To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervis...To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervised masked contrastive learning and domain adaptation(SSMCL-DA)method for gearbox fault diagnosis under variable conditions.Initially,during the unsupervised pre-training phase,a dual signal augmentation strategy is devised,which simultaneously applies random masking in the time domain and random scaling in the frequency domain to unlabeled samples,thereby constructing more challenging positive sample pairs to guide the encoder in learning intrinsic features robust to condition variations.Subsequently,a ConvNeXt-Transformer hybrid architecture is employed,integrating the superior local detail modeling capacity of ConvNeXt with the robust global perception capability of Transformer to enhance feature extraction in complex scenarios.Thereafter,a contrastive learning model is constructed with the optimization objective of maximizing feature similarity across different masked instances of the same sample,enabling the extraction of consistent features from multiple masked perspectives and reducing reliance on labeled data.In the final supervised fine-tuning phase,a multi-scale attention mechanism is incorporated for feature rectification,and a domain adaptation module combining Local Maximum Mean Discrepancy(LMMD)with adversarial learning is proposed.This module embodies a dual mechanism:LMMD facilitates fine-grained class-conditional alignment,compelling features of identical fault classes to converge across varying conditions,while the domain discriminator utilizes adversarial training to guide the feature extractor toward learning domain-invariant features.Working in concert,they markedly diminish feature distribution discrepancies induced by changes in load,rotational speed,and other factors,thereby boosting the model’s adaptability to cross-condition scenarios.Experimental evaluations on the WT planetary gearbox dataset and the Case Western Reserve University(CWRU)bearing dataset demonstrate that the SSMCL-DA model effectively identifies multiple fault classes in gearboxes,with diagnostic performance substantially surpassing that of conventional methods.Under cross-condition scenarios,the model attains fault diagnosis accuracies of 99.21%for the WT planetary gearbox and 99.86%for the bearings,respectively.Furthermore,the model exhibits stable generalization capability in cross-device settings.展开更多
This study addresses the risk of privacy leakage during the transmission and sharing of multimodal data in smart grid substations by proposing a three-tier privacy-preserving architecture based on asynchronous federat...This study addresses the risk of privacy leakage during the transmission and sharing of multimodal data in smart grid substations by proposing a three-tier privacy-preserving architecture based on asynchronous federated learning.The framework integrates blockchain technology,the InterPlanetary File System(IPFS)for distributed storage,and a dynamic differential privacy mechanism to achieve collaborative security across the storage,service,and federated coordination layers.It accommodates both multimodal data classification and object detection tasks,enabling the identification and localization of key targets and abnormal behaviors in substation scenarios while ensuring privacy protection.This effectively mitigates the single-point failures and model leakage issues inherent in centralized architectures.A dynamically adjustable differential privacy mechanism is introduced to allocate privacy budgets according to client contribution levels and upload frequencies,achieving a personalized balance between model performance and privacy protection.Multi-dimensional experimental evaluations,including classification accuracy,F1-score,encryption latency,and aggregation latency,verify the security and efficiency of the proposed architecture.The improved CNN model achieves 72.34%accuracy and an F1-score of 0.72 in object detection and classification tasks on infrared surveillance imagery,effectively identifying typical risk events such as not wearing safety helmets and unauthorized intrusion,while maintaining an aggregation latency of only 1.58 s and a query latency of 80.79 ms.Compared with traditional static differential privacy and centralized approaches,the proposed method demonstrates significant advantages in accuracy,latency,and security,providing a new technical paradigm for efficient,secure data sharing,object detection,and privacy preservation in smart grid substations.展开更多
In federated learning,backdoor attacks have become an important research topic with their wide application in processing sensitive datasets.Since federated learning detects or modifies local models through defense mec...In federated learning,backdoor attacks have become an important research topic with their wide application in processing sensitive datasets.Since federated learning detects or modifies local models through defense mechanisms during aggregation,it is difficult to conduct effective backdoor attacks.In addition,existing backdoor attack methods are faced with challenges,such as low backdoor accuracy,poor ability to evade anomaly detection,and unstable model training.To address these challenges,a method called adaptive simulation backdoor attack(ASBA)is proposed.Specifically,ASBA improves the stability of model training by manipulating the local training process and using an adaptive mechanism,the ability of the malicious model to evade anomaly detection by combing large simulation training and clipping,and the backdoor accuracy by introducing a stimulus model to amplify the impact of the backdoor in the global model.Extensive comparative experiments under five advanced defense scenarios show that ASBA can effectively evade anomaly detection and achieve high backdoor accuracy in the global model.Furthermore,it exhibits excellent stability and effectiveness after multiple rounds of attacks,outperforming state-of-the-art backdoor attack methods.展开更多
Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global...Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global model through compromised updates,posing significant threats to model integrity and becoming a key focus in FL security.Existing backdoor attack methods typically embed triggers directly into original images and consider only data heterogeneity,resulting in limited stealth and adaptability.To address the heterogeneity of malicious client devices,this paper proposes a novel backdoor attack method named Capability-Adaptive Shadow Backdoor Attack(CASBA).By incorporating measurements of clients’computational and communication capabilities,CASBA employs a dynamic hierarchical attack strategy that adaptively aligns attack intensity with available resources.Furthermore,an improved deep convolutional generative adversarial network(DCGAN)is integrated into the attack pipeline to embed triggers without modifying original data,significantly enhancing stealthiness.Comparative experiments with Shadow Backdoor Attack(SBA)across multiple scenarios demonstrate that CASBA dynamically adjusts resource consumption based on device capabilities,reducing average memory usage per iteration by 5.8%.CASBA improves resource efficiency while keeping the drop in attack success rate within 3%.Additionally,the effectiveness of CASBA against three robust FL algorithms is also validated.展开更多
While reinforcement learning-based underwater acoustic adaptive modulation shows promise for enabling environment-adaptive communication as supported by extensive simulation-based research,its practical performance re...While reinforcement learning-based underwater acoustic adaptive modulation shows promise for enabling environment-adaptive communication as supported by extensive simulation-based research,its practical performance remains underexplored in field investigations.To evaluate the practical applicability of this emerging technique in adverse shallow sea channels,a field experiment was conducted using three communication modes:orthogonal frequency division multiplexing(OFDM),M-ary frequency-shift keying(MFSK),and direct sequence spread spectrum(DSSS)for reinforcement learning-driven adaptive modulation.Specifically,a Q-learning method is used to select the optimal modulation mode according to the channel quality quantified by signal-to-noise ratio,multipath spread length,and Doppler frequency offset.Experimental results demonstrate that the reinforcement learning-based adaptive modulation scheme outperformed fixed threshold detection in terms of total throughput and average bit error rate,surpassing conventional adaptive modulation strategies.展开更多
