Abstracts:This article proposes a robust multiarea distribution system state estimation method for interval estimation of state variables based on a physics-informed decentralized graphical representation network and Gaussian process (GP)-aided multiarea state estimators. The real-time and pseudomeasurements are first cast to a graph with tree topology and a graph attention-based representation network is employed to capture the structural information between measurements from the historical data. A centralized pretraining and distributed inference framework is developed to extract essential global information from historical data and extend it to various subregions. Then, the robust nodal features extracted by the graphical network are fed into the GP with a multitask kernel for multiarea state estimation. The adopted kernel can find relevance between tasks for different subregions that are useful for the multiarea state estimation. The embedding of structural information in the representation network enables the proposed method to achieve robustness in the presence of outliers. The adopted kernel further allows us to reduce the reliance on network communication and achieve accurate multiarea state estimation. It also offers the ability to quantify the uncertainty of state variables, yielding more valuable estimation outcomes. Experimental results demonstrate the effectiveness of the proposed method in handling abnormal data and accurately quantifying the uncertainty of state variables.

Abstracts:Face information in public images is vulnerable to tampering. Some studies have used pre-embedded watermarks to detect tampering but cannot recover the original face. To address this, we propose a proactive face hiding network that conceals face information in the background region for the first time. Our framework includes three U-Net-based modules: a preparation network, an encoder, and a recovery network. Special loss functions are designed to achieve our objective of recovering the original face from a protected image attacked by face swapping. In addition, we develop a neural network-based JPEG simulator and a differentiable simulator, offering a fresh perspective on addressing the robustness problem associated with JPEG compression. Our method generates protected images with a peak signal-to-noise ratio (PSNR) of 40.947 dB in experiments. Even after different face attacks, the recovered images maintain PSNR between 28.368 and 33.847 dB. After the attack of JPEG compression, the PSNR of the recovered image decreases by a maximum of 2.142 dB. Our scheme effectively generates high-quality protected images that resist face swapping and JPEG compression attacks, enabling recovery of the original faces.

Abstracts:Iterative learning control (ILC) achieves high control precision across various motion systems during repetitive tracking tasks by successively updating the compensation. In identical control circumstances, tracking errors primarily comprise repetitive components to be eliminated, induced by the input signals and regular system disturbances. In order to mitigate repetitive errors efficiently, conventional ILC methods treat the high-frequency components of tracking errors as nonrepetitive noise and disturbances, employing the low-pass filter to exclude these components. However, this specific frequency criterion is not entirely accurate, as there can be interference in the low-frequency range and effective components in the high-frequency range. Therefore, an energy-selected ILC is proposed in this article to identify these components, thereby enhancing the filtering validity. The proposed method proposes a novel energy criterion to construct the robust filter, improving the capability to distinguish repetitive components. Based on this advanced robust filter, the trajectory modification is designed as the learning filter to accelerate the convergence rate. The stability and convergence of this method are thoroughly proven and analyzed. Various comparative experiments have been conducted to illustrate the effectiveness of this novel energy-selected ILC approach. Generally, the proposed method has the following superiorities: it achieves high control precision across various motion scenarios; it broadens wide bandwidth applicable in high-frequency and nonsmooth circumstances; and it has an accurate error analysis suitable for practical applications. Meanwhile, it improves the control performance of classic ILC while maintaining its ease of implementation.

Abstracts:To address the high-dimensional and complex scheduling issues in the low-carbon economic dispatch (LCED) with carbon capture plants, in this article, we propose a novel safe reinforcement learning (SRL) based on heterogeneous action space representation, which can make fast decisions for both optimal power flow and carbon capture operation. First, SRL is designed based on the feasible set to ensure that the dispatch results continuously remain within the preset range. Then, to tackle the problem of having a large number of discrete and continuous variables in the LCED, this article employs a parameterized Markov process to represent these discrete–continuous actions and uses a conditional variational autoencoder to depict heterogeneous space. To learn the correlation between discrete and continuous action spaces, a mechanism for approximating action space based on small-sample behavior cloning is proposed, and a method based on dynamic time warping for calculating environment similarity is designed for determining the value of the regularization term. Finally, numerical simulations validate the superiority and scalability of the proposed method in enhancing decision-making efficiency and promoting the low-carbon economic operation of the power system.

