Search

Coin mixing can be used to protect the identity privacy of Bitcoin owners, by engaging a set of middlepersons (i.e., ${Mix}$ ) to temporarily hold on to the transacting Bitcoins and remove the linkage between the transacting parties. However, existing schemes are generally not scalable due to limitations associated with the anonymity set, and self-credibility. In this paper, we propose an efficient coin mixing scheme (CoinLayering). To achieve strong anonymity, CoinLayering randomly selects two sets of middlepersons to respectively execute Bitcoin holding and Bitcoin trading. The seller can also select lower-loaded sets of middlepersons in the shortest time possible. We also design two coin mixing protocols (CoinLayering-PA and CoinLayering-PB) to mitigate the risk due to misbehaving middlepersons and ${Supervisor}$ . We then mathematically prove that CoinLayering achieves both strong anonymity and self-credibility, and evaluate its performance to demonstrate its scalability.

Unmanned aerial vehicles (UAVs) serve as aerial base stations to provide controlled wireless connections for ground users. Due to their constraints on both mobility and energy consumption, a key problem is how to deploy UAVs adaptively in a geographic area with changing traffic demand of mobile users, while meeting the aforementioned constraints. In this paper, we propose a QoE-driven and energy-efficient adaptive deployment strategy for multi-UAV networks based on hybrid deep reinforcement learning to solve the problem of incomplete information game, where the UAVs can adjust their moving directions and distance to serve users who move randomly in the target area. Through the hybrid deep reinforcement learning with centralized training and distributed testing, UAVs can be trained offline to obtain the global state information and learn a completely distributed control strategy, with which each UAV only needs to take actions based on its observed state in the real deployment to be fully adaptive. Moreover, in order to improve the speed and effect of learning, we improve hybrid reinforcement learning, by adding genetic algorithms and TD-error-based resampling optimization mechanism. Simulation results show that the hybrid deep reinforcement learning algorithm has better efficiency and robustness in multi-UAV control, and has better performance in terms of QoE, energy consumption and average throughput, by which average throughput can be increased by 20% to 60%.

By leveraging the power of crowd, the prevalence of mobile devices in mobile crowd-sensing (MCS) networks helps and provides a wide range of sensing services through collecting and sharing sensing data. However, the diverse behaviours of mobile users affect the quality of sensing service. Unreliable or malicious users and platforms could provide untrusted data to gain illegal benefits, which deteriorates the practicability of MCS. Besides, due to individual selfishness, mobile users are reluctant to participant in sensing tasks. It is desirable to design a trust and incentive scheme to improve the service efficiency of MCS. In this paper, we propose a novel trust-driven contract incentive framework in MCS, which guarantees the service quality, while stimulating mobile users to join sensing tasks. The basic idea is to establish a trust evaluation and contract incentive scheme for mobile users and platforms. We first design a trust evaluation scheme between mobile users and sensing platforms based on the historical interactions to derive the reliability value of sensing platform. Then, the trust threshold is formulated to filter out malicious sensing platforms. By considering the privacy preferences of mobile users, we establish a contract incentive scheme to maximize the utility of both mobile users and sensing platforms. The design objective is to derive a set of optimal contracts under both discrete and continuous contract models. Meanwhile, the designed contracts guarantee the individual rationality (IR) and incentive compatibility (IC) properties. Finally, simulations are conducted to evaluate the effectiveness of the proposed trust-driven contract incentive scheme, and results demonstrate that the proposed scheme can jointly improve the quality of sensing service and maximize the utilities of mobile users and sensing platforms.

Tree density is a significant parameter for forest. However, it has always been challenging to use remote-sensing techniques to acquire it efficiently and effectively on a large spatial scale. As a matter of fact, tree counting can be regarded as end-to-end density regression, which uses a remote-sensing image as an input and generates a tree density map as an output. Therefore, the tree number can be easily obtained by the summation of density values in an area. To this end, a deep neural network has been constructed to aggregate multiple decoding paths to extract hierarchical features at different encoding stages, in order to merge tree features of multiple scales in remote-sensing images. At the same time, a hybrid loss function has been proposed to effectively guide the network training and enhance the model ability. A tree count dataset consisting of 2400 sample pairs has been constructed for training and validation. When compared with other popular counting networks, it has been found that the proposed network achieved the best results with a relative mean absolute error (rMAE) of 16.72%, a root mean squared error (RMSE) of 77.96, and an R² of 0.96, evidencing that this is a promising method for estimating tree numbers on a large spatial scale.

Cloud operating system (Cloud OS) is the heart of cloud management platform that takes control of various cloud resources. Therefore, it attracts numerous attacks, especially unauthorized access. Many existing works adopt role-based access control (RBAC) model for Cloud OS access control and token-based approaches as user credentials of sessions or transactions between users and cloud, but they fail to resist privilege abuse caused by RBAC policy rules tampering or token hijacking. To addresses this challenging problem, we propose a secure access control framework suitable for resource-centric Cloud OS. For one thing, we propose a new authorization model with cryptographically protected RBAC policy rules. To solve the policy decision problem caused by encrypted policy rules in this model, an approach is developed to transform it into permission searching problem and we further propose a policy decision scheme based on this. For another thing, we achieve user token unlinkability and token-replay-attack resistance by introducing randomization mechanism and leveraging one-show token technique. A proof of concept implementation has been developed and the proposed scheme is proven secure and efficient by security analysis and the performance evaluation.

