Abstract:

Abstract: To address the issue of analyzing the complex dynamic behaviors of gait, such as slip and bounce, in passive walking robots, an arc-based bipedal passive walking robot model considering foot-ground contact was developed, to analyze various walking modes exhibited with complex walking conditions with changes in ground surface materials. It could reveal the gait instability mechanisms with multi-parameter co-evolution. Firstly, the Hertz contact model and Coulomb friction model was introduced to describe the normal contact force and tangential friction force between the foot and the ground. The second kind of Lagrangian equation was used for dynamic modeling. Secondly, by changing contact parameters to simulate different walking surfaces, numerical simulations were employed to study phenomena such as slipping, bouncing, and falling during walking, so as to obtaining foot-ground contact conditions that optimize the robot′s walking performance. Finally, tools such as bifurcation diagrams and time response plots were used to analyze the walking gait of the robot during the co-evolution of various structural parameters. The research results indicated that reducing the friction coefficient and damping coefficient could cause slipping and bouncing in the gait, respectively, and continued reduction could cause the robot to fall. Among the structural parameters, reducing the hip joint mass and moment of inertia, and increasing the foot radius were more conducive to the robot adapting to a wide range of walking slope angles. Seemed that regardless of how the structural parameters evolved, the robot would experience gait instability due to slipping as the slope angle increased.

Abstract: This study aimed to address the estimation and compensation of unmatched uncertainties with unknown bounds in sliding mode control. A non-singular terminal sliding mode controller was designed for a four-wheel mobile robot. To ensure the existence of the sliding surface, Gaussian process regression (GPR) was employed for online estimation of the unmatched uncertainties. GPR not only estimated the bound of the uncertainties but also provided the mean and variance, which allows for a more robust estimation. On the one hand, the use of GPR for uncertainty estimation could help avoid the use of high-gain control, thereby reducing control chattering. On the other hand, the uncertainty compensation based on the estimates from GPR could enhance the adaptability of the modelbased sliding mode control algorithm. Additionally, an adaptive terminal sliding mode controller was designed based on the proximal policy optimization (PPO) algorithm. A reward function was constructed with the objective of improving control accuracy and minimizing control input chattering, which could enable the adaptive adjustment of the sliding mode controller’s parameters. The stability of the non-singular terminal sliding mode controller was proven through Lyapunov stability analysis. The effectiveness of the proposed control algorithm is validated through numerical simulations. The results demonstrated that the adaptive terminal sliding mode controller based on GPR significantly reduced by 90% while achieving high control accuracy, outperforming traditional control methods.

Abstract: In view of the problems of protective equipment detection, such as information interference, uneven illumination and occlusion in the construction scene, a lightweight algorithm with improved YOLOv8n was proposed, which was called YOLO-LA . Firstly, the weighted bidirectional feature pyramid network BiFPN was introduced into the neck, and the underlying details and high-level semantic information were improved through multi-path interactive fusion, the multi-scale feature fusion performance was enhanced, and the detection accuracy of the model for small targets in complex scenes was improved. Secondly, the C2f-ContextGuided module was used to transform the backbone network in the baseline model, and the global context information was used to calculate the weight vector, to refine the joint features of the local features and the surrounding context features, so as to improve the feature extraction ability of the model and reduce the complexity of the model. Then, a new LSCD lightweight detection head was proposed, which used shared convolution to reduce the number of parameters and computations of the model. Finally, EIoU was used to replace the original CIoU, and the border regression was optimized, and the convergence speed and regression accuracy of the algorithm were improved. Compared with the baseline model YOLOv8n, the number of parameters, the amount of computation, and the size of the model were reduced by 61.5%, 43.2% and 58.7%, respectively, and the mAP@0.5 was increased by 1.4 percentage points, and the FPS was 253 frames/s, which could meet the requirements of real-time, accuracy and lightweight of protective equipment wearing detection.

Abstract: In order to solve the task autonomy assignment problem of AMR in flexible production, a multi-agent deep deterministic policy gradient (MADDPG) algorithm based on improved multi-agent reinforcement learning algorithm was adopted. The attention mechanism was introduced to improve the algorithm. Firstly, the framework of centralized training decentralized execution was adopted, and then the action and state of AMR were set. Secondly, according to the size of the reward value, the coverage degree of the task node and the completion effect of the task were determined. The simulation results showed that the average reward value of MADDPG algorithm increase 3 than other algorithms, and the training times were reduced by 300 times. It could have faster learning speed and more stable convergence process while ensuring the completion of task allocation.

