition traditional operation costs, losses farm de-loaded operation, environmental benefits safety costs...
The convolution neural networks (CNNs) are widely used in computer vision applications nowadays. However, the trends of higher accuracy and resolution generate larger networks, indicating that computation I/O bandwidth key bottlenecks to reach performance. Xilinx's latest 7nm Versal ACAP platform with AI-Engine (AIE) cores can deliver up-to 8x silicon compute density at 50% power consumption compared traditional FPGA solutions. In this paper, we propose XVDPU: AIE-based int8-precision CNN...
Advanced Driver Assistance Systems (ADAS) help the driver in driving process by detecting objects, doing basic classification, implementing safety guards and so on. Convolution Neural Networks (CNN) has been proved to be an essential support ADAS. We designed architecture named Aristotle execute neural networks for both object detection semantic segmentation on FPGA. DNNDK (Deep Learning Development Toolkit), a full-stack software tool, with tens of compilation optimization techniques is...
Advanced Driver-Assistance Systems (ADAS) can help drivers in the driving process and increase safety by automatically detecting objects, doing basic classification, implementing safeguards, etc. ADAS integrate multiple subsystems including object detection, scene segmentation, lane so on. Most algorithms are now designed for one specific task, while such separate approaches will be inefficient which consists of many modules. In this paper, we establish a multi-task learning framework...
Today, convolutional neural networks (CNNs) are widely used in computer vision applications. However, the trends of higher accuracy and resolution generate larger networks. The requirements computation or I/O key bottlenecks. In this article, we propose XVDPU: AI Engine (AIE)-based CNN accelerator on Versal chips to meet heavy requirements. To resolve IO bottleneck, adopt several techniques improve data reuse reduce An arithmetic logic unit is further proposed that can better balance...
Convolutional Neural Network (CNN) and Recurrent (RNN) have made great progress in machine learning community. Combining CNN RNN can accomplish more general complex tasks. Many specially designed hardware accelerators on FPGA or ASIC been proposed for RNN, yet few of them focus CNN-RNN-based models purpose applications. In this paper, we propose a complete design framework deploying general-purpose CNNRNN-based FPGAs. We use Deephi Aristotle Descartes IPs to build an efficient reconfigurable...
3-Dimensional (3D) convolutional neural networks (CNN) are widely used in the field of disparity estimation. However, 3D CNN is more computationally dense than 2D due to increase dimension. To enable practical applications autonomous driving, robotics, and other scenarios on embedded devices, we propose a unified 2D/3D accelerator (A-U3D) design. This design unifies standard / transposed convolution into convolution, respectively. Our processing unit can support same mode without additional...
Since wind and solar power are generally featured with randomness, intermittence volatility, the high proportion of new energy will adversely affect reliability grid. Energy storage can improve flexibility system also be employed to matter large-scale grid connection. In this context, article takes centralized as research object. order give full play regulation storage, remaining capacity is used reserve on basis peak-shaving valley-filling. article, Firstly, operation characteristics...
With the gradual increase in penetration of new energy distribution network, overall capacity generating units has declined, adequacy system is insufficient, and problem supply assurance become more serious. Therefore, this paper proposes a demand-side resource optimal scheduling model for active network with sufficient assurance. Firstly, different characteristics four resources, storage, heat cold storage electric vehicle, as well orderly power enterprises are analyzed, user satisfaction...
Low bit quantization of neural network is required on edge devices to achieve lower power consumption and higher performance. 8bit or binary either consumes a lot resources has accuracy degradation. Thus, full-process hardware-friendly solution 4A4W (activations 4bit weights 4bit) proposed better accuracy/resource trade-off. It doesn't contain any additional floating operations comparable full-precision. We also implement low-precision accelerator for CNN (LPAC) the Xilinx FPGA, which takes...
In recent years, the centralized interconnection of renewable energy sources has led to a surge in reserve demand power grids. However, relying solely on conventional for risk management and control obviously cannot meet operational requirements grid. view shortcomings existing optimal configuration reserve, this paper studies unconventional namely emergency which can participate short-term regulation system additional or cost. paper, based method, with goal minimizing total cost regional...