This paper presents an online smart meter measurement error estimation algorithm. Extended Kalman filter (EKF) and limit memory recursive least square (LMRLS) methods are used for remote calibration of a large amount of user-side smart meters. Then, a modified joint estimation model is obtained by selecting the estimation step that conforms to the actual working condition and filtering the abnormal estimation value according to the line loss rate characteristics. Finally, based on the experimental data obtained by the program-controlled load simulation system, the precision of metering error estimation is verified. The results show that the method improves the precision of error estimation by analyzing the coupling between line loss rates and metering error estimation. By using the limited memory RLS algorithm, the influence of old measured data on error parameter estimation is reduced so that new data can be added to correct error parameter estimation to enhance the precision of the real-time smart meter error estimation.

Imagine that we traveled into the future where more than 50% of electricity production is generated by wind and solar. Will you wonder how the grid will operate in an overcast, windless, extremely cold/hot day, where half of its generation fleet may become unavailable?

With the development of communication 5g networks and technologies, there is an increasing shortage of spectrum resources. The scarcity of the spectrum resource makes market-driven spectrum auction become an important means of spectrum allocation, and due to the complexity of the network environment, the security of spectrum auctions can not be ignored. Most existing secure spectrum auction schemes introduce a semi-honest agent to complete spectrum auction. However, the hypothetical semi-honest model does not guarantee the security of spectrum auction in the actual application scenario, which may lead to potential security threats: the agent may reveal the privacy of bidders, agent or auctioneer may collude with the bidder to manipulate the spectrum auction, and so on. In this paper, a secure spectrum auction scheme without a trusted party is proposed based on the smart contract technology, and the smart contract written into the blockchain replaces the traditional semi-honest agent to cooperate with auctioneer server to complete the auction. To ensure the security of our scheme, a secure spectrum auction protocol is designed, in which the Software Guard Extensions (SGX) technology and Paillier cryptosystem are used to protect the bidders' bids’ privacy. Public verification is provided in our protocol by using extensive Pedersen commitment, which can prevent the auctioneer server and the bidder from colluding with each other and verify group bid sum values. Finally, the security analysis is given to propose several types of attacks that can be defended. Besides, theoretical analysis and simulation experiments of our protocol are also provided.

This article presents a hardware-in-the-loop (HIL) based modeling approach for simulating impacts of unreliable communication on the performance of centralized volt-var control and for developing an encoding method to mitigate the impacts. First, an asynchronous real-time HIL simulation platform is introduced to enable multi-rate co-simulation of a distribution system with many inverter-based distributed energy resources (DERs). The distribution system is modeled by milliseconds phasor-based models and the DERs are modeled by micro-seconds power electronic models. Communication connections between a centralized volt-var controller (modeled externally to the HIL testbed) and smart inverters are built by implementing Modbus links and the Long Term Evolution network. On this co-simulation platform, an enhanced, augmented Lagrangian multiplier based encoded data recovery (EALM-EDR) algorithm for mitigating the impact of unreliable communication is developed and validated. Simulation results demonstrate the efficacy of using the HIL-based co-simulation platform as a power grid digital twin for developing algorithms that coordinate a large number of heterogeneous control systems through wired and wireless communication links.

State-of-the-art mechanisms against eavesdropping first encrypt all packet payloads in the application layer and then split the packets into multiple network paths. However, versatile eavesdroppers could simultaneously intercept several paths to intercept all the packets, classify the packets into streams using transport fields, and analyze the streams by brute-force. In this article, we propose a programming protocol-independent packet processors (P4)-based network immune scheme (P4NIS) against the intractable eavesdropping. Specifically, P4NIS is equipped with three lines of defenses to provide a softwarized network immunity. Packets are successively processed by the third, second, and first line of defenses. The third line basically encrypts all packet payloads in the application layer using cryptographic mechanisms. Additionally, the second line re-encrypts all packet headers in the transport layer to distribute the packets from one stream into different streams, and disturbs eavesdroppers to classify the packets correctly. Besides, the second line adopts a programmable design for dynamically changing encryption algorithms. Complementally, the first line uses programmable forwarding policies which could split all the double-encrypted packets into different network paths disorderly. Using a paradigm of programmable data planes—P4, we implement P4NIS and evaluate its performances. Experimental results show that P4NIS can increase difficulties of eavesdropping and transmission throughput effectively compared with state-of-the-art mechanisms. Moreover, if P4NIS and state-of-the-art mechanisms have the same level of defending eavesdropping, P4NIS can decrease the encryption cost by 69.85%–81.24%.

Unmanned Aerial Vehicle (UAV) Networks have recently attracted great attention as being able to provide convenient and fast wireless connections. One central question is how to allocate a limited number of UAVs to provide wireless services across a large number of regions, where each region has dynamic arriving flows and flows depart from the system once they receive the desired amount of service (referred to as the flow-level dynamic model). In this article, we propose a MaxWeight-type scheduling algorithm taking into account sharp flow-level dynamics that efficiently redirect UAVs across a large number of regions. However, in our considered model, each flow experiences an independent fading channel and will immediately leave the system once it completes its service, which makes its evolution quite different from the traditional queueing model for wireless networks. This poses significant challenges in our performance analysis. Nevertheless, we incorporate sharp flow-dynamic into the Lyapunov-drift analysis framework, and successfully establish both throughput and heavy-traffic optimality of the proposed algorithm. Extensive simulations are performed to validate the effectiveness of our proposed algorithm.