Abstract: In order to improve the adaptability of lower limb rehabilitation robot, such as its difficulty to respond to the patient′s intention in real time, and to adapt to individual motion needs, in this study, an adaptive control method of lower limb exoskeleton was proposed based on myoelectric prediction model. By collecting the surface EMG signals of biceps femoris, rectus femoris and lateral femoris muscles, the EMG prediction model was constructed to predict the expected motion trajectory of patients. Aiming at the uncertainties and model errors of the system, an adaptive sliding mode controller was designed to dynamically adjust sliding mode parameters according to muscle activation, so as to improve the tracking accuracy and compliance of the robot. The myoelectric model and sliding mode controller were tested in an experiment with 5 healthy subjects. The results showed that the RMSE of the model was 7.94 for the hip joint and 9.31 for the knee joint, which could meet the need of trajectory generation. Compared with the traditional PID control, the tracking accuracy of the adaptive sliding mode controller was improved by 28%, which proved the effectiveness of the method.

Abstract: To address the challenges posed by the intermittency and randomness of photovoltaic (PV) power output to maintaining stable power system frequency, a rapid frequency regulation method based on the twin delayed deep deterministic policy gradient (TD3) algorithm was proposed. No need to rely on specific mechanistic models, this method could tackle the strong uncertainties associated with PV power generation. Firstly, a simplified model of the PV power generation system was constructed. Secondly, a novel frequency controller was designed leveraging the TD3 algorithm. Lastly, the proposed control strategy was compared with traditional droop control, sliding mode control, and a control strategy based on the deep deterministic policy gradient (DDPG) algorithm. The results demonstrated that, in two scenarios, single-step and continuous-step load disturbances respectively, the frequency deviations based on the proposed control strategy were significantly lower than those of the other three control algorithms. Specifically, the integral of time-weighted absolute error (ITAE) criterion showed a reduction of 41.7% and 31.8% compared to the worst-performing droop control, thoroughly validating the superiority of the proposed control strategy in terms of both dynamic and steady-state performance during frequency regulation.

Abstract: To improve the accuracy of wind power interval prediction, in this study, a combined deep learningbased wind power interval prediction model was proposed. Firstly, to address the imbalance between global optimization ability and local exploration in the traditional dung beetle optimization (DBO) algorithm, an improved version POTDBO was introduced. This algorithm enhanced the global search capability and improved the local search strategy. By optimizing the decomposition number K and penalty factor β in the variational mode decomposition (VMD), thus it improved the decomposition performance of VMD. Secondly, based on the optimized VMD decomposition results, a combined deep learning model, POTDBO-VMD-CNN-BiLSTM, was established. In this model, convolutional neural networks (CNN) were used to extract the spatial features of wind power, and a bidirectional long short-term memory (BiLSTM) network was applied to capture both historical and future signal features in the data. The individual components were predicted and then combined to reconstruct the wind power prediction accurately. To perform interval prediction for wind power, in this study the non-parametric kernel density estimation (KDE) method was introduced to fit the prediction errors of the combined model, then to obtain wind power interval prediction results at different confidence levels. Finally, the proposed model was validated using actual operation data from a wind farm in Xinjiang. Simulation results showed that, at a 95% confidence level, compared to the Gaussian and T-distribution models, the proposed method reduced the prediction interval coverage width (CWC) by 0.103 6 and 0.171 4, respectively, while improving the interval prediction accuracy.

Abstract: In view of the power fluctuation and randomness existing in the operation of wind turbine networks, to improve the accuracy of wind speed prediction and the stability of wind turbine operation, in this study a VMDLSTM short-term wind speed prediction model was proposed based on the beluga whale optimization and the whale optimization algorithm. Firstly, the Beluga optimization algorithm was used to optimize the number of modes and penalty factors in VMD to obtain the reorganized subsequence. For parameters such as the number of hidden layer nodes, the maximum number of training generations, and the initial learning rate in LSTM, the whale optimization algorithm was used to determine these parameters. Finally, the monomer transplantation ability of LSTM was utilized to predict the data. The results indicated that the VMD-LSTM prediction model based on BWO and WOA proposed in this study achieved RMSE, MAE, and MAPE values of 0.223 4, 0.172 7, and 0.083 7, respectively, on the test set, all of which were lower than those of other comparative models. This validated the effectiveness of the proposed model in short-term wind speed prediction.