In recent years,researchers have leveraged single-agent reinforcement learning to boost educational outcomes and deliver personalized interventions;yet this paradigm provides no capacity for inter-agent interaction.Mu...In recent years,researchers have leveraged single-agent reinforcement learning to boost educational outcomes and deliver personalized interventions;yet this paradigm provides no capacity for inter-agent interaction.Multi-agent reinforcement learning(MARL)overcomes this limitation by allowing several agents to learn simultaneously within a shared environment,each choosing actions that maximize its own or the group's rewards.By explicitly modeling and exploiting agent-to-agent dynamics,MARL can align those interactions with pedagogical goals such as peer tutoring,collaborative problem-solving,or gamified competition,thus opening richer avenues for adaptive and socially informed learning experiences.This survey investigates the impact of MARL on educational outcomes by examining evidence of its effectiveness in enhancing learner performance,engagement,equity,and reducing teacher workload compared to single agent or traditional approaches.It explores the educational domains and pedagogical problems addressed by MARL,identifies the algorithmic families used,and analyzes their influence on learning.The review also assesses experimental settings and evaluation metrics to determine ecological validity,and outlines current challenges and future research directions in applying MARL to education.展开更多
Designing appropriate loss functions is critical to the success of supervised learning models.However,most conventional losses are fixed and manually designed,making them suboptimal for diverse and dynamic learning sc...Designing appropriate loss functions is critical to the success of supervised learning models.However,most conventional losses are fixed and manually designed,making them suboptimal for diverse and dynamic learning scenarios.In this work,we propose an Adaptive Meta-Loss Network(Adaptive-MLN)that learns to generate taskagnostic loss functions tailored to evolving classification problems.Unlike traditional methods that rely on static objectives,Adaptive-MLN treats the loss function itself as a trainable component,parameterized by a shallow neural network.To enable flexible,gradient-free optimization,we introduce a hybrid evolutionary approach that combines GeneticAlgorithms(GA)for global exploration and Evolution Strategies(ES)for local refinement.This co-evolutionary process dynamically adjusts the loss landscape,improvingmodel generalization without relying on analytic gradients or handcrafted heuristics.Experimental evaluations on synthetic tasks and the CIFAR-10 andMNIST datasets demonstrate that our approach consistently outperforms standard losses such as Cross-Entropy and Mean Squared Error in terms of accuracy,convergence,and adaptability.展开更多
With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy...With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy.However,efficient client selection and adaptive weight allocation in heterogeneous and non-IID environments remain challenging.To address these issues,we propose Federated Learning with Client Selection and Adaptive Weighting(FedCW),a novel algorithm that leverages adaptive client selection and dynamic weight allocation for optimizing model convergence in real-time vehicular networks.FedCW selects clients based on their Euclidean distance from the global model and dynamically adjusts aggregation weights to optimize both data diversity and model convergence.Experimental results show that FedCW significantly outperforms existing FL algorithms such as FedAvg,FedProx,and SCAFFOLD,particularly in non-IID settings,achieving faster convergence,higher accuracy,and reduced communication overhead.These findings demonstrate that FedCW provides an effective solution for enhancing the performance of FL in heterogeneous,edge-based computing environments.展开更多
Agile lithology identification can assist mining by providing important information in the exploration and production of mineral resources.This study proposes a new lithology recognition procedure using video-logging ...Agile lithology identification can assist mining by providing important information in the exploration and production of mineral resources.This study proposes a new lithology recognition procedure using video-logging of boreholes with an endoscope,applied to six production blocks in a limestone quarry.Images are automatically extracted from the videos and the lithology is classified into three classes based on clay content,i.e.massive limestone,brecciated limestone,and high amount of clay.The image quality is evaluated with a gray pixel intensity threshold and three no-reference image quality metrics,i.e.perception-based image quality evaluator,natural image quality evaluator,and blind/referenceless image spatial quality evaluator.After removing low-quality images,7583 images are retained and used for developing lithology classification models using six optimized classification techniques.The contrast-limited adaptive histogram equalization(CLAHE)technique is used to improve image quality.Ten color characteristics involving three percentiles of red,green and blue pixel intensities,together with color counting and five texture characteristics-correlation,entropy,homogeneity,contrast and energy-are used as inputs.Bayesian optimized light gradient boosting machine model performs best,with an overall accuracy of 88.04%,and a precision on the classes of massive limestone,brecciated limestone and high amount of clay of 90.72%,83.52%and 85.29%,respectively,for the testing set.The feature importance scores show that the color counting is the most significant parameter for the development of the classification model.Compared with previous image-based methodologies,this study provides a more flexible and cheaper procedure to identify lithology.展开更多
Accurate temperature control and effective oxide removal are essential for achieving high-quality epitaxial growth in molecular beam epitaxy(MBE).However,traditional methods often rely on manual identification of refl...Accurate temperature control and effective oxide removal are essential for achieving high-quality epitaxial growth in molecular beam epitaxy(MBE).However,traditional methods often rely on manual identification of reflection high-energy electron diffraction(RHEED)patterns.This process is heavily influenced by the grower’s experience,leading to issues with reproducibility and limiting the potential for automation.In this report,we propose an unsupervised learning framework for realtime RHEED analysis during the deoxidation process.By incorporating temporal similarity constraints into contrastive learning,our model generates smooth and interpretable feature trajectories that illustrate transitions in the deoxidation state,thus eliminating the need for manual labeling.The model,pre-trained using grouped contrastive loss,shows significant improvement in RHEED feature boundary discrimination and localization of critical regions.We evaluated its generalizability through two transfer learning strategies:calibration-free clustering and few-shot fine-tuning.The pre-trained model achieved a clustering accuracy of 88.1%for GaAs deoxidation samples without additional labels and reached an accuracy of 94.3%to 95.5%after fine-tuning with just five sample pairs across GaAs,Ge,and InAs substrates.This framework is optimized for resource-constrained edge devices,allowing for real-time,plug-and-play integration with existing MBE systems and swift adaptation across various materials and equipment.This work paves the way for greater automation and improved reproducibility in semiconductor manufacturing.展开更多
