With the increased dependency on Wide Area Network (WAN) signals in power systems for critical functions, cyber-attacks on the communication lines can lead to catastrophic results, such as the complete shutdown of a c...
详细信息
ISBN:
(纸本)9798350372410;9798350372403
With the increased dependency on Wide Area Network (WAN) signals in power systems for critical functions, cyber-attacks on the communication lines can lead to catastrophic results, such as the complete shutdown of a country's power grid. Detecting these potential attacks on WAN signals is of utmost importance. In this work, we explore several data-driven approaches for detecting stealthy false data injection attacks on a critical component of the power system, the Wide Area Damping control (WADC). After extensive simulations on the two-area test system for data collection, we test four machine/deep learning approaches to detect the attack injections in real time. The four techniques are eXtreme Gradient Boosting (XGBoost), Gradient Boosting Classifier (GBC), Long Short Term Memory (LSTM), and Convolutional Neural Networks (CNNs). We consider two cases;the first is where all voltage and current signals are sent to the central WADC. The second is when only the required tie line power measurement is sent. In both cases, LSTM results in the highest detection accuracy, exceeding 99.5% in the first case and 94.2% in the second case. These results reflect the importance of redundant data availability.
Iterative learningcontrol (ILC) is excellently applied to output tracking for systems that perform repetitive tasks. Classical norm optimal ILC (NOILC) approach yields optimal control inputs by minimizing a quadratic...
详细信息
In machine learning-driven landslide prediction models, data quality directly determines the accuracy of landslide predictions. Currently, data preprocessing methods mainly involve fitting the data using moving averag...
详细信息
Electric power steering (EPS) systems form a fundamental unit in modern automotive vehicles, providing motor assistance to aid the driver's manual maneuvering. Preventing loss of assist (LoA) in advance is critica...
详细信息
Electric power steering (EPS) systems form a fundamental unit in modern automotive vehicles, providing motor assistance to aid the driver's manual maneuvering. Preventing loss of assist (LoA) in advance is critical for EPS systems to mitigate fatal accidents and reduce maintenance costs. As part of the recent interest in intelligent software-defined vehicles (SDV), data-driven approaches have gained much attention to overcome the limitations of conventional fail-safe mechanisms and provide value-added maintenance strategies. While related works in the field have shown promising results, they are limited to proof-of-concept studies validated under simulation and test-bed environments. Here, we present a novel deep learning (DL)-based method to detect EPS performance degradation using experimental data acquired from a commercial vehicle' s controller area network (CAN) bus. Our approach initially proposes a neural fault observer model and its adversarial learning scheme to represent the EPS system's normal operating dynamics. We demonstrate that our proposed model can detect degradation levels down to ten percent from normal conditions under various driving scenarios based on an anomaly detection mechanism that outperforms baseline methods in quantitative and qualitative measures. Furthermore, we provide physically relevant intuitions of our closed-box model's inference mechanism based on its attention-based saliency map to strengthen the reliability aspect of our data-driven approach. Lastly, we demonstrate that a quantized model can operate in real-time on an automotive electronic control unit (ECU) device.
With the rapid development of imaging sensor technology in the field of remote sensing, multi-modal remote sensing data fusion has emerged as a crucial research direction for land cover classification tasks. While dif...
详细信息
With the rapid development of imaging sensor technology in the field of remote sensing, multi-modal remote sensing data fusion has emerged as a crucial research direction for land cover classification tasks. While diffusion models have made great progress in generative models and image classification tasks, existing models primarily focus on single-modality and single-client control, that is, the diffusion process is driven by a single modal in a single computing node. To facilitate the secure fusion of heterogeneous data from clients, it is necessary to enable distributed multi-modal control, such as merging the hyperspectral data of organization A and the LiDAR data of organization B privately on each base station client. In this study, we propose a multi-modal collaborative diffusion federated learning framework called FedDiff. Our framework establishes a dual-branch diffusion model feature extraction setup, where the two modal data are inputted into separate branches of the encoder. Our key insight is that diffusion models driven by different modalities are inherently complementary in terms of potential denoising steps on which bilateral connections can be built. Considering the challenge of private and efficient communication between multiple clients, we embed the diffusion model into the federated learning communication structure, and introduce a lightweight communication module. Qualitative and quantitative experiments validate the superiority of our framework in terms of image quality and conditional consistency. To the best of our knowledge, this is the first instance of deploying a diffusion model into a federated learning framework, achieving optimal both privacy protection and performance for heterogeneous data. Our FedDiff surpasses existing methods in terms of performance on three multi-modal datasets, achieving a classification average accuracy of 96.77% while reducing the communication cost.
In this article, an H-infinity cooperative optimal output regulation (H-infinity) method based on policy iteration is proposed for unknown multiagent systems (MASs). First, the dynamics matrix and states of the unknow...
详细信息
In this article, an H-infinity cooperative optimal output regulation (H-infinity) method based on policy iteration is proposed for unknown multiagent systems (MASs). First, the dynamics matrix and states of the unknown leader are estimated by designing a data-driven distributed observer. Then, the H-infinity problem is decomposed into a static constrained optimization problem and a zero-sum game dynamic optimization problem. Optimal feedback gains are obtained by designing a model-free three-phases off-policy learning algorithm for solving the zero-sum game optimization problem. In two initialization phases, the stabilizing gains are obtained via iterating the poles of the closed-loop system into the stabilization region without a priori knowledge of the MASs. The feedforward control policy is obtained by solving the static constrained optimization problem. H-infinity is realized by means of a data-based distributed feed-forward feedback controller. For the proposed method, we prove that even if state estimation errors are present, stabilizing gains can still be obtained and that optimal solutions to GAREs can be obtained as the estimation errors asymptotically converge to zero. Finally, the effectiveness of the algorithm is verified by simulation.
This paper proposes a fast terminal sliding mode control based on adaptive neural network to address the problem of external interference and internal uncertainty in trajectory tracking control for a six degree of fre...
详细信息
Vehicle lane change is one of the most common driving behaviors. It is of great significance to ensure that the autonomous vehicle can change lanes safely and accurately. Therefore, in this paper, a model predictive c...
详细信息
Improving system-level resiliency of networked microgrids against adversarial cyber-attacks is an important aspect in the current regime of increased inverter-based resources (IBRs). To achieve that, this paper contri...
详细信息
Improving system-level resiliency of networked microgrids against adversarial cyber-attacks is an important aspect in the current regime of increased inverter-based resources (IBRs). To achieve that, this paper contributes in designing a hierarchical control layer, in conjunction with the existing control layers, resilient to adversarial attack signals. Considering model complexities, unknown dynamical behaviors of IBRs, and privacy issues regarding data sharing in multi-party-owned microgrids, designing such a control layer is non-trivial. Here, to tackle these issues, a novel federated reinforcement learning (Fed-RL) method is proposed. To grasp the interconnected dynamics of networked microgrids, the paper develops Federated Soft Actor-Critic (FedSAC) algorithm following the vertical structure of implementing Fed-RL. Next, utilizing the OpenAI Gym interface, we built a custom set-up in GridLAB-D/HELICS co-simulation platform, named Resilient RL Co-simulation (ResRLCoSIM), to train the RL agents with ieee 123-bus benchmark comprising 3 interconnected microgrids. Finally, the learned policies in the simulation are transferred to the real-time hardware-in-the-loop (HIL) test-bed developed using the high-fidelity Hypersim platform. Experiments show that the simulator-trained RL controllers achieve desirable performance with the test-bed platform, validating the minimization of the sim-to-real gap.
暂无评论