Abstracts:In the Industrial Internet of Things (IIoT), ensuring timely and secure data transmission between sensors and edge devices is paramount, particularly when dealing with sensitive information captured by high-resolution image sensors. However, existing methods often struggle to strike the delicate balance between security and efficiency, resulting in either vulnerable transmissions or significant processing delays. This article proposes a novel image encryption algorithm specifically designed for IIoT environments with constrained resources. The proposed algorithm leverages a new chaotic model that combines a cyclic construction with a 1-D perturbed logistic map. The proposed chaos model produces a single data sequence, subsequently utilized to generate three matrices mirroring the size of the image. Among these matrices, two contribute to crafting a permutation matrix for randomizing pixel positions, while the third matrix facilitates diffusion operations. The chaotic data sequence and generated matrices are preprocessed and can be used for encrypting multiple images under the same session key, enhancing efficiency. Encryption utilizes a single round that combines diffusion and permutation operations simultaneously, further reducing processing time. Experimental results demonstrate its effectiveness on an IoT camera sensor for encryption and a separate device for decryption. Statistical tests confirm the robustness of the encrypted images against various attacks, including correlation analysis, histogram analysis, differential attacks, and key and plaintext sensitivity. Furthermore, comparisons with existing image encryption techniques showcase the proposed algorithm's superior security and efficiency. Notably, it effectively encrypts images of varying sizes, making it suitable for deployment in IIoT environments.

Abstracts:Probability forecasting is a powerful tool for quantifying uncertainty in short-term load forecasting. However, its performance may be hampered by excessive feature redundancy and the quantile crossing phenomenon. To overcome these challenges, this study proposes a novel deep noncrossing quantile method with multi-information fusion for day-ahead load probabilistic density forecasting. This method extracts different types of input features through distinct neural networks, and can reduce the redundancy of feature information. Based on the positive differences among output values from neural networks, a novel quantile noncrossing strategy is introduced. This strategy, integrated within the neural network, eliminates quantile crossing phenomena and enhances the interpretability of model during the training process. Experimental results show that the proposed model reduces the quantile loss by 11% to 31%, produces prediction intervals with higher quality, precision, and no crossovers.

Abstracts:Vold–Kalman filter (VKF) is a powerful tool for time-frequency (TF) decomposition of nonstationary signals. However, the overdependence on instantaneous frequency (IF) estimation, neglect of nonlinear initial phase, and improper bandwidth selection against noise interference limit its practical performance in mechatronic transmission fault diagnosis under variable speed conditions. This article proposes a novel signal processing method named iterative adaptive Vold–Kalman filter (IAVKF) to tackle the challenges in VKF and realize accurate IF estimation and fault dynamic feature extraction. Specifically, an improved VKF model is developed with the consideration of nonlinear initial phase and discrepancies between true and estimated IFs. Then, the estimated IF is refined by the recovered envelope to ameliorate TF resolution. Finally, an iterative bandwidth adaptation step is developed based on signal orthogonality to reduce noise interference and ensure algorithm convergence. Numerical analysis and two engineering applications in mechatronic transmission fault diagnosis are conducted, showing that IAVKF provides higher accuracy and efficiency in fault feature extraction and IF estimation.

Abstracts:Incompleteness and irregularity are inherent challenges in 3-D point clouds, which hinder point-cloud-based 3-D object or scene understanding, especially for complicated industrial scenarios. Previous point cloud completion methods typically suffer from the nonuniform density of the restored point cloud. In this article, we proposed a global and uniform point cloud completion algorithm via a density-sensitive transformer for reconstructing complete and fine-grained 3-D small-scale industrial components. We first designed a position and feature encoding module to aggregate discriminative point cloud features. Then, we constructed a density-sensitive transformer structure with a coarse point cloud generator that allowed for recovering the global structure and uniform local details of the 3-D object. The experimental results on a real-world industrial component dataset and a public multiobject dataset demonstrated that our method achieved state-of-the-art performance with the former CD-$\mathcal{l}$1 of 8.17 × 10-3, avg-NUC of 0.61 and the latter CD-$\mathcal{l}$1 of 5.33 × 10-3, avg-NUC of 0.74. Our method better balanced the global point cloud integrity and density uniformity in a coarse-to-fine pipeline, which benefited high-quality 3-D object reconstruction. The method has been applied to practical aero-engine component reconstruction in the real-world scenario.

Abstracts:Compressed sensing (CS) for sensor-near vibration diagnostics represents a suitable approach for the design of network-efficient structural health monitoring systems. This article presents a solution for vibration analysis based on deep neural networks (DNNs) trained on compressed data. The envisioned maintenance system consists of a network of sensing nodes orchestrated by a very constrained centralizing unit. The latter is equipped with a microcontroller unit (MCU) that predicts the health state using the aggregated information. As a major contribution, the DNN architectures are generated automatically from the data through a procedure inspired by hardware-aware (HW) neural architecture search (NAS), called as HW-NAS-CS, which is uniquely refined with additional constraints that consider both the peculiarities of CS parameters and the limitation of embedded devices. The proposed approach has been validated using two real-world SHM datasets for vibration damage identification and eventually deployed on a low-end computing platform (the STM32L5 MCU). Results demonstrate that DNNs combined with adapted CS schemes can attain classification scores always above 90% even in case of very huge compression levels (higher than 64x): these performances significantly improve the ones attained by state-of-the-art approaches in the field, with the utmost advantage of being portable on embedded devices.