The prediction of hospital emergency room visits (ERV) for respiratory diseases after the outbreak of PM2.5 is of great importance in terms of public health, medical resource allocation, and policy decision support. Recently, the machine learning methods bring promising solutions for ERV prediction in view of their powerful ability of short-term forecasting, while their performances still exist unknown. Therefore, we aim to check the feasibility of machine learning methods for ERV prediction of respiratory diseases. Three different machine learning models, including autoregressive integrated moving average (ARIMA), multilayer perceptron (MLP), and long short-term memory (LSTM), are introduced to predict daily ERV in urban areas of Beijing, and their performances are evaluated in terms of the mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R2). The results show that the performance of ARIMA is the worst, with a maximum R2 of 0.70 and minimum MAE, RMSE, and MAPE of 99, 124, and 26.56, respectively, while MLP and LSTM perform better, with a maximum R2 of 0.80 (0.78) and corresponding MAE, RMSE, and MAPE of 49 (33), 62 (42), and 14.14 (9.86). In addition, it demonstrates that MLP cannot detect the time lag effect properly, while LSTM does well in the description and prediction of exposure-response relationship between PM2.5 pollution and infecting respiratory disease.

In an islanded microgrid with distributed generators (DGs), on account of the influence of feeder impedance, load and other factors, it is difficult to dispatch output power reasonably by means of droop control, therefore the secondary control is employed to solve this problem. However, the convention methods may fail to achieve the expected effect. In this context, an improved distributed control strategy based on multi-agent consensus algorithm is proposed in this paper. A virtual leader is set to track the balance of power. Through the mutual communication between DGs, it can realize the flexible distribution of load power and has good convergence. By introducing the integral controller on the basis of conventional droop control, the droop curve can move adaptively to maintain frequency and voltage stable. Finally, the effectiveness of the proposed scheme and the performance of anti-communication failure are verified in MATLAB/Simulink simulation platform.

Multicamera or multilight-source is generally used to estimate the 3-D coordinates of the cornea center for 3-D gaze estimation in the existing gaze trackers. Although some model-based methods use one-camera systems to estimate 3-D gazes, which include some user-dependent eye parameters preset by fixed values, a one-camera-one-light-source system is still unable to achieve 3-D gaze estimation by considering individual differences. In this article, an iris feature-based method using one camera and one light source for 3-D gaze estimation is proposed. The iris radius and the kappa angle are determined during user calibration. The optical axis is reconstructed by the iris center and its normal vector from only the iris features. Therefore, with the real-time estimation of kappa angle, the 3-D gaze is estimated by an optimization method. Computer simulations and practical experiments have been performed to analyze the feasibility and robustness of the proposed method. This article’s innovation lies in achieving 3-D gaze estimation based on only one-camera and one-light source with calibrated eye-specific parameters, which breaks through the limitation of the existing methods using a 3-D cornea center on the system’s hardware requirements of 3-D gaze estimation. The simplified hardware system has great application values, especially in the portable mobile devices widely used today. Moreover, a binocular model is used to optimize the results of the point of regard, which improves the accuracy of 3-D gaze estimation effectively.

Comprehensive understanding of the built environment, especially the urban form, is a prerequisite for building a smart city. Data analytics of urban fabric metrics using quantitative methods is critical to understanding a city’s complexity. This paper aims to study urban fabric using comprehensive computation methods. A series of morphological indexes of urban blocks are established to measure the blocks’ overall features and subtle differences. This study uses multiple statistical methods with computation techniques and machine learning to fulfill factor analysis and clustering to classify major block types and their spatial distribution, and this study aims to precisely position the important continuous zone and fracture locations within the study area based on a geo-information system (GIS), effectively revealing the potential morphological order of different block types in the urban fabric. The study provides accurate basis and technical support for the optimization of urban construction. It has important and practical significance for promoting the scientific and reasonable implementation of a new type of urbanization.

This paper presents a novel battery model parameterization method using actual field measurement and manufacturer datasheet for real-time hardware-in-the-loop (HIL) applications. It is critical that real-time HIL models can accurately reproduce field test results so that tests can be conducted on HIL testbeds instead of in the field. In the past, numerical heuristic optimization algorithms were often used to derive parameters for battery models. However, the deterministic algorithms often reach a locally optimal solution and stochastic heuristic searching strategies suffer from low searching efficiency. Therefore, in this paper, we propose a global-local searching enhanced genetic algorithm (GL-SEGA). By applying the generalized opposition-based learning mechanism, GL-SEGA can efficiently explore the global solution space. By using the trust-region-reflective method to perform the local search, the GL-SEGA can improve the accuracy and convergence in its local exploitations. Field measurements and manufactory datasheets are used to test and validate the accuracy and robustness of the GL-SEGA algorithm.

A high level of renewable resource penetration could lead to displacement of conventional synchronous generators from dispatch, and consequently, reduce the system inertia. The decline in the system inertia will increase the primary frequency response (PFR) need requiring to maintain grid frequency stability. This study proposes a new market design to co-optimise energy, inertia and PFR while considering uncertainties in renewable energy productions. A full dynamic model of Electric Reliability Council of Texas (ERCOT) interconnection is used to quantify the PFR requirement for the system inertia. The proposed method explicitly incorporates frequency dynamics and uncertainties in energy production from renewable resources in the scheduling process, which is formulated as a stochastic unit commitment problem. A case study of ERCOT grid demonstrated that the proposed stochastic scheduling of energy, inertia and PFR could yield a more cost-effective solution than the traditional deterministic formulation for the grids with a high penetration of renewable resources.

Cornea radius estimation is a key technique for 3D gaze estimation in the single-camera 3D gaze tracking system. Traditional methods with one-camera-one-light-source systems or one-camera-two-light-source systems cannot achieve 3D gaze estimation. The 3D line-of-sight can be estimated only when the cornea radius is pre-calibrated by the user. A cornea radius calibration method based on the iris radius is proposed in this paper for 3D gaze estimation in remote one-camera-two-light-source systems. We first calibrate the iris radius based on the binocular strategy, estimate the spatial iris center using the calibrated iris radius, and then calibrate the cornea radius by a set of non-linear equations under the constraint of equivalent distances from the cornea center to the iris edge points. The calibrated cornea radius is verified by binocular optimization constraints. Simulations and physical experiments validate the effectiveness of the proposed method. The iris-based cornea radius calibration approach is novel; it can be used to obtain the cornea radius and 3D gaze using remote one-camera-one-light-source or one-camera-multi-light-source systems.