This paper investigates the adaptive optimal tracking control(AOTC)for underactuated surface vessels(USVs).Compared to the majority of existing studies,the control strategy in this paper innovatively combines an exten...This paper investigates the adaptive optimal tracking control(AOTC)for underactuated surface vessels(USVs).Compared to the majority of existing studies,the control strategy in this paper innovatively combines an extended state observer(ESO)with reinforcement learning(RL).The designed ESO has high estimation accuracy and robust disturbance rejection capabilities for the unmeasurable information for USVs.To obtain the AOTC,the actor–critic(AC)networks based on RL are constructed to solve the Hamilton–Jacobi–Bellman(HJB)equations.Due to the uncertainties,it is challenging to obtain the optimal controller by directly solving the HJB equations.To address this issue,this paper employs neural networks(NNs)to approximate the uncertainties and solves the optimal controller via AC-RL and ESO.In addition,the adaptive parameters of the optimal controller is trained in parallel with AC networks,which can ensure that the trained networks can further improve tracking performance.The boundedness of AOTC for USVs is shown by Lyapunov stability theorem.Finally,simulation results demonstrate the effectiveness of the proposed algorithm.展开更多
As joint operations have become a key trend in modern military development,unmanned aerial vehicles(UAVs)play an increasingly important role in enhancing the intelligence and responsiveness of combat systems.However,t...As joint operations have become a key trend in modern military development,unmanned aerial vehicles(UAVs)play an increasingly important role in enhancing the intelligence and responsiveness of combat systems.However,the heterogeneity of aircraft,partial observability,and dynamic uncertainty in operational airspace pose significant challenges to autonomous collision avoidance using traditional methods.To address these issues,this paper proposes an adaptive collision avoidance approach for UAVs based on deep reinforcement learning.First,a unified uncertainty model incorporating dynamic wind fields is constructed to capture the complexity of joint operational environments.Then,to effectively handle the heterogeneity between manned and unmanned aircraft and the limitations of dynamic observations,a sector-based partial observation mechanism is designed.A Dynamic Threat Prioritization Assessment algorithm is also proposed to evaluate potential collision threats from multiple dimensions,including time to closest approach,minimum separation distance,and aircraft type.Furthermore,a Hierarchical Prioritized Experience Replay(HPER)mechanism is introduced,which classifies experience samples into high,medium,and low priority levels to preferentially sample critical experiences,thereby improving learning efficiency and accelerating policy convergence.Simulation results show that the proposed HPER-D3QN algorithm outperforms existing methods in terms of learning speed,environmental adaptability,and robustness,significantly enhancing collision avoidance performance and convergence rate.Finally,transfer experiments on a high-fidelity battlefield airspace simulation platform validate the proposed method's deployment potential and practical applicability in complex,real-world joint operational scenarios.展开更多
To address the challenge of achieving decentralized,scalable,and adaptive control for large-scale multiple unmanned aerial vehicle(multi-UAV)swarms in dynamic urban environments with obstacles and wind perturbations,w...To address the challenge of achieving decentralized,scalable,and adaptive control for large-scale multiple unmanned aerial vehicle(multi-UAV)swarms in dynamic urban environments with obstacles and wind perturbations,we proposed a hybrid framework integrating adaptive reinforcement learning(RL),multi-modal perception fusion,and enhanced pigeon flock optimization(PFO)with curiosity-driven exploration to enable robust autonomous and formation control.The framework leverages meta-learning to optimize RL policies for real-time adaptation,fuses sensor data for precise state estimation,and enhances PFO with learned leader-follower dynamics and exploration rewards to maintain cohesive formations and explore uncertain areas.For swarms of 10–30 UAVs,it achieves 34%faster convergence,61%reduced stability root mean square error(RMSE),88%fewer collisions and 85.6%–92.3%success rates in target detection and encirclement,outperforming standard multi-agent RL,pure PFO,and single-modality RL.Three-dimensional trajectory visualizations confirm cohesive formations,collision-free maneuvers,and efficient exploration in urban search-and-rescue scenarios.Innovations include meta-RL for rapid adaptation,multi-modal fusion for robust perception,and curiosity-driven PFO for scalable,decentralized control,advancing real-world multi-UAV swarm autonomy and coordination.展开更多
Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the pro...Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the proposed wearable wristband with selfsupervised contrastive learning excels at dynamic motion tracking and adapts rapidly across multiple scenarios.It features a four-channel sensing array composed of an ionic hydrogel with hierarchical microcone structures and ultrathin flexible electrodes,resulting in high-sensitivity capacitance output.Through wireless transmission from a Wi-Fi module,the proposed algorithm learns latent features from the unlabeled signals of random wrist movements.Remarkably,only few-shot labeled data are sufficient for fine-tuning the model,enabling rapid adaptation to various tasks.The system achieves a high accuracy of 94.9%in different scenarios,including the prediction of eight-direction commands,and air-writing of all numbers and letters.The proposed method facilitates smooth transitions between multiple tasks without the need for modifying the structure or undergoing extensive task-specific training.Its utility has been further extended to enhance human–machine interaction over digital platforms,such as game controls,calculators,and three-language login systems,offering users a natural and intuitive way of communication.展开更多
With increasing density and heterogeneity in unlicensed wireless networks,traditional MAC protocols,such as Carrier Sense Multiple Access with Collision Avoidance(CSMA/CA)in Wi-Fi networks,are experiencing performance...With increasing density and heterogeneity in unlicensed wireless networks,traditional MAC protocols,such as Carrier Sense Multiple Access with Collision Avoidance(CSMA/CA)in Wi-Fi networks,are experiencing performance degradation.This is manifested in increased collisions and extended backoff times,leading to diminished spectrum efficiency and protocol coordination.Addressing these issues,this paper proposes a deep-learning-based MAC paradigm,dubbed DL-MAC,which leverages spectrum data readily available from energy detection modules in wireless devices to achieve the MAC functionalities of channel access,rate adaptation,and channel switch.First,we utilize DL-MAC to realize a joint design of channel access and rate adaptation.Subsequently,we integrate the capability of channel switching into DL-MAC,enhancing its functionality from single-channel to multi-channel operations.Specifically,the DL-MAC protocol incorporates a Deep Neural Network(DNN)for channel selection and a Recurrent Neural Network(RNN)for the joint design of channel access and rate adaptation.We conducted real-world data collection within the 2.4 GHz frequency band to validate the effectiveness of DL-MAC.Experimental results demonstrate that DL-MAC exhibits significantly superior performance compared to traditional algorithms in both single and multi-channel environments,and also outperforms single-function designs.Additionally,the performance of DL-MAC remains robust,unaffected by channel switch overheads within the evaluation range.展开更多