Abstract: In order to improve resident living environment along the metro access line, one section of Zhengzhou metro access line was taken as the example, the fast multi-pole boundary element method was considered, the sound barrier model was constructed with Virtual Lab acoustic simulation software, the wheel rail noise of low-speed subway train was simplified as a two-point sound source, the insertion loss of sound pressure level was adopted to characterize the noise reduction effectiveness of the sound barrier, and the influence of some factors on noise reduction effectiveness were analyzed such as the sound source sound pressure level, sound barrier height, distance between sound barriers and sound sources, the top structure of the sound barrier and the location in the sound shadow area. According to the simulation results, the vertical Y-shaped reflective sound barrier with 4 m height and 3 m distance from the center of dual sound sources was set up on the metro access line, the noise signals were in-situ measured, the noise characteristic parameters were obtained, and the simulation results had good agreement with the test results, the simulation calculation was verified to be reasonable. The closest distance between buildings along the metro access line and the center of the track is 30 meters, the noise of residential buildings with sound barrier along the line meet with relevant regulatory requirements, Y-shaped sound barrier had better noise reduction performance, and the relevant achievements could provide parameters for the noise reduction design and performance evaluation of metro access line.

Abstract: There are problems in the optimization of mountainous highway routes, including complex evaluation indicators, difficulty in quantifying qualitative indicators, and the weight of subjectively set indicators do not match the actual situation. In this study, based on the principles of technology, economy, and safety, the Hall three-dimensional structure was to analyzed to explore the influencing factors of mountainous highway routes, and to construct an evaluation index system. cloud model theory was introduced to quantify qualitative evaluation indicators, and the variable weight theory was used to modify the constant weight of evaluation indicators considering the impact of the actual state of evaluation indicators on the evaluation results. Finally, a method for optimizing mountainous highway routes based on TOPSIS was proposed. This method was applied to the optimization of routes for the fourth risk point of the Alcia Highway Project in Bolivia. The results indicated that cloud models could effectively solve the problem of uncertainty in qualitative indicators, which made them difficult to quantify. The closeness degree of the three route schemes was 0.833, 0.606, and 0.684, respectively. Compared with traditional constant weight, the variable weight theory highlighted the influence of extreme indicators on the evaluation results in the evaluation process, and the results were more realistic.

Abstract: To tackle frequent subway delays and interruptions, to address the shortcomings of common emergency strategies, in this study, an emergency strategy for localized subway line interruptions was proposed based on resilience theory. Firstly, the evolution process of subway line performance during emergencies were thoretically analyzed. Subsequently, a resilience evaluation model for subway lines was constructed using indicators such as line connectivity, delay resistance index, and passenger retention rate. Finally, aiming to enhance subway line resilience, a multi-section single-track bidirectional operation emergency strategy was proposed for common incidents of partial interruption on one side of subway tracks. This strategy divided the operable track in the opposite direction of the accident site into several sections, with a train operating in a single-track bidirectional mode within each section. To assess the effectiveness of the proposed strategy, a case study was conducted on the Line 4 of the Changsha subway during weekday morning peak hours. The results indicated that during emergencies, the majority of passengers experienced wait times concentrated within 5-7 minutes, with passenger waiting rates dropping below 50% within 10 minutes. Compared to traditional shuttle operation and single-track bidirectional operation schemes, the multi-section single-track bidirectional operation scheme improved passenger retention rates by 42.4% and 12.7%, respectively, with a corresponding increase in line resilience of 145% and 50%.

Abstract: Aiming at the problems of large labor demand and high transport loss in manual assembly of prefabricated ecological brick slope protection site, and considering the resource utilization of silt waste, a method of using silt as a cementitious material to prepare self-compacting concrete (SCC) for non-structural ecological slope protection was proposed. This approach replaced prefabricated ecological bricks and could reduce costs while increasing efficiency. In this study, the physicochemical properties of silt were analyzed and the effects of silt admixture on working properties, mechanical properties, and pore structure was investigated. The results indicated that the working performance of SCC declined with the addition of silt. To meet SCC specifications and construction requirements, the amount of water reducing agent increased correspondingly with the increase in silt mixing. The mechanical properties of the material decreased as the amount of silt increases. Specifically, the 28-day compressive strength, split tensile strength, and modulus of elasticity were all reduced. This was due to the effect of silt on the pore structure of SCC, which led to an increase in cumulative pore volume and porosity.The results of this study showed that SCC with 10% silt incorporated could reach the strength of C30 at a slump extension of 700 mm, which verified its feasibility as an ecological slope protection material.