The load profile at a feeder-head is usually known to utility engineers while the nodal load profiles are not. However, the nodal load profiles are increasingly important for conducting time-series analysis in distribution systems. Therefore, in this paper, we present a pivot-point based, two-stage feeder load disaggregation algorithm using smart meter data. The two stages are load profile selection (LPS) and load profile allocation (LPA). In the LPS stage, a random load profile selection process is first executed to meet the load diversity requirement. Then, a few pairs of pivot points are selected as the matching targets. After that, a matching algorithm will run repetitively to select one load profile at a time for matching the reference load profile at the pivot points. In the LPA stage, the LPS selected load profiles are allocated to each load node on the feeder considering distribution transformer loading limits, load composition, and square-footage. The proposed method is validated using actual data collected in a North Carolina service area. Simulation results show that the proposed method can generate a unique load shape for each load node while match the shape of their aggregated profile with the actual feeder head load profile.

In this paper, we present a practical secret key management for data relay security of bluetooth-connected drones. Time-varying received signal strengths between the drones and the ground sensing nodes are quantized to generate the secret key pairs, where the quantization interval is adjusted to reduce the number of mismatched secret key bits. To validate the key management performance, a multihop aerial relay system testbed is developed based on the MX400 drone platform and the bluetooth low energy radio transceiver.

Forest inventory at single-tree level is of great importance to modern forest management. The inventory contains two critical parameters about trees, including their numbers and spatial locations. Traditional methods to catalogue single trees are laborious, while deep neural networks enable to discover the multi-scale features hidden in images and thus make it possible to count trees with remote sensing imagery. In this study, four different tree counting networks, which were constructed by remodeling four different classical deep convolutional neural networks, were evaluated to determine their abilities to grasp the relationship between remote sensing images and tree locations for automatic tree counting end-to-end. To this end, a tree counting dataset was constructed with remote sensing images of 0.8-m spatial resolution in distinct regions. This dataset consisted of 24 GF-II images and the corresponding manually annotated locations of trees based on these images. Thereafter, a large number of experiments were conducted to examine the performance of these networks in regards to tree counting. The results demonstrated that all networks could achieve the competitive performance (above 0.91) in terms of the determination coefficient (R2) between the ground truth and the estimated values. The average accuracy of the Encoder-Decoder Network (one of the four networks) was greater than 91.58% and its R2 was equal to 0.97, achieving the best performance. It has been found that the deep learning is an efficient and effective means for tree counting task.

This paper introduces a supervised machine learning approach to predict and schedule the real-time operation mode of the next operation interval for residential PV/Battery systems controlled by mode-based controllers. The performance of the mode-based economic model-predictive control (EMPC) approach is used as the benchmark. The residential load and PV data used in the paper are 1-minute data downloaded from the the Pecan Street Project website. The optimal operation mode for each control interval is first derived from the historical data used as the training set. Then, four machine learning algorithms (i.e. neural network, support vector machine, logistic regression, and random forest algorithms) are applied. We compared the performance of the four algorithms when using different number of features and length of the training sets extracted from different months of the year. Simulation results show that using the ML approach can effectively improve the performance of the mode-based control system and reduce the computation effort of local controllers because the training can be completed on a cloud-based ML engine. The work presented in this paper paves the way for using a shared-learning platform to design controllers of residential PV/storage systems. This may significantly reduce the cost for implementing such systems.

This paper presents a two-stage hierarchical voltage-load sensitivity matrix (VLSM)-based demand response (DR) algorithm for regulating the voltage at distribution feeders while fulfilling the transmission system DR requests. This allows distributed energy resources to be aggregated to provide transmission-level services without causing operational issues in the distribution grids. At the beginning of each operation interval, a transmission system controller will issue DR commands to distribution system controllers. In the first stage, the distribution system controller uses a VLSM-based dispatch algorithm to dispatch the available DR resources on the distribution feeder with an objective to minimize the total voltage deviations at the lowest cost. Then, power flow studies are conducted assuming that the DR commands have been executed. If voltage violations are detected, a second-stage VLSM-based DR dispatch is performed to remove those violations. After that, the upper and lower DR limits calculated for the next operation interval are sent back to the transmission system controller so the transmission optimization algorithm can use them as operational constraints to make subsequent decisions. The DR resources include smart photovoltaic (PV) inverters that can curtail real power and provide reactive power support, controllable loads and capacitor banks. The IEEE 123-bus test system with 5-min PV and load data is used to evaluate the performance of the algorithm. Simulation results show that the proposed algorithm can fulfill the transmission-level DR requests while maintaining the voltage in the distribution system within limits.

In this paper, we solve the day-ahead scheduling programming problem for commercial building-level consumers with combined time-of-use ($/kWh) and demand ($/kW) pricing plans. Aiming at minimizing the monthly charge, the problem formulation is proposed as a generic algorithm that produces day-ahead building operation schedules. It considers the influences of daily peak load during on-peak hours and daily energy consumption on the monthly charge. The aggregation model of building-level space conditionings is built for scheduling demo. To obtain the near global optimum, a multi sub-swarms particle swarm optimization (MSPSO) is proposed by introducing the ideas of mutation operation, work hierarchy and iterative regrouping into particle swarm optimization (PSO). It improves the population diversity for enhancing the global searching ability and the ability of escaping from local optimum. The PSO comparison shows that MSPSO has better convergence performance with higher stability compared with some classical PSOs. Furthermore, a commercial office-style building with space conditionings is simulated. Using TOU and demand pricing plan from Duke Energy, numerical results demonstrate that the proposed day-ahead scheduling algorithm and the improved MSPSO can reduce the monthly charge by 30% and 17%, respectively.