Machine learning methods are widely used to evaluate the risk of small-and mediumsized enterprises(SMEs)in supply chain finance(SCF).However,there may be problems with data scarcity,feature redundancy,and poor predict...Machine learning methods are widely used to evaluate the risk of small-and mediumsized enterprises(SMEs)in supply chain finance(SCF).However,there may be problems with data scarcity,feature redundancy,and poor predictive performance.Additionally,data collected over a long time span may cause differences in the data distribution,and classic supervised learning methods may exhibit poor predictive abilities under such conditions.To address these issues,a domain-adaptation-based multistage ensemble learning paradigm(DAMEL)is proposed in this study to evaluate the credit risk of SMEs in SCF.In this methodology,a bagging resampling algorithm is first used to generate a dataset to address data scarcity.Subsequently,a random subspace is applied to integrate various features and reduce feature redundancy.Additionally,a domain adaptation approach is utilized to reduce the data distribution discrepancy in the cross-domain.Finally,dynamic model selection is developed to improve the generalization ability of the model in the fourth stage.A real-world credit dataset from the Chinese securities market was used to validate the effectiveness and feasibility of the multistage ensemble learning paradigm.The experimental results demonstrated that the proposed domain-adaptation-based multistage ensemble learning paradigm is superior to principal component analysis,joint distribution adaptation,random forest,and other ensemble and transfer learning methods.Moreover,dynamic model selection can improve the model generalization performance and prediction precision of minority samples.This can be considered a promising solution for evaluating the credit risk of SMEs in SCF for financial institutions.展开更多
This paper proposes virtual impedance adaptation of the lower-limb exoskeleton for human performance augmentation(LEHPA) based on deep reinforcement learning(VIADRL) to mitigate reliance on model accuracy and address ...This paper proposes virtual impedance adaptation of the lower-limb exoskeleton for human performance augmentation(LEHPA) based on deep reinforcement learning(VIADRL) to mitigate reliance on model accuracy and address the ever-changing human-exoskeleton interaction(HEI) dynamics. The classical sensitivity amplification control strategy is expanded to the virtual impedance control strategy with more learnable virtual impedance parameters. The adjustment of these virtual impedance parameters is formalized as finding the optimal policy for a Markov Decision Process and can then be effectively resolved using deep reinforcement learning algorithms. To ensure safe and efficient policy training, a multibody simulation environment is established to facilitate the training process, supplemented by the innovative hybrid inverse-forward dynamics simulation approach for executing the simulation. For comparison purposes, the SADRL strategy is introduced as a benchmark. A novel control performance evaluation method based on the HEI forces at the back, thighs, and shanks is proposed to quantitatively evaluate the performance of our proposed VIADRL strategy. The VIADRL controller is systematically compared with the SADRL controller at five selected walking speeds. The lumped ratio of HEI forces under the SADRL strategy relative to those under the SADRL strategy is as low as 0.81 in simulation and approximately 0.89 on the LEHPA prototype. The overall reduction of HEI forces demonstrates the superiority of the VIADRL strategy in comparison to the SADRL strategy.展开更多
The environment has an important impact on maize(Zea mays L.)production,making it necessary to identify plant adaptation regions that are suitable for different maize varieties.Traditional methods using field trials a...The environment has an important impact on maize(Zea mays L.)production,making it necessary to identify plant adaptation regions that are suitable for different maize varieties.Traditional methods using field trials are costly and restricted to a limited number of areas.Identifying adaptation regions based on climate data has great potential,but a basic understanding and a prediction approach for diverse maize varieties are lacking.Here,we collected a representative dataset comprising 32,840 data points from the National Maize Variety Trial Data Management Platform.We employed three traits to characterize the adaptability of different maize varieties:PH(plant height),DTS(days to silking),and yield.First,we quantified the contributions of variety(V),environment(E),and V×E to variance in the three adaptationrelated traits.The mean contributions of E to variance in PH,DTS,and yield were 54.50%,82.87%,and 75.92%,respectively,suggesting that environmental effects are crucial for phenotype construction.Second,we analyzed correlations between the three traits and three environmental indices:GDD(growing degree days),PRE(precipitation),and SSD(sunshine duration).The highest absolute correlation coefficients between phenotypes and environmental indices were 0.15–0.69 at the whole-data level.To predict variety adaptation on a national scale,we modeled the three traits using environmental indices and best linear unbiased predictors(BLUPs)via the random forest algorithm.The predictive abilities of our models for PH,DTS,and yield were 0.90(MAE=9.95 cm),0.99(MAE=1.09 d),and 0.95(MAE=0.55 t ha^(−1)),respectively,indicating that our proposed framework can predict adaptationrelated traits for diverse maize varieties in China.展开更多
Autonomous legged robots,capable of navigating uneven terrain,can perform a diverse array of tasks.However,designing locomotion controllers remains challenging.In particular,designing a controller based on durable and...Autonomous legged robots,capable of navigating uneven terrain,can perform a diverse array of tasks.However,designing locomotion controllers remains challenging.In particular,designing a controller based on durable and reliable proprioceptive sensors,is essential for achieving adaptability.Presently,the controller must either be manually designed for specific robots and tasks,or developed using machine-learning techniques,which require extensive training time and result in complex controllers.Inspired by animal locomotion,we propose a simple yet comprehensive closed-loop modular framework that utilizes minimal proprioceptive feedback(i.e.,the Coxa-Femur(CF)joint angle),enabling a quadruped robot to efficiently navigate unpredictable and uneven terrains,including the step and slope.The framework comprises a basic neural control network capable of rapidly learning optimized motor patterns,and a straightforward module for sensory feedback sharing and integration.In a series of experiments,we show that integrating sensory feedback into the base neural control network aids the robot in continually learning robust motor patterns on flat,step,and slope terrain,compared with the open-loop base framework.Sharing sensory feedback information across the four legs enables a quadruped robot to proactively navigate unpredictable steps with minimal interaction.Furthermore,the controller remains functional even in the absence of sensor signals.This control configuration was successfully transferred to a physical robot without any modifications.展开更多
Low-carbon smart parks achieve selfbalanced carbon emission and absorption through the cooperative scheduling of direct current(DC)-based distributed photovoltaic,energy storage units,and loads.Direct current power li...Low-carbon smart parks achieve selfbalanced carbon emission and absorption through the cooperative scheduling of direct current(DC)-based distributed photovoltaic,energy storage units,and loads.Direct current power line communication(DC-PLC)enables real-time data transmission on DC power lines.With traffic adaptation,DC-PLC can be integrated with other complementary media such as 5G to reduce transmission delay and improve reliability.However,traffic adaptation for DC-PLC and 5G integration still faces the challenges such as coupling between traffic admission control and traffic partition,dimensionality curse,and the ignorance of extreme event occurrence.To address these challenges,we propose a deep reinforcement learning(DRL)-based delay sensitive and reliable traffic adaptation algorithm(DSRTA)to minimize the total queuing delay under the constraints of traffic admission control,queuing delay,and extreme events occurrence probability.DSRTA jointly optimizes traffic admission control and traffic partition,and enables learning-based intelligent traffic adaptation.The long-term constraints are incorporated into both state and bound of drift-pluspenalty to achieve delay awareness and enforce reliability guarantee.Simulation results show that DSRTA has lower queuing delay and more reliable quality of service(QoS)guarantee than other state-of-the-art algorithms.展开更多
基金supported by the National Natural Science Foundation of China Funded Project(Project Name:Research on Robust Adaptive Allocation Mechanism of Human Machine Co-Driving System Based on NMS Features,Project Approval Number:52172381).
