Abstract:A complete communication system must encompass a full two-way communication link, which includes both uplink and downlink. The uplink has always been a challenge for two-way communication systems. In recent years, the rapid development of visible light communication technology, with its advantages of no electromagnetic radiation, large communication capacity, and environmental friendliness, can serve as a supplement to traditional uplink solutions. The article first introduces the application scenarios and system composition of visible light communication, and then provides a review of the current research status of visible light uplink at home and abroad in recent years. In addition, it presents various schemes for visible light uplink, such as visible light with radio frequency, visible light with visible light, visible light with power line carrier, and single-source reverse modulation technology. Finally, it summarizes the current issues faced by the visible light communication uplink and summarizes the advantages and disadvantages of various schemes, as well as prospects for future development trends.

Abstract:The research on pre-trained large models has made remarkable achievements in recent years, this paper reviews the application of pre-trained large model in robotics. Traditional deep learning models in robots were trained on small datasets customized for specific tasks, which limits their adaptability in different applications. In contrast, large models pre-trained on internet-scale data appear to have superior generalization capabilities and in some cases show an exploratory ability to find one-shot solutions where they are not present in the training data. The underlying model has the potential to enhance the various components of a robot’s autonomous task, from perception to decision making and control. This paper examines recent papers that use or build large models to solve robotics problems, exploring how large models can help improve robots’ capabilities in the areas of perception, decision making, and control, thereby promoting the application of large robot models in more fields. Meanwhile, the challenges that hinder the application of large models in robotic autonomous systems were discussed in this paper, such as data scarcity in robotic applications, the variability of robots themselves, the limitations of multimodal representations, and real-time performance, and provides opportunities and potential approaches for future improvements.

Abstract:Aiming at improving the accuracy and real-time performance of zoom imaging depth measurement, based on the given system design configuration, a new monocular visual depth measurement method with liquid optical control is proposed by utilizing liquid lens adjustment characteristics and neural network technology. Firstly, to eliminate the influence of optical axis drift induced by the liquid gravity factor on the measurement results, the ratio of target image area is adopted as the feature parameter. A target area calculation method based on chain code classification and strip segmentation is presented. Then, in order to describe the mapping relationship between liquid lens parameters, image feature quantity and target depth, a neural network model of liquid monocular depth measurement is constructed, and the model parameters are optimized by genetic algorithm. Furthermore, the focal power function is obtained by calibrating the parameters of the liquid lens. The neural network trained on the dataset for depth measurement has an average prediction relative error of 0.799%. Finally, an experiment is designed to test and verify the method. The average depth measurement error of targets with different distances is 2.86%, and the average measurement speed is 108.2 ms. The measurement error for targets of different shapes at a distance of 1 000 mm shall not exceed 3.60%. The results show that the monocular vision method combining liquid optical control and neural network prediction can achieve high-precision and fast depth measurement, and has good generalization performance for different shapes of objects. The research provide a new technical idea for overcoming the existing limitations of zoom imaging ranging method.

Abstract:The combination of sparse sampling and image restoration can not only compress data capacity, but also improve imaging speed, which is of great significance for the development of high-resolution LiDAR imaging technology. In order to improve the restoration effect of sparse sampled images, a new residual channel attention network block was designed in the paper, and the residual channel attention block was introduced into a deep unfolding network based on compressed sensing iterative soft threshold method to suppress the blurring phenomenon caused by the loss of high-frequency information in image restoration and reconstruction, forming a new method for the restoration and reconstruction of sparse sampled LiDAR images. This method combines the advantages of traditional compressed sensing reconstruction methods and neural network methods, and has a faster reconstruction speed compared to traditional compressed sensing reconstruction methods. Compared with existing neural network methods, it enhances structural insight and improves the problem of image blur in reconstruction. The validation calculations using Middlebury Stereo Data 2006 as the test dataset show that our method not only has better image reconstruction quality compared to SDA, ReconNet, TVAL3, D-AMP, and IRCNN methods, but also has higher computational efficiency; When the sparse sampling ratio is 25%, the peak signal-to-noise ratio (PSNR) of the restored image is more than 1.6 dB higher than other methods, making it an ideal method for restoring sparse LiDAR images with good overall performance.