Direct imaging of three-dimensional microstructure via X-ray diffraction-based techniques gives valuable insight into the crystallographic features that influence materials properties and performance. For instance, X-ray diffraction tomography provides information on grain orientation, position, size, and shape in a bulk specimen. As such techniques become more accessible to researchers, demands are placed on processing the datasets that are inherently “noisy,” multi-dimensional, and multimodal. To fulfill this need, we have developed a one-of-a-kind function package, PolyProc, that is compatible with a range of data shapes, from planar sections to time-evolving and three-dimensional orientation data. Our package comprises functions to import, filter, analyze, and visualize the reconstructed grain maps. To accelerate the computations in our pipeline, we harness computationally efficient approaches: for instance, data alignment is done via genetic optimization; grain tracking through the Hungarian method; and feature-to-feature correlation through k-nearest neighbors algorithm. As a proof-of-concept, we test our approach in characterizing the grain texture, topology, and evolution in a polycrystalline Al–Cu alloy undergoing coarsening.

The deployment of drone small cells has emerged as a promising solution to agile provisioning of Internet backbone access for Internet of Things devices, and many other types of users/devices. In this paper, we consider the problem of deploying a set of drone cells operating on multiple channels in a target area to provide access to the backbone/core network, which is formulated as a combinatorial network utility maximization problem. Since an offline and centralized solution to such a problem is not feasible, a low-complexity and distributed online algorithm is highly desired. Therefore, we propose a measurement-aided dynamic planning (MAD-P) algorithm, where the dispatched drones perform position and channel configurations autonomously on the fly based on the real-time measurement of network throughput to solve the problem in a distributed fashion during flight with minimal centralized control. We prove that the proposed MAD-P algorithm is asymptotically optimal, and investigate how long it takes for the convergence to stationarity under the MAD-P algorithm by giving a mixing time analysis. We also derive an upper bound of the performance gap in presence of measurement errors. Simulation results are provided to validate our analytic results and demonstrate the effectiveness of our algorithm.

Exploration in environments with sparse rewards remains a challenging problem in Deep Reinforcement Learning (DRL). For the off-policy method, it usually needs a large number of training samples. With the growing dimensions of state and action space, this method becomes more and more sample-inefficient. In this paper, we propose a novel fast exploration method for off-policy reinforcement learning, called Density-based Intrinsic Motivation and Self-adaptive Action Noise (DIMSAN). Our main contribution is twofold: (1) We propose a Density-based Intrinsic Motivation (DIM) method. It introduces a new intrinsic-reward generation mechanism based on samples’ density estimation during experience replay and encourages the agent to seek novel and unfamiliar states. (2) We propose a Self-adaptive Action Noise (SAN) to deal with the exploration-exploitation tradeoffs, which could automatically change the exploration step through adding adaptive action space noise. The synergy between DIM and SAN could guide the agent to search the state and action space with high efficiency. We evaluate our method on the benchmark manipulation tasks and the designed challenging ones. Empirical results show that our method outperforms the existing methods in terms of convergence speed and sample efficiency, especially in challenging tasks.

A web application was developed to provide a dynamic feeding program for PIC maternal dam lines during gilt development, gestation, peri-partum, lactation and wean-to-service interval (WSI). These recommendations for each production phase are based on peer-reviewed large-scale commercial research. The tool was developed using the Shiny package of R and includes CSS themes, html widgets, and JavaScript actions. Inputs, include pigs weaned per sow per year (PWSY), farrowing rate (FR), total born (TB), replacement rate, the existing feeding program of the breeding herd and the diet energy and standardized ileal digestible (SID) lysine (Lys) levels. Outputs include the feeding program, nutrient specifications and estimates of economic opportunity and performance improvement. The feeding program is based on user-defined energy and SID Lys levels for the gestation and lactation diets and the PIC nutrient recommendations for the breeding herd. Correspondingly, recommended specifications of other nutrients in the diets are provided and calculated based on the recommended feeding program. The tool provides economic and productivity opportunity analysis by comparing the PIC recommendations and the current user feeding programs. Improvement in PWSY is driven by the energy intake impact on caliper score during breeding and farrowing and consequently to FR and TB. This web application can be used by nutritionists and production managers to compare their current feeding practices to PIC recommendations for highly prolific sows. This will aid in their decision-making process regarding nutrition and feeding programs considering productivity and profitability outcomes.

This paper presents a networked hardware-in-the-loop (HIL) simulation system for modeling large-scale power systems. Researchers have developed many HIL test systems for power systems in recent years. Those test systems can model both microsecond-level dynamic responses of power electronic systems and millisecond-level transients of transmission and distribution grids. By integrating individual HIL test systems into a network of HIL test systems, we can create large-scale power grid digital twins with flexible structures at required modeling resolution that fits for a wide range of system operating conditions. This will not only significantly reduce the need for field tests when developing new technologies but also greatly shorten the model development cycle. In this paper, we present a networked OPAL-RT based HIL test system for developing transmission-distribution coordinative Volt-VAr regulation technologies as an example to illustrate system setups, communication requirements among different HIL simulation systems, and system connection mechanisms. Impacts of communication delays, information exchange cycles, and computing delays are illustrated. Simulation results show that the performance of a networked HIL test system is satisfactory.