Abstract: Urban drainage models (UDMs) often involved a large number of complex parameters, leading to significant challenges for model calibration. Existing studies focused on calibration methods but neglected the impact of monitoring point placement on the calibration effect. It led to the issues of poor generalization ability and low reliability of the model. In this study, the real-world UDM in the town of Bellinge, Denmark was taken as a case study and optimization methods were utilied to determine the optimal layouts of sensors under different numbers, therefore the impact of the number and layout of sensors on the UDM calibration were analyzed. Results showed that increasing the number of sensors and optimizing the sensor layout could significantly improve the accuracy and robustness of the model calibration. For instance, the average prediction accuracy of the calibration model at flooding nodes with the optimized layout scheme with five water level sensors improved by about 53% compared with the spatially uniform layout scheme. The study also revealed that the UDM calibration problem faced the challenge of "parameter equifinality", which hindered the accurate deduction of true parameter values. However, a calibration model that met practical requirements could still be obtained when there was sufficient observation information. For instance, increasing the number of sensors from one to more than five enhanced the overall prediction performance of the calibration model by over 38%.

Abstract: Aimed at the problem that the state parameters were difficult to obtain and the analysis results were single in the process of vehicle yaw stability analysis, a two-degree-of-freedom vehicle model was established as a reference model for yaw stability analysis and state estimation. The phase plane was constructed by using the sideslip angle and its angular velocity to analyze the yaw stability of the vehicle, and the adaptive phase plane stability domain based on multi-layer perceptron (MLP) was designed. According to the real-time state of the vehicle and the phase plane stability region, the yaw stability evaluation index was constructed. A vehicle state estimation algorithm based on extended Kalman filter (EKF) was designed, and a vehicle yaw stability analysis method based on state estimation was proposed. In order to verify the effectiveness and practicability of the proposed yaw stability analysis method, simulation tests at 100 km/h and real vehicle tests at 30 km/h were conducted in double lane change conditions. The simulation and real vehicle test results showed that the average error of sideslip angle estimation based on state estimation was less than 0.1°, and the average error of longitudinal velocity estimation was less than 0.03 m/s. This method could quantify the yaw stability from 0 to 1 based on the estimated vehicle state parameter input, reflecting the dynamic changes in vehicle yaw stability.

Abstract: Dynamic impedance matching technology can improve the output power and energy conversion efficiency of piezoelectric transducers. Most of the existing dynamic impedance matching method was intelligent numerical optimization algorithm, but the intelligent algorithm had some problems, such as complex modeling, long iteration time and large amount of calculation. In order to solve these problems, a dynamic impedance matching method based on data fitting was proposed, and the corresponding T-shaped impedance matching network was designed. The proposed method obtained the observed values of the resistance and reactance components of the transducer at the corresponding frequency by fine-tuning the operating frequency, and obtained a set of equivalent circuit parameters of the transducer with the highest degree of fitting based on the minimum residual sum of squares. Combined with the relevant formulas, the matching network element parameters and the series resonant frequency of the transducer were calculated. The frequency tracking function was further implemented. The dynamic impedance matching method was simulated in Python, and the simulation circuit was built in MATLAB/Simulink to simulate the effect of impedance matching and frequency tracking. The results showed that the proposed method could accurately obtain the equivalent circuit parameters of the transducer, the voltage and current signals at both ends of the matching Ttype impedance matching network were basically in phase, the active power was significantly improved, the matching effect was good, and the matching speed was also significantly improved compared with the genetic algorithm.