Abstract:Influenced by factors such as observation conditions and acquisition scenarios, underwater optical image data usually presents the characteristics of high-dimensional small samples and is easily accompanied with noise interference. Resulting in insufficient robust performance of many dimensionality reduction methods in their recognition process. To solve this problem, we propose a novel 2DPCA method for underwater optical image recognition, called arbitrary triangle structure 2DPCA (ATS-2DPCA). On the basis of considering the relationship between reconstruction error and variance of projection data, ATS-2DPCA can successfully match the flexible robust distance metric mechanism, which effectively improves the accuracy of underwater optical image recognition under noise interference environment and achieves reasonable protection of the geometric structure of the data. In this paper, we theoretically prove the availability and convergence of the proposed method and use three underwater optical image databases for experimental verification. The optimal recognition accuracy is 89.07%, 88.52%, and 86.00%, respectively. The extensive experimental results show that ATS-2DPCA has more outstanding performance than other 2DPCA-based methods.

Abstract:Due to the dense presence of multiple targets and complex background information in remote sensing images, existing detection algorithms often struggle to achieve satisfactory precision in small target detection. To address this issue, we propose a small target rotation box detection algorithm for remote sensing images called DRS-YOLO, based on reparameterized generalized pyramids and dilated residuals. First, to overcome the shortcomings of the backbone network in feature extraction, we enhance semantic information by incorporating a dilated residual module into the neck of the network, building upon the YOLOv8OBB framework. Second, to improve the network's performance in detecting multi-scale targets and facilitate the flow of low-level feature information to high-level features, we replace the neck structure with a reparameterized generalized feature pyramid network for more efficient multi-scale feature fusion, which aids in capturing high-level semantics and low-level spatial details. Finally, to further enhance the network's performance in detecting small targets, we propose the SPPFI module to expand the receptive field, thereby improving detection accuracy for remote sensing targets. Experimental results demonstrate that the improved algorithm achieves an increase in detection precision of 1.5% and 9.8% on the public DIOR and HRSC2016 datasets, respectively, compared to the original YOLOv8sOBB baseline network, indicating a significant enhancement in small target detection performance for remote sensing images.

Abstract:The multi-dimensional defect detection of cylindrical battery shells is a critical technology for ensuring the quality and safety of lithium batteries. Due to the different processes involved in production and transportation, defects may occur in each part of the cylindrical battery shells. To solve the problem of low detection accuracy in existing methods when handling the diverse and variably scaled defects of cylindrical battery shells, this study designs an image acquisition system based on the characteristics of each part of the battery shells and constructs a multi-dimensional defect dataset. Additionally, a multi-dimensional defect detection technology is proposed based on an improved YOLOv8n. Firstly, the switchable atrous convolution is used in the C2f module to improve the multi-scale feature extraction capability. Secondly, the down sampling module is refined by combining average pooling and max pooling strategies, reducing the spatial dimensions of feature maps while retaining key information. Finally, the LSKA attention mechanism is introduced to enhance the fusion effect of multi-scale features. Experimental results show that the improved YOLOv8n model achieves an average detection accuracy of 77.4% on a custom cylindrical battery shell defect dataset, which is 4.3% higher than the original model. Furthermore, the computational load is reduced by 17%, the model size is only 6 MB, and the detection speed reaches 177 FPS, meeting the requirements for real-time industrial mass detection.

Abstract:Tongue diagnosis in traditional Chinese medicine (TCM) judges the deficiency and strength of internal organs as well as the vitality of functions by observing tongue features. It has the advantages of being non-invasive and convenient. Accompanied by the rapid development and wide application of computer vision technology, it is crucial to develop a model that can perform automatic detection, extraction and recognition of tongue features. Toward demands for digital tongue diagnosis in traditional Chinese medicine clinic and health monitoring, an automatic detection model for tongue tooth mark and fissure features was proposed based on improved RetinaNet. The SimPSA-ResNet and SimSPPF module were introduced into the backbone of RetinaNet to enhance the feature extraction capability and robustness of the network. Meanwhile, the multi-level feature pyramid network structure was improved to ensure that the model can better integrate information from different scales, thereby focusing more accurately on the key information pertinent to tongue features. Finally, to further streamline the model’s output, redundant output feature layers were eliminated and integrated with the Attention-guided Spatial Feature Fusion structure. This step helps retain important features while improving the utilization of information within the network. The improved RetinaNet model was trained and predicted by using the self-built tongue image dataset, and the mean average precision(mAP) reaches 94.37%, which is 2.77% higher than that of the original algorithm. Experimental results conclusively demonstrate that the improved RetinaNet model can effectively elevate the detection accuracy of tongue tooth mark and fissure features. This advancement holds tremendous potential for facilitating daily self-examination, health management and assisting doctors in diagnosis.