Computer vision with state-of-the-art deep learning models has achieved huge success in the field of Optical Character Recognition (OCR) including text detection and recognition tasks recently. However, Key Information Extraction (KIE) from documents as the downstream task of OCR, having a large number of use scenarios in real-world, remains a challenge because documents not only have textual features extracting from OCR systems but also have semantic visual features that are not fully exploited and play a critical role in KIE. Too little work has been devoted to efficiently make full use of both textual and visual features of the documents. In this paper, we introduce PICK, a framework that is effective and robust in handling complex documents layout for KIE by combining graph learning with graph convolution operation, yielding a richer semantic representation containing the textual and visual features and global layout without ambiguity. Extensive experiments on real-world datasets have been conducted to show that our method outperforms baselines methods by significant margins. Our code is available at https://github.com/wenwenyu/PICK-pytorch., Comment: Accepted by ICPR2020. Code at https://github.com/wenwenyu/PICK-pytorch

This paper presents a fast power-setpoint tracking algorithm to enable utility-scale photovoltaic (PV) systems to provide high quality grid services such as power reserves and fast frequency response. The algorithm unites maximum power-point estimation (MPPE) with flexible power-point tracking (FPPT) control to improve the performance of both algorithms, achieving fast and accurate PV power-setpoint tracking even under rapid solar irradiance changes. The MPPE is developed using a real-time, nonlinear curve-fitting approach based on the Levenberg-Marquardt algorithm. A modified adaptive FPPT based on the Perturb and Observe technique is developed for the power-setpoint tracking. By using MPPE to decouple the impact of irradiance changes on the measured PV output power, we develop a fast convergence technique for tracking power-reference changes within three FPPT iterations. Furthermore, to limit the maximum output power ripple, a new design is introduced for the steady-state voltage step size of the adaptive FPPT. The proposed algorithm is implemented on a testbed consisting of a 500 kVA three-phase, single-stage, utility-scale PV system on the OPAL-RT eMEGASIM platform. Results show that the proposed method outperforms the state-of-the-art., Comment: Accepted by IEEE Transactions on Sustainable Energy

Mobile cloud computing (MCC) is embracing rapid development these days and able to provide data outsourcing and sharing services for cloud users with pervasively smart mobile devices. Although these services bring various conveniences, many security concerns such as illegally access and user privacy leakage are inflicted. Aiming to protect the security of cloud data sharing against unauthorized accesses, many studies have been conducted for fine-grained access control using ciphertext-policy attribute-based encryption (CP-ABE). However, a practical and secure data sharing scheme that simultaneously supports fine-grained access control, large university, key escrow free, and privacy protection in MCC with expressive access policy, high efficiency, verifiability, and exculpability on resource-limited mobile devices has not been fully explored yet. Therefore, we investigate the challenge and propose an Efficient and Multiauthority Large Universe Policy-Hiding Data Sharing (EMA-LUPHDS) scheme. In this scheme, we employ fully hidden policy to preserve the user privacy in access policy. To adapt to large scale and distributed MCC environment, we optimize multiauthority CP-ABE to be compatible with large attribute universe. Meanwhile, for the efficiency purpose, online/offline and verifiable outsourced decryption techniques with exculpability are leveraged in our scheme. In the end, we demonstrate the flexibility and high efficiency of our proposal for data sharing in MCC by extensive performance evaluation.

In this paper, we propose an end-to-end trainable text recognition model that consists of an auxiliary augmentation module and a text recognizer. Well-established traditional methods in computer vision (such as binarization, sharpening and morphological operations) play critical roles in image preprocessing. These operations are proven to be particularly useful for downstream computer vision tasks. In order to achieve better results, case-by-case hyperparameter adjustment is often required. Inspired by traditional CV methods, we propose an auxiliary network to mimic traditional CV operations. The auxiliary network acts like an image preprocessing module to extract rich augmented features from the input image to ease the downstream recognition difficulty. We studied three types of augmentation modules with parameters that can be learned directly via gradient back-propagation. This way, our method combines traditional CV techniques and deep neural network by joint learning. The proposed method is extensively tested on major benchmark datasets to show that it can boost the performance of the recognizers, especially for degraded text images in various challenging conditions.

Deep learning-based scene text detection can achieve preferable performance, powered with sufficient labeled training data. However, manual labeling is time consuming and laborious. At the extreme, the corresponding annotated data are unavailable. Exploiting synthetic data is a very promising solution except for domain distribution mismatches between synthetic datasets and real datasets. To address the severe domain distribution mismatch, we propose a synthetic-to-real domain adaptation method for scene text detection, which transfers knowledge from synthetic data (source domain) to real data (target domain). In this paper, a text self-training (TST) method and adversarial text instance alignment (ATA) for domain adaptive scene text detection are introduced. ATA helps the network learn domain-invariant features by training a domain classifier in an adversarial manner. TST diminishes the adverse effects of false positives (FPs) and false negatives (FNs) from inaccurate pseudo-labels. Two components have positive effects on improving the performance of scene text detectors when adapting from synthetic-to-real scenes. We evaluate the proposed method by transferring from SynthText, VISD to ICDAR2015, ICDAR2013. The results demonstrate the effectiveness of the proposed method with up to 10% improvement, which has important exploration significance for domain adaptive scene text detection. Code is available at https://github.com/weijiawu/SyntoReal_STD.