文摘To address the issue of scarce labeled samples and operational condition variations that degrade the accuracy of fault diagnosis models in variable-condition gearbox fault diagnosis,this paper proposes a semi-supervised masked contrastive learning and domain adaptation(SSMCL-DA)method for gearbox fault diagnosis under variable conditions.Initially,during the unsupervised pre-training phase,a dual signal augmentation strategy is devised,which simultaneously applies random masking in the time domain and random scaling in the frequency domain to unlabeled samples,thereby constructing more challenging positive sample pairs to guide the encoder in learning intrinsic features robust to condition variations.Subsequently,a ConvNeXt-Transformer hybrid architecture is employed,integrating the superior local detail modeling capacity of ConvNeXt with the robust global perception capability of Transformer to enhance feature extraction in complex scenarios.Thereafter,a contrastive learning model is constructed with the optimization objective of maximizing feature similarity across different masked instances of the same sample,enabling the extraction of consistent features from multiple masked perspectives and reducing reliance on labeled data.In the final supervised fine-tuning phase,a multi-scale attention mechanism is incorporated for feature rectification,and a domain adaptation module combining Local Maximum Mean Discrepancy(LMMD)with adversarial learning is proposed.This module embodies a dual mechanism:LMMD facilitates fine-grained class-conditional alignment,compelling features of identical fault classes to converge across varying conditions,while the domain discriminator utilizes adversarial training to guide the feature extractor toward learning domain-invariant features.Working in concert,they markedly diminish feature distribution discrepancies induced by changes in load,rotational speed,and other factors,thereby boosting the model’s adaptability to cross-condition scenarios.Experimental evaluations on the WT planetary gearbox dataset and the Case Western Reserve University(CWRU)bearing dataset demonstrate that the SSMCL-DA model effectively identifies multiple fault classes in gearboxes,with diagnostic performance substantially surpassing that of conventional methods.Under cross-condition scenarios,the model attains fault diagnosis accuracies of 99.21%for the WT planetary gearbox and 99.86%for the bearings,respectively.Furthermore,the model exhibits stable generalization capability in cross-device settings.
基金funded by the National Natural Science Foundation of China,grant number 61605004the Fundamental Research Funds for the Central Universities,grant number FRF-TP-19-016A2Guizhou Power Grid Co.,Ltd.2024 first batch of services(2024-2026 technology R&D services for science and technology projects(in addition to national and SGCC key projects)),grant number 060100KC23100012。
文摘This study addresses the risk of privacy leakage during the transmission and sharing of multimodal data in smart grid substations by proposing a three-tier privacy-preserving architecture based on asynchronous federated learning.The framework integrates blockchain technology,the InterPlanetary File System(IPFS)for distributed storage,and a dynamic differential privacy mechanism to achieve collaborative security across the storage,service,and federated coordination layers.It accommodates both multimodal data classification and object detection tasks,enabling the identification and localization of key targets and abnormal behaviors in substation scenarios while ensuring privacy protection.This effectively mitigates the single-point failures and model leakage issues inherent in centralized architectures.A dynamically adjustable differential privacy mechanism is introduced to allocate privacy budgets according to client contribution levels and upload frequencies,achieving a personalized balance between model performance and privacy protection.Multi-dimensional experimental evaluations,including classification accuracy,F1-score,encryption latency,and aggregation latency,verify the security and efficiency of the proposed architecture.The improved CNN model achieves 72.34%accuracy and an F1-score of 0.72 in object detection and classification tasks on infrared surveillance imagery,effectively identifying typical risk events such as not wearing safety helmets and unauthorized intrusion,while maintaining an aggregation latency of only 1.58 s and a query latency of 80.79 ms.Compared with traditional static differential privacy and centralized approaches,the proposed method demonstrates significant advantages in accuracy,latency,and security,providing a new technical paradigm for efficient,secure data sharing,object detection,and privacy preservation in smart grid substations.
文摘In federated learning,backdoor attacks have become an important research topic with their wide application in processing sensitive datasets.Since federated learning detects or modifies local models through defense mechanisms during aggregation,it is difficult to conduct effective backdoor attacks.In addition,existing backdoor attack methods are faced with challenges,such as low backdoor accuracy,poor ability to evade anomaly detection,and unstable model training.To address these challenges,a method called adaptive simulation backdoor attack(ASBA)is proposed.Specifically,ASBA improves the stability of model training by manipulating the local training process and using an adaptive mechanism,the ability of the malicious model to evade anomaly detection by combing large simulation training and clipping,and the backdoor accuracy by introducing a stimulus model to amplify the impact of the backdoor in the global model.Extensive comparative experiments under five advanced defense scenarios show that ASBA can effectively evade anomaly detection and achieve high backdoor accuracy in the global model.Furthermore,it exhibits excellent stability and effectiveness after multiple rounds of attacks,outperforming state-of-the-art backdoor attack methods.
基金supported by the National Natural Science Foundation of China(Grant No.62172123)the Key Research and Development Program of Heilongjiang Province,China(GrantNo.2022ZX01A36).
文摘Federated Learning(FL)protects data privacy through a distributed training mechanism,yet its decentralized nature also introduces new security vulnerabilities.Backdoor attacks inject malicious triggers into the global model through compromised updates,posing significant threats to model integrity and becoming a key focus in FL security.Existing backdoor attack methods typically embed triggers directly into original images and consider only data heterogeneity,resulting in limited stealth and adaptability.To address the heterogeneity of malicious client devices,this paper proposes a novel backdoor attack method named Capability-Adaptive Shadow Backdoor Attack(CASBA).By incorporating measurements of clients’computational and communication capabilities,CASBA employs a dynamic hierarchical attack strategy that adaptively aligns attack intensity with available resources.Furthermore,an improved deep convolutional generative adversarial network(DCGAN)is integrated into the attack pipeline to embed triggers without modifying original data,significantly enhancing stealthiness.Comparative experiments with Shadow Backdoor Attack(SBA)across multiple scenarios demonstrate that CASBA dynamically adjusts resource consumption based on device capabilities,reducing average memory usage per iteration by 5.8%.CASBA improves resource efficiency while keeping the drop in attack success rate within 3%.Additionally,the effectiveness of CASBA against three robust FL algorithms is also validated.
基金funding from the National Key Research and Development Program of China(No.2018YFE0110000)the National Natural Science Foundation of China(No.11274259,No.11574258)the Science and Technology Commission Foundation of Shanghai(21DZ1205500)in support of the present research.
文摘While reinforcement learning-based underwater acoustic adaptive modulation shows promise for enabling environment-adaptive communication as supported by extensive simulation-based research,its practical performance remains underexplored in field investigations.To evaluate the practical applicability of this emerging technique in adverse shallow sea channels,a field experiment was conducted using three communication modes:orthogonal frequency division multiplexing(OFDM),M-ary frequency-shift keying(MFSK),and direct sequence spread spectrum(DSSS)for reinforcement learning-driven adaptive modulation.Specifically,a Q-learning method is used to select the optimal modulation mode according to the channel quality quantified by signal-to-noise ratio,multipath spread length,and Doppler frequency offset.Experimental results demonstrate that the reinforcement learning-based adaptive modulation scheme outperformed fixed threshold detection in terms of total throughput and average bit error rate,surpassing conventional adaptive modulation strategies.