Abstract:In dynamic control engineering, precise measurement of pendulum angle is crucial for achieving swing suppression. While binocular pendulum angle measurement methods rely on camera calibration results, changes in camera positions can introduce measurement deviations and pose challenges for efficient and stable measurements. To address these issues, this study proposes a monocular non-calibrated measurement method. The proposed method involves several key steps. First, continuous target image capture is combined with Kalman filtering and color space transformation to locate marker positions. Second, coordinate transformation based on cross-ratio is applied, comparing the current position with the initial position to calculate real-time pendulum angles through trigonometric transformations. Finally, a smoothing process using a combination of Gaussian and mean filtering is employed, and the results are validated against attitude sensor measurements. Experimental results demonstrate the effectiveness of the proposed method. Comparisons with binocular methods under variable-speed and constant-speed motion states yield the following findings: In variable-speed motion, the maximum pendulum angle is 1.871°, with a maximum angle error of 0.184°compared to the sensor. This represents an accuracy improvement of 0.018°over the binocular method. In constant-speed motion, the maximum pendulum angle is 3.075°, with a maximum angle error of 0.259°compared to the sensor, showing an accuracy improvement of 0.021 over the binocular method. Moreover, the computational efficiency is enhanced by 133.3%. When camera position deviations occur, the calibrated method produces maximum angle errors exceeding 0.5°compared to the sensor, while the non-calibrated method maintains errors below 0.3°. This non-calibrated approach effectively eliminates deviations, enabling accurate measurement of dynamic pendulum angles.

Abstract:The stability of transmission lines is a crucial guarantee for the normal operation of the power grid. To prevent accidents caused by accidental contact with conductors during line construction, this paper proposes a feature extraction network based on a multi-branch dual attention mechanism, DAMF-NET, addressing the low accuracy and poor reliability of existing detection methods. This algorithm enhances the network’s focus on local features of target information by constructing a multi-branch dual attention mechanism, optimizing the feature extraction process. A multi-branch lightweight feature fusion network is proposed to reinforce the global multi-scale semantic information and feature significance under dense tasks, thereby improving the completeness of image features. A small object detection network is introduced to mitigate network scale variance and enhance the sensitivity of small object detection. By employing focal loss and EIoU optimized loss functions, the method reduces noise generated by positive and negative sample imbalance, accelerating the convergence speed of model training. Finally, a state recognition algorithm based on risk area localization is designed and deployed in the intelligent detection system of construction machinery. Experiments show that this method has better average precision compared to most current detection models, indicating its research significance in the detection of construction machinery and intelligent inspection.

Abstract:To reveal the effects of different signal types, as well as their frequencies and amplitudes, on the accuracy of parameter identification and control performance in the signal injection method for permanent magnet synchronous motor，a comparative analysis through simulation and experimentation was conducted on four signal injection methods: square wave, trapezoidal wave, triangular wave, and sinusoidal wave. Firstly, according to the characteristics of the injected signals, square and trapezoidal waves are categorized as locally constant signals, whereas triangular and sinusoidal waves are categorized as globally time-varying signals. Based on the forgetting factor recursive least square, two categories of parameter identification models for signal injection methods are constructed. Secondly, simulation analysis is performed to investigate the effects of the frequency and amplitude of the two categories of injected signals on the identification results of the parameters pending identification. Comprehensively considering the identification accuracy of each parameter requiring identification, the frequency and amplitude of the injected signals is judiciously chosen. On this basis, comparative analysis of the parameter identification results is conducted and the control performance of the system is evaluated among the four signal injection methods. Finally, an experimental platform is set up to validate the parameter identification through experiments. The outcomes demonstrate that although parameter identification accuracy of locally constant signal injection is higher than that of globally time-varying signal injection, the former has a greater influence on control performance of the system compared to the latter. Among the tested signal injection methods, trapezoidal wave injection achieves the highest identification accuracy, while triangular wave injection has the least effects on the control performance of the system.