Comparing with raw textual data, information in tabular format is more compact and concise, and easier for comparison, retrieval, and understanding. Furthermore, there are many demands to detect and extract tables from photos in the era of Mobile Internet. However, most of the existing table detection methods are designed for scanned document images or Portable Document Format (PDF). And tables in the real world are seldom collected in the current mainstream table detection datasets. Therefore, we construct a dataset named NTable for camera-based table detection. NTable consists of a smaller-scale dataset NTable-ori, an augmented dataset NTable-cam, and a generated dataset NTable-gen. The experiments demonstrate deep learning methods trained on NTable improve the performance of spotting tables in the real world. We will release the dataset to support the development and evaluation of more advanced methods for table detection and other further applications in the future.

Over the past few years, various word-level textual attack approaches have been proposed to reveal the vulnerability of deep neural networks used in natural language processing. Typically, these approaches involve an important optimization step to determine which substitute to be used for each word in the original input. However, current research on this step is still rather limited, from the perspectives of both problem-understanding and problem-solving. In this paper, we address these issues by uncovering the theoretical properties of the problem and proposing an efficient local search algorithm (LS) to solve it. We establish the first provable approximation guarantee on solving the problem in general cases. Extensive experiments involving 5 NLP tasks, 8 datasets and 26 NLP models show that LS can largely reduce the number of queries usually by an order of magnitude to achieve high attack success rates. Further experiments show that the adversarial examples crafted by LS usually have higher quality, exhibit better transferability, and can bring more robustness improvement to victim models by adversarial training.

This paper presents a control strategy for multi-machine parallel permanent magnet synchronous generator (PMSG) grid-connected system, which combines the advantages of virtual synchronous generators (VSG) and direct current transmission to achieve large-scale wind power integration. The grid-connected inverter adopts the VSG control strategy to simulate the port characteristics of the traditional synchronous generator, which can actively provide the inertia support for the system when the power system fluctuates. Based on the power-DC voltage drooping characteristics, a machine-side DC droop control strategy is constructed. This control strategy can significantly reduce the number of inverters and realize the collection and transmission of DC-type power, thereby saving economic costs and reducing the cost. Many problems caused by the internal AC collection process of wind farms. Finally, the performance and effectiveness of the proposed method are validated by the simulation results.

Change detection for high-resolution remote sensing images is more and more widespread in the application of monitoring the Earth’s surface. However, on the one hand, the ground truth could facilitate the distinction between changed and unchanged areas, but it is hard to acquire them. On the other hand, due to the complexity of remote sensing images, it is difficult to extract features of difference, let alone the construction of the classification model that performs change detection based on the features of difference in each pixel pair. Aiming at these challenges, this paper proposes a weakly supervised change detection method based on edge mapping and Stacked Denoising Auto-Encoders (SDAE) network called EM-SDAE. We analyze the difference in edge maps of bi-temporal remote sensing images to acquire part of the ground truth at a relatively low cost. Moreover, we design a neural network based on SDAE with a deep structure, which extracts the features of difference so as to efficiently classify changed and unchanged regions after being trained with the ground truth. In our experiments, three real sets of high-resolution remote sensing images are employed to validate the high efficiency of our proposed method. The results show that accuracy can even reach up to 91.18% with our method. In particular, compared with the state-of-the-art work (e.g., IR-MAD, PCA-k-means, CaffeNet, USFA, and DSFA), it improves the Kappa coefficient by 27.19% on average.

Exploration in environments with sparse feedback remains a challenging research problem in reinforcement learning (RL). When the RL agent explores the environment randomly, it results in low exploration efficiency, especially in robotic manipulation tasks with high dimensional continuous state and action space. In this paper, we propose a novel method, called Augmented Curiosity-Driven Experience Replay (ACDER), which leverages (i) a new goal-oriented curiosity-driven exploration to encourage the agent to pursue novel and task-relevant states more purposefully and (ii) the dynamic initial states selection as an automatic exploratory curriculum to further improve the sample-efficiency. Our approach complements Hindsight Experience Replay (HER) by introducing a new way to pursue valuable states. Experiments conducted on four challenging robotic manipulation tasks with binary rewards, including Reach, Push, Pick&Place and Multi-step Push. The empirical results show that our proposed method significantly outperforms existing methods in the first three basic tasks and also achieves satisfactory performance in multi-step robotic task learning.

Robotic grasping in unstructured dense clutter remains a challenging task and has always been a key research direction in the field of robotics. In this paper, we propose a novel robotic grasping system that could use the synergies between pushing and grasping actions to automatically grasp the objects in dense clutter. Our method involves using fully convolutional action-value functions (FCAVF) to map from visual observations to two action-value tables in a Q-learning framework. These two value tables infer the utility of pushing and grasping actions, and the highest value with the corresponding location and orientation means the best place to execute action for the end effector. For better grasping, we introduce an active pushing mechanism based on a new metric, called Dispersion Degree, which describes how spread out the objects are in the environment. Then we design a coordination mechanism to apply the synergies of different actions based on the action-values and dispersion degree of the objects and make the grasps more effective. Experimental results show that our proposed robotic grasping system can greatly improve the robotic grasping success rate in dense clutter and also has the capability to be generalized to the new scenarios.

Unmanned aerial vehicles (UAVs) can be used as air base stations to provide fast wireless connections for ground users. Due to their constraints on both mobility and energy consumption, a key problem is how to deploy UAVs adaptively in a geographic area with changing traffic demand of mobile users, while meeting the aforemetioned constraints. In this paper, we propose an adaptive deployment strategy for UAV-aided networks based on hybrid deep reinforcement learning, where a UAV can adjust its movement direction and distance to serve users who move randomly in the target area. Through hybrid deep reinforcement learning, UAVs can be trained offline to obtain the global state information and learn a completely distributed control strategy, with which each UAV only needs to take actions based on its observed state in the real deployment to be fully adaptive. Moreover, in order to improve the speed and effect of learning, we improve hybrid reinforcement learning, by adding genetic algorithms and TD-error-based resampling optimization mechanism. Simulation results show that the hybrid deep reinforcement learning algorithm has better efficiency and robustness in multi-UAV control, and has better performance in terms of coverage, energy consumption and average throughput, by which average throughput can be increased by 20% to 60%.