文摘In recent years,researchers have leveraged single-agent reinforcement learning to boost educational outcomes and deliver personalized interventions;yet this paradigm provides no capacity for inter-agent interaction.Multi-agent reinforcement learning(MARL)overcomes this limitation by allowing several agents to learn simultaneously within a shared environment,each choosing actions that maximize its own or the group's rewards.By explicitly modeling and exploiting agent-to-agent dynamics,MARL can align those interactions with pedagogical goals such as peer tutoring,collaborative problem-solving,or gamified competition,thus opening richer avenues for adaptive and socially informed learning experiences.This survey investigates the impact of MARL on educational outcomes by examining evidence of its effectiveness in enhancing learner performance,engagement,equity,and reducing teacher workload compared to single agent or traditional approaches.It explores the educational domains and pedagogical problems addressed by MARL,identifies the algorithmic families used,and analyzes their influence on learning.The review also assesses experimental settings and evaluation metrics to determine ecological validity,and outlines current challenges and future research directions in applying MARL to education.
基金supported by the National Natural Science Foundation of China(NSFC)under Grant number:82171965.
文摘Designing appropriate loss functions is critical to the success of supervised learning models.However,most conventional losses are fixed and manually designed,making them suboptimal for diverse and dynamic learning scenarios.In this work,we propose an Adaptive Meta-Loss Network(Adaptive-MLN)that learns to generate taskagnostic loss functions tailored to evolving classification problems.Unlike traditional methods that rely on static objectives,Adaptive-MLN treats the loss function itself as a trainable component,parameterized by a shallow neural network.To enable flexible,gradient-free optimization,we introduce a hybrid evolutionary approach that combines GeneticAlgorithms(GA)for global exploration and Evolution Strategies(ES)for local refinement.This co-evolutionary process dynamically adjusts the loss landscape,improvingmodel generalization without relying on analytic gradients or handcrafted heuristics.Experimental evaluations on synthetic tasks and the CIFAR-10 andMNIST datasets demonstrate that our approach consistently outperforms standard losses such as Cross-Entropy and Mean Squared Error in terms of accuracy,convergence,and adaptability.
文摘With the increasing complexity of vehicular networks and the proliferation of connected vehicles,Federated Learning(FL)has emerged as a critical framework for decentralized model training while preserving data privacy.However,efficient client selection and adaptive weight allocation in heterogeneous and non-IID environments remain challenging.To address these issues,we propose Federated Learning with Client Selection and Adaptive Weighting(FedCW),a novel algorithm that leverages adaptive client selection and dynamic weight allocation for optimizing model convergence in real-time vehicular networks.FedCW selects clients based on their Euclidean distance from the global model and dynamically adjusts aggregation weights to optimize both data diversity and model convergence.Experimental results show that FedCW significantly outperforms existing FL algorithms such as FedAvg,FedProx,and SCAFFOLD,particularly in non-IID settings,achieving faster convergence,higher accuracy,and reduced communication overhead.These findings demonstrate that FedCW provides an effective solution for enhancing the performance of FL in heterogeneous,edge-based computing environments.
基金the DigiEcoQuarry project,funded by the European Union's Horizon 2020 research and innovation program under Grant Agreement No.101003750supported by the China Scholarship Council(Grant No.202006370006).
文摘Agile lithology identification can assist mining by providing important information in the exploration and production of mineral resources.This study proposes a new lithology recognition procedure using video-logging of boreholes with an endoscope,applied to six production blocks in a limestone quarry.Images are automatically extracted from the videos and the lithology is classified into three classes based on clay content,i.e.massive limestone,brecciated limestone,and high amount of clay.The image quality is evaluated with a gray pixel intensity threshold and three no-reference image quality metrics,i.e.perception-based image quality evaluator,natural image quality evaluator,and blind/referenceless image spatial quality evaluator.After removing low-quality images,7583 images are retained and used for developing lithology classification models using six optimized classification techniques.The contrast-limited adaptive histogram equalization(CLAHE)technique is used to improve image quality.Ten color characteristics involving three percentiles of red,green and blue pixel intensities,together with color counting and five texture characteristics-correlation,entropy,homogeneity,contrast and energy-are used as inputs.Bayesian optimized light gradient boosting machine model performs best,with an overall accuracy of 88.04%,and a precision on the classes of massive limestone,brecciated limestone and high amount of clay of 90.72%,83.52%and 85.29%,respectively,for the testing set.The feature importance scores show that the color counting is the most significant parameter for the development of the classification model.Compared with previous image-based methodologies,this study provides a more flexible and cheaper procedure to identify lithology.
基金supported by the Beijing Natural Science Foundation(Grant Nos.F251036 and L248103)CAS Project for Young Scientists in Basic Research(Grant Nos.YSBR-090 and YSBR-05)National Natural Science Foundation of China(Grant No.62274159).
文摘Accurate temperature control and effective oxide removal are essential for achieving high-quality epitaxial growth in molecular beam epitaxy(MBE).However,traditional methods often rely on manual identification of reflection high-energy electron diffraction(RHEED)patterns.This process is heavily influenced by the grower’s experience,leading to issues with reproducibility and limiting the potential for automation.In this report,we propose an unsupervised learning framework for realtime RHEED analysis during the deoxidation process.By incorporating temporal similarity constraints into contrastive learning,our model generates smooth and interpretable feature trajectories that illustrate transitions in the deoxidation state,thus eliminating the need for manual labeling.The model,pre-trained using grouped contrastive loss,shows significant improvement in RHEED feature boundary discrimination and localization of critical regions.We evaluated its generalizability through two transfer learning strategies:calibration-free clustering and few-shot fine-tuning.The pre-trained model achieved a clustering accuracy of 88.1%for GaAs deoxidation samples without additional labels and reached an accuracy of 94.3%to 95.5%after fine-tuning with just five sample pairs across GaAs,Ge,and InAs substrates.This framework is optimized for resource-constrained edge devices,allowing for real-time,plug-and-play integration with existing MBE systems and swift adaptation across various materials and equipment.This work paves the way for greater automation and improved reproducibility in semiconductor manufacturing.
基金supported by the National Natural Science Foundation of China under Grants 62203338,62173259 and U1913602Zhejiang Provincial Natural Science Foundation of China under Grant LZ24F0390006the Postdoctoral Science Foundation of China under Grant 2022M722485.