Abstract:To address the challenge of separating pipeline leakage signals from complex background noise and the difficulty of extracting small leakage features, a denoising method that uses the RIME to improve VMD is proposed. This method calculates the Bubble entropy value of the denoised signal for feature extraction, and then identifies the small leakage condition of the pipeline using an improved ELM optimized by RIME. First, RIME is used to improve the selection of key parameters for VMD, achieving adaptive decomposition. The mutual information entropy value between the IMFs generated by VMD is used as the fitness function value in the parameter optimization of the Rime algorithm, establishing a denoising method for water pipeline leakage signals based on RIME-VMD. Experiments have shown that compared to other heuristic optimization algorithms, the RIME-VMD method has the highest SNR of 23.922 dB, indicating that the reconstructed signal filtered by this method has the highest proportion of effective signal. The RIME-VMD method also has the lowest MAE and MSE, at 0.187 and 0.056 respectively, indicating that the reconstructed signal contains the least noise. Second, a method for extracting features from pipeline micro leakage signals using Bubble entropy is proposed, and these features are input into a model with ELM parameters optimized by RIME for pipeline leakage detection. Ultimately, the classification accuracy of the RIME-ELM model reached 95.71%, which is a 37.4% improvement compared to directly inputting fault feature vectors into the ELM, verifying the effectiveness of the proposed method.

Abstract:The whale optimization algorithm (WOA) is a highly competitive and efficient swarm intelligence optimization algorithm. In comparison to other intelligent optimization algorithms, WOA offers a simple structure, fewer parameters, and robust optimization capabilities. However, the conventional WOA exhibits slow convergence and falls into local optima easily. To address these issues, this paper proposes an improved whale optimization algorithm (IWOA). The algorithm adopts an adaptive update mechanism, inspired by particle swarm optimization, incorporating the individual’s historical best position during the optimization process, and dynamically adjusts the weights of the global best and individual best positions through an adaptive strategy to avoid getting trapped in local optima; at the same time, through neighborhood search strategy, neighborhood updates are carried out around the global historical optimal position in the later stage of iteration to improve the algorithm’s optimization ability. 16 typical benchmark test functions and 8 composite functions from the CEC2014 test set are selected for simulation experiments, compared to other traditional and improved swarm intelligence optimization algorithms, IWOA demonstrates superior convergence accuracy and speed, validating its effectiveness; and IWOA is applied to two engineering design problems, welding beam and pressure vessel design, compared with WOA, the economic cost is saved by 3.94% and 5.58%, respectively, verifying the effectiveness of the algorithm.

Abstract:Nonlinear elements cause significant modeling errors and control delays in speaker control processes, which affect the precise control of the speaker voice coil’s motion. This not only improves sound quality but also reduces mechanical wear and aging. This paper addresses the problems of modeling errors and control delays in the fine control of the speaker’s voice coil. We design a backstepping sliding mode controller based on an improved RBF-MLP neural network, solving the issues of control interference caused by nonlinear elements in electric speakers and the insufficient accuracy of the classical RBF network in fitting complex nonlinear models. By introducing perception layers, adaptive learning mechanisms, and generalized radial basis function, the improved RBF-MLP network reduces the mean squared error of nonlinear function fitting by more than 5% compared to the classical network, enhancing its ability to capture complex nonlinear characteristics of the speaker system and improving model fitting accuracy. A simulation environment was built to evaluate the control performance of the speaker system under different frequency, amplitude, and load conditions, focusing on control precision, system delay, and chattering problems. The experimental results show that under varying frequency and load conditions, the control delay is reduced to an average of 0.15 ms, and control errors are decreased by 39%. Furthermore, the improved control method maintains excellent robustness and stability under complex load and frequency variations. These results demonstrate the broad application potential of the improved controller in electric speaker control systems.