Abstract: Most of the existing studies took the entire linear worktable as the research target, and merely obtained the local influence laws of different errors on the one-dimensional linear worktable and failed to fully disclosed the coupling mechanisms among different types of errors as well as the influence laws on the overall motion error of the worktable. In response to this deficiency, in this study, the linear worktable with double guide rails and four sliders was taken as the research target. By means of the equivalent stiffness of the dynamic interface, the constraint stiffness matrix of the sliders was solved. Then, combined with the position and posture state of the worktable after the coordinated deformation of the sliders with the double constraints of the guide rails and the worktable, the error coupling model of the one-dimensional linear worktable was established. Based on the error coupling model, multiple parallelism errors were mapped onto the initial errors of the sliders through homogeneous transformation, and ultimately, the mapping model from the parallelism error of guide rail installation to the motion error of the worktable was established. Based on the mapping model, the influence laws of parallelism errors on the motion error of the worktable were simulated and analyzed, and the consistency of the results was verified through finite element simulation and experiments. The maximum error between the parallelism error mapping model and the finite element model did not exceed 9.8%.This study indicated that different types of parallelism errors had significantly different degrees of influence on the linear worktable. By reducing the parallelism errors of yaw and pitch, the overall error of the worktable could be greatly reduced. The proposed method provided a powerful theoretical support for the error compensation and precision design of linear worktables.

Abstract: Keyphrase prediction often fail to fully utilize the complex hierarchies and semantic information within text structures. To address this issue, a novel keyphrase prediction method that enhanced semantic representation, called adaptive contrastive learning for keyphrase prediction(ACL-KP) was proposed. This method introduced an adaptive weighting mechanism to dynamically adjust sample weights, to solve the problem of distinguishing true samples from noise samples during contrastive learning, thereby reducing the impact of misidentifying noise samples and optimizing spatial representation. Additionally, to increase the diversity of training data, Gaussian white noise was incorporated to automatically generate some challenging virtual samples, thus enhancing the semantic representation of documents and keywords. Experimental results on multiple public datasets in the keyphrase prediction field showed that the model improved performance by 2% to 17% in F1@5 and F1@M metrics compared to current state-of-the-art models. Compared to sequence-to-sequence models and unified models, the proposed model demonstrated a more significant performance advantage.

Abstract: The robot detection methods based on GNN ignored the importance of minority class nodes when dealing with class imbalance problems, and did not consider the unique connectivity problem of graph structures, resulting in unsatisfactory node classification performance. Therefore, in response to the shortcomings of existing solutions, in this study, a class imbalanced node classification algorithm was proposed based on minority class weighted and abnormal connectivity margin loss, which extended the traditional imbalanced classification idea in the field of machine learning to graph structured data. Based on GraphSMOTE, minority class weighted aggregation was performed to enhance the feature aggregation of minority nodes. In the oversampling stage, the SMOTE algorithm was used to process imbalanced data, which considering node representation and topology structure. Simultaneously training an edge generator to model relational information and introducing anomalous connectivity margin loss to improve GNN′s perception of connectivity anomalies and enhance the model′s learning of connectivity information.Finally, experiments were conducted on publicly available Weibo, Twitter fake accounts, and BlogCatalog datasets. The comparison results with the five baselines of SMOTE, Re-weight, GraphSMOTE, DR-GCN, and mGNN showed that the average ACC of the algorithm proposed in this study reached 84.3%, with an accuracy improvement of 1.3% compared to the mGNN model on the Kaggle dataset.

Abstract: The intelligent insertion assistance technology of aviation electrical connectors based on augmented reality or robotic arms relied on precise prior information such as precise contact positions and sorting information. However, the current acquisition of prior information relied entirely on manual collection, with low accuracy and completeness. To address this issue, in this study, a 3D model-based visual recognition method of aviation electrical connector′s contacts was proposed. This method coupled deep learning and image processing methods. The precise detection and positioning information acquisition of the 3D model contact for aviation electrical connectors was achieved through a two-step contact precise positioning method based on deep learning. Then, based on the circular layering idea, the positioned contacts were completed and sorted. Finally, fully automated, intelligent, and precise visual recognition of complex aviation electrical connector model contacts was achieved, and accurate contact positions and sorting information were obtained. The experiment showed that the contact recognition method proposed in this study was superior to the single deep learning method in both recognition rate and positioning accuracy. Among them, the fusion of YOLOv7 had the best effect, with an average recognition rate of 97.85%, an average positioning error of 0.025 mm, an average positioning time of 69 ms, a missing contact completion rate of 100%, and a sorting accuracy rate of 100%. It could provide accurate and effective prior information for intelligent insertion assistance of aviation electrical connectors based on augmented reality or robotic arms.