文摘This paper investigates the adaptive optimal tracking control(AOTC)for underactuated surface vessels(USVs).Compared to the majority of existing studies,the control strategy in this paper innovatively combines an extended state observer(ESO)with reinforcement learning(RL).The designed ESO has high estimation accuracy and robust disturbance rejection capabilities for the unmeasurable information for USVs.To obtain the AOTC,the actor–critic(AC)networks based on RL are constructed to solve the Hamilton–Jacobi–Bellman(HJB)equations.Due to the uncertainties,it is challenging to obtain the optimal controller by directly solving the HJB equations.To address this issue,this paper employs neural networks(NNs)to approximate the uncertainties and solves the optimal controller via AC-RL and ESO.In addition,the adaptive parameters of the optimal controller is trained in parallel with AC networks,which can ensure that the trained networks can further improve tracking performance.The boundedness of AOTC for USVs is shown by Lyapunov stability theorem.Finally,simulation results demonstrate the effectiveness of the proposed algorithm.
基金supported by the National Key Research and Development Program of China(No.2022YFB4300902).
文摘As joint operations have become a key trend in modern military development,unmanned aerial vehicles(UAVs)play an increasingly important role in enhancing the intelligence and responsiveness of combat systems.However,the heterogeneity of aircraft,partial observability,and dynamic uncertainty in operational airspace pose significant challenges to autonomous collision avoidance using traditional methods.To address these issues,this paper proposes an adaptive collision avoidance approach for UAVs based on deep reinforcement learning.First,a unified uncertainty model incorporating dynamic wind fields is constructed to capture the complexity of joint operational environments.Then,to effectively handle the heterogeneity between manned and unmanned aircraft and the limitations of dynamic observations,a sector-based partial observation mechanism is designed.A Dynamic Threat Prioritization Assessment algorithm is also proposed to evaluate potential collision threats from multiple dimensions,including time to closest approach,minimum separation distance,and aircraft type.Furthermore,a Hierarchical Prioritized Experience Replay(HPER)mechanism is introduced,which classifies experience samples into high,medium,and low priority levels to preferentially sample critical experiences,thereby improving learning efficiency and accelerating policy convergence.Simulation results show that the proposed HPER-D3QN algorithm outperforms existing methods in terms of learning speed,environmental adaptability,and robustness,significantly enhancing collision avoidance performance and convergence rate.Finally,transfer experiments on a high-fidelity battlefield airspace simulation platform validate the proposed method's deployment potential and practical applicability in complex,real-world joint operational scenarios.
基金supported by the National Natural Science Foundation of China(No.62350048)。
文摘To address the challenge of achieving decentralized,scalable,and adaptive control for large-scale multiple unmanned aerial vehicle(multi-UAV)swarms in dynamic urban environments with obstacles and wind perturbations,we proposed a hybrid framework integrating adaptive reinforcement learning(RL),multi-modal perception fusion,and enhanced pigeon flock optimization(PFO)with curiosity-driven exploration to enable robust autonomous and formation control.The framework leverages meta-learning to optimize RL policies for real-time adaptation,fuses sensor data for precise state estimation,and enhances PFO with learned leader-follower dynamics and exploration rewards to maintain cohesive formations and explore uncertain areas.For swarms of 10–30 UAVs,it achieves 34%faster convergence,61%reduced stability root mean square error(RMSE),88%fewer collisions and 85.6%–92.3%success rates in target detection and encirclement,outperforming standard multi-agent RL,pure PFO,and single-modality RL.Three-dimensional trajectory visualizations confirm cohesive formations,collision-free maneuvers,and efficient exploration in urban search-and-rescue scenarios.Innovations include meta-RL for rapid adaptation,multi-modal fusion for robust perception,and curiosity-driven PFO for scalable,decentralized control,advancing real-world multi-UAV swarm autonomy and coordination.
基金supported by the Research Grant Fund from Kwangwoon University in 2023,the National Natural Science Foundation of China under Grant(62311540155)the Taishan Scholars Project Special Funds(tsqn202312035)the open research foundation of State Key Laboratory of Integrated Chips and Systems.
文摘Wearable wristband systems leverage deep learning to revolutionize hand gesture recognition in daily activities.Unlike existing approaches that often focus on static gestures and require extensive labeled data,the proposed wearable wristband with selfsupervised contrastive learning excels at dynamic motion tracking and adapts rapidly across multiple scenarios.It features a four-channel sensing array composed of an ionic hydrogel with hierarchical microcone structures and ultrathin flexible electrodes,resulting in high-sensitivity capacitance output.Through wireless transmission from a Wi-Fi module,the proposed algorithm learns latent features from the unlabeled signals of random wrist movements.Remarkably,only few-shot labeled data are sufficient for fine-tuning the model,enabling rapid adaptation to various tasks.The system achieves a high accuracy of 94.9%in different scenarios,including the prediction of eight-direction commands,and air-writing of all numbers and letters.The proposed method facilitates smooth transitions between multiple tasks without the need for modifying the structure or undergoing extensive task-specific training.Its utility has been further extended to enhance human–machine interaction over digital platforms,such as game controls,calculators,and three-language login systems,offering users a natural and intuitive way of communication.
基金supported in part by the National Key R&D Program of China under Grant 2021YFB1714100in part by the Shenzhen Science and Technology Program,China,under Grant JCYJ20220531101015033.
文摘With increasing density and heterogeneity in unlicensed wireless networks,traditional MAC protocols,such as Carrier Sense Multiple Access with Collision Avoidance(CSMA/CA)in Wi-Fi networks,are experiencing performance degradation.This is manifested in increased collisions and extended backoff times,leading to diminished spectrum efficiency and protocol coordination.Addressing these issues,this paper proposes a deep-learning-based MAC paradigm,dubbed DL-MAC,which leverages spectrum data readily available from energy detection modules in wireless devices to achieve the MAC functionalities of channel access,rate adaptation,and channel switch.First,we utilize DL-MAC to realize a joint design of channel access and rate adaptation.Subsequently,we integrate the capability of channel switching into DL-MAC,enhancing its functionality from single-channel to multi-channel operations.Specifically,the DL-MAC protocol incorporates a Deep Neural Network(DNN)for channel selection and a Recurrent Neural Network(RNN)for the joint design of channel access and rate adaptation.We conducted real-world data collection within the 2.4 GHz frequency band to validate the effectiveness of DL-MAC.Experimental results demonstrate that DL-MAC exhibits significantly superior performance compared to traditional algorithms in both single and multi-channel environments,and also outperforms single-function designs.Additionally,the performance of DL-MAC remains robust,unaffected by channel switch overheads within the evaluation range.
基金supported by grants from the National Natural Science Foundation of China(No.72361014)the Technical Field Fund of Basic Research Strengthening Program(Project No.2021-JCJQ-JJ-0003)+1 种基金the Major Program of the National Social Science Foundation of China(No.19ZDA103)the Science and Technology Project of Jiangxi Provincial Department of Education(No.GJJ2200526).