With the ubiquitous deployment of mobile sensors and smart devices, the scope of Internet of things (IoT) has extended to the space of mobile networks, where IoT terminals are moving around instead of being fixed in buildings, ground infrastructures, etc. In this paper, we consider such mobile Internet of things (MIoT), and propose an autonomous rate control (RC) scheme for the uplink transmission from MIoT terminals to access stations. A deep reinforcement learning (DRL) based approach is designed to capture the channel variations of the link and to improve the effectiveness of the rate selection for each egress frame. Extensive simulations are conducted for MIoT terminals including vehicles and UAVs and show significant throughput performance improvement comparing with traditional methods, as well as the robustness and scalability of the DRL-RC algorithm. The proposed DRL-RC can provide inspirations for efficient and scalable link adaptation schemes for MIoT terminals.

This paper presents a microgrid power flow control for islanded microgrids consisting of photovoltaic (PV) system, battery energy storage system (BESS) and diesel generator. The proposed control ensures automatic coordination among the microgrid units. To achieve the coordination, the local control of PV system and BESS are configured to regulate power flow without relying on communications. The generator is controlled as the back-up source and its connection depends on battery state of charge (SOC). A state machine with hysteresis settings is used to trigger smooth transition among operation modes. Not only does the proposed controller consider battery SOC and power limits, but also integrates three-stage charging control as battery charging mechanism, which is not considered in the existing literature. The proposed control is validated for an islanded hybrid microgrid using an OPAL-RT real-time simulator.

In China, the government and the cigarette industry yearly lose millions in sales and tax revenue because of imitation cigarettes. Usually, visual observation is not enough to identify counterfeiting. An auxiliary analytical method is needed for cigarette brands identification. To this end, we developed a portable, low-cost electronic nose (e-nose) system for brand recognition of cigarettes. A gas sampling device was designed to reduce the influence caused by humidity fluctuation and the volatile organic compounds (VOCs) in the environment. To ensure the uniformity of airflow distribution, the structure of the sensing chamber was optimized by computational fluid dynamics (CFD) simulations. The e-nose system is compact, portable, and lightweight with only 15 cm in side length and the cost of the whole device is less than $100. Results from the machine learning algorithm showed that there were significant differences between 5 kinds of cigarettes we tested. Random Forest (RF) has the best performance with accuracy of 91.67% and K Nearest Neighbor (KNN) has the accuracy of 86.98%, which indicated that the e-nose was able to discriminate samples. We believe this portable, cheap, reliable e-nose system could be used as an auxiliary screen technique for counterfeit cigarettes.

Due to improving computational capacity of supercomputers, transmitting encrypted packets via one single network path is vulnerable to brute-force attacks. The versatile attackers secretly eavesdrop all the packets, classify packets into different streams, performs an exhaustive search for the decryption key, and extract sensitive personal information from the streams. However, new Internet Protocol (IP) brings great opportunities and challenges for preventing eavesdropping attacks. In this paper, we propose a Programming Protocol-independent Packet Processors (P4) based Network Immune Scheme (P4NIS) against the eavesdropping attacks. Specifically, P4NIS is equipped with three lines of defense to improve the network immunity. The first line is promiscuous forwarding by splitting all the traffic packets in different network paths disorderly. Complementally, the second line encrypts transmission port fields of the packets using diverse encryption algorithms. The encryption could distribute traffic packets from one stream into different streams, and disturb eavesdroppers to classify them correctly. Besides, P4NIS inherits the advantages from the existing encryption-based countermeasures which is the third line of defense. Using a paradigm of programmable data planes—P4, we implement P4NIS and evaluate its performances. Experimental results show that P4NIS can increase difficulties of eavesdropping significantly, and increase transmission throughput by 31.7% compared with state-of-the-art mechanisms.

In this paper, a multi-hop airborne system is built based on Bluetooth 5 connected autonomous drones to relay real-time data of Internet of Things (IoT). A new lightweight Onboard Bluetooth Transceiver (OBT) is developed for reliable drone-to-drone and drone-to-ground communications. A graphical user interface is presented to monitor real-time flight trajectory of the drones and end-to-end data delivery. Outdoor experiments are conducted in real world to test autonomous flight control of the drones and received signal strength of the OBT communications.

Forced oscillations are caused by sustained cyclic disturbances. This paper presents a machine learning (ML) based time-series classification method that uses the synchrophasor measurements to locate the sources of forced oscillations for fast disturbance removal. Sequential feature selection is used to identify the most informative measurements of each power plant so that multivariate time series (MTS) can be constructed. By training the Mahalanobis matrix, we measure and compare the distance between the MTSs. Templates for representing each class is constructed to reduce the size of training datasets and improve the online matching efficiency. Dynamic time warping (DTW) algorithm is used to align the out-of-sync MTSs to account for oscillation detection errors. The algorithm is validated on two test systems: the IEEE 39-bus system and the WECC 179-bus system. When a forced oscillation occurs, MTSs will be constructed by designated PMU measurements. Then, the MTSs will be classified by the trained classifiers, the class membership of which corresponds to the location of each oscillation source. Simulation results show that the proposed method can be used online to identify the forced oscillation sources with high accuracy. The robustness of the proposed algorithm in the presence of oscillation detection errors is also quantified., submitted to IEEE Trans. on Smart Grid