文摘Machine learning methods are widely used to evaluate the risk of small-and mediumsized enterprises(SMEs)in supply chain finance(SCF).However,there may be problems with data scarcity,feature redundancy,and poor predictive performance.Additionally,data collected over a long time span may cause differences in the data distribution,and classic supervised learning methods may exhibit poor predictive abilities under such conditions.To address these issues,a domain-adaptation-based multistage ensemble learning paradigm(DAMEL)is proposed in this study to evaluate the credit risk of SMEs in SCF.In this methodology,a bagging resampling algorithm is first used to generate a dataset to address data scarcity.Subsequently,a random subspace is applied to integrate various features and reduce feature redundancy.Additionally,a domain adaptation approach is utilized to reduce the data distribution discrepancy in the cross-domain.Finally,dynamic model selection is developed to improve the generalization ability of the model in the fourth stage.A real-world credit dataset from the Chinese securities market was used to validate the effectiveness and feasibility of the multistage ensemble learning paradigm.The experimental results demonstrated that the proposed domain-adaptation-based multistage ensemble learning paradigm is superior to principal component analysis,joint distribution adaptation,random forest,and other ensemble and transfer learning methods.Moreover,dynamic model selection can improve the model generalization performance and prediction precision of minority samples.This can be considered a promising solution for evaluating the credit risk of SMEs in SCF for financial institutions.
文摘This paper proposes virtual impedance adaptation of the lower-limb exoskeleton for human performance augmentation(LEHPA) based on deep reinforcement learning(VIADRL) to mitigate reliance on model accuracy and address the ever-changing human-exoskeleton interaction(HEI) dynamics. The classical sensitivity amplification control strategy is expanded to the virtual impedance control strategy with more learnable virtual impedance parameters. The adjustment of these virtual impedance parameters is formalized as finding the optimal policy for a Markov Decision Process and can then be effectively resolved using deep reinforcement learning algorithms. To ensure safe and efficient policy training, a multibody simulation environment is established to facilitate the training process, supplemented by the innovative hybrid inverse-forward dynamics simulation approach for executing the simulation. For comparison purposes, the SADRL strategy is introduced as a benchmark. A novel control performance evaluation method based on the HEI forces at the back, thighs, and shanks is proposed to quantitatively evaluate the performance of our proposed VIADRL strategy. The VIADRL controller is systematically compared with the SADRL controller at five selected walking speeds. The lumped ratio of HEI forces under the SADRL strategy relative to those under the SADRL strategy is as low as 0.81 in simulation and approximately 0.89 on the LEHPA prototype. The overall reduction of HEI forces demonstrates the superiority of the VIADRL strategy in comparison to the SADRL strategy.
基金funded by the National Science and Technology Major Project(2022ZD0115703)the Beijing Postdoctoral Research Foundation(2023-ZZ-116).
文摘The environment has an important impact on maize(Zea mays L.)production,making it necessary to identify plant adaptation regions that are suitable for different maize varieties.Traditional methods using field trials are costly and restricted to a limited number of areas.Identifying adaptation regions based on climate data has great potential,but a basic understanding and a prediction approach for diverse maize varieties are lacking.Here,we collected a representative dataset comprising 32,840 data points from the National Maize Variety Trial Data Management Platform.We employed three traits to characterize the adaptability of different maize varieties:PH(plant height),DTS(days to silking),and yield.First,we quantified the contributions of variety(V),environment(E),and V×E to variance in the three adaptationrelated traits.The mean contributions of E to variance in PH,DTS,and yield were 54.50%,82.87%,and 75.92%,respectively,suggesting that environmental effects are crucial for phenotype construction.Second,we analyzed correlations between the three traits and three environmental indices:GDD(growing degree days),PRE(precipitation),and SSD(sunshine duration).The highest absolute correlation coefficients between phenotypes and environmental indices were 0.15–0.69 at the whole-data level.To predict variety adaptation on a national scale,we modeled the three traits using environmental indices and best linear unbiased predictors(BLUPs)via the random forest algorithm.The predictive abilities of our models for PH,DTS,and yield were 0.90(MAE=9.95 cm),0.99(MAE=1.09 d),and 0.95(MAE=0.55 t ha^(−1)),respectively,indicating that our proposed framework can predict adaptationrelated traits for diverse maize varieties in China.
基金supported by the National Natural Science Foundation of China(Grant Nos.62233008 and 51705247)the State Key Laboratory of Mechanics and Control for Aerospace Structures of Nanjing University of Aeronautics and Astronautics.
文摘Autonomous legged robots,capable of navigating uneven terrain,can perform a diverse array of tasks.However,designing locomotion controllers remains challenging.In particular,designing a controller based on durable and reliable proprioceptive sensors,is essential for achieving adaptability.Presently,the controller must either be manually designed for specific robots and tasks,or developed using machine-learning techniques,which require extensive training time and result in complex controllers.Inspired by animal locomotion,we propose a simple yet comprehensive closed-loop modular framework that utilizes minimal proprioceptive feedback(i.e.,the Coxa-Femur(CF)joint angle),enabling a quadruped robot to efficiently navigate unpredictable and uneven terrains,including the step and slope.The framework comprises a basic neural control network capable of rapidly learning optimized motor patterns,and a straightforward module for sensory feedback sharing and integration.In a series of experiments,we show that integrating sensory feedback into the base neural control network aids the robot in continually learning robust motor patterns on flat,step,and slope terrain,compared with the open-loop base framework.Sharing sensory feedback information across the four legs enables a quadruped robot to proactively navigate unpredictable steps with minimal interaction.Furthermore,the controller remains functional even in the absence of sensor signals.This control configuration was successfully transferred to a physical robot without any modifications.
基金supported by the Science and Technology Project of State Grid Corporation of China under grant 52094021N010(5400-202199534A-0-5-ZN)。
文摘Low-carbon smart parks achieve selfbalanced carbon emission and absorption through the cooperative scheduling of direct current(DC)-based distributed photovoltaic,energy storage units,and loads.Direct current power line communication(DC-PLC)enables real-time data transmission on DC power lines.With traffic adaptation,DC-PLC can be integrated with other complementary media such as 5G to reduce transmission delay and improve reliability.However,traffic adaptation for DC-PLC and 5G integration still faces the challenges such as coupling between traffic admission control and traffic partition,dimensionality curse,and the ignorance of extreme event occurrence.To address these challenges,we propose a deep reinforcement learning(DRL)-based delay sensitive and reliable traffic adaptation algorithm(DSRTA)to minimize the total queuing delay under the constraints of traffic admission control,queuing delay,and extreme events occurrence probability.DSRTA jointly optimizes traffic admission control and traffic partition,and enables learning-based intelligent traffic adaptation.The long-term constraints are incorporated into both state and bound of drift-pluspenalty to achieve delay awareness and enforce reliability guarantee.Simulation results show that DSRTA has lower queuing delay and more reliable quality of service(QoS)guarantee than other state-of-the-art algorithms.