Abstract:Aiming at the poor robustness of deep reinforcement learning for fault diagnosis in strong noise interference environments, a reinforcement learning fault diagnosis method with noise interference environment adaptation is proposed. The efficient channel attention mechanism based deep residual shrinkage network (ECA-DRSN) is taken as the basic framework of Q-network to avoid the phenomenon of gradient vanishing caused by the complex structure of Q-network. In the ECA-DRSN，the efficient channel attention mechanism is used to adaptively adjust the softening threshold，and the dilated convolution is introduced in the convolution layer of the residual shrinkage unit to obtain the fault characteristics in different scales under the noise environment. Meanwhile, the exponential linear unit is used as the activation function to further enhance the noise robustness. A quantized reward function based on signal-to-noise ratio is designed to stimulate self-directed exploratory learning of Agent. Combining the dueling Q network learning mechanism with the prioritized experience replay mechanism, the optimal diagnostic strategy of agent is generated and applied to identify the equipment fault states under noise interference environments. Example analysis results show that the recognition accuracy of bearing and gearbox faults using the method of this paper can reach 98.13% and 93.45%, respectively, and has better robustness to different intensity noise and adaptability to the environment.

Abstract:Motor imagery is a classic research paradigm in the field of brain computer interfaces, which aims to study the information transmission and control of external electronic devices solely through brain imagination. The Common spatial paternal algorithm is an indispensable classic feature extraction algorithm in motor imagery research. This algorithm can obtain highly discriminative features by maximizing inter class variance, thereby obtaining models with good classification performance. However, the common spatial paternal algorithm is sensitive to noise and other interferences, and requires as much inter class information as possible, resulting in poor performance in non-invasive brain imaging research. To address this problem, a data processing algorithm based on phase information in frequency domain and trend information in time domain is proposed. The phase residual sequence is constructed using the instantaneous phase sequence and the empirical mode decomposition residual component of electroencephalogram signals. The algorithm maximizes brain neural activity information while eliminating interference from external or other noise, and extracts more discriminative features through common spatial paternal algorithm, ultimately obtaining a classification model with strong recognition and generalization performance. The experimental results show that the proposed method has an average classification accuracy of 88.19% among 52 subjects, which is higher than the original sequence’s 79.67%. At the same time, it exhibits more stable classification performance in motor imagery data of different subjects, proving that the method has good recognition and generalization ability in electroencephalogram-based motor imagery classification.

Abstract:High-tensile bolts are critical components in key structures such as suspension bridges and wind turbine towers, where the quality of their installation significantly impacts the stability and safety of the entire structure. However, traditional torque methods struggle to accurately measure the axial preload of bolts, making it difficult to assess structural stability effectively. To address this issue, a bolt axial force detection system was developed based on the Acoustoelastic effect, incorporating circuits such as high-voltage excitation and voltage-controlled gain. The system detects echo signals through an excitation piezoelectric sensor and utilizes a cross-correlation algorithm to calculate the time-of-flight difference, which reflects the stress state of the bolt. Stress-transition time calibration experiments were conducted, along with comparative testing between the ultrasonic and torque methods for wind power bolts. The experimental results indicate that when the axial preload of a bolt reaches 40% of its rated value, the system achieves a stress measurement error rate of ≤2.81%, with a resolution of up to 0.250 7 kN. This meets the requirements for accurate axial preload measurement. Compared with the traditional torque method, the ultrasonic approach demonstrates a clear advantage in reducing measurement error and improving resolution during bolt service, offering a reliable technical solution for stress measurement in industrial applications.

Abstract:With the development of biomedical field, wireless implantable devices are playing an increasingly important role. Taking into account the power loss and specific absorption rate (SAR) of electromagnetic waves in human tissues, the selection of frequency for implantable electrodes is of great significance. To address this issue, CST software is used to build a high-precision voxel model of the human head, simulate and explore the transmission characteristics of electromagnetic waves in the commonly used human communication frequency band of 400 MHz~6.5 GHz in the head tissue, analyze the distribution of electromagnetic fields, power density losses, and radiation hazard SAR values. The results show that the path loss of electromagnetic waves in human head increases with the depth of the tissue, and the channel loss in the frequency band below 3.5 GHz is smaller, which power density decreases less than 10 dB in the tissue within 25 mm of the skin. The human tissue has different absorption capabilities for different frequencies of electromagnetic radiation, and the SAR value of high-frequency electromagnetic radiation is generally smaller than that of low-frequency electromagnetic radiation. Among them, the SAR of 1.8 GHz reached the maximum value of 1.71 W/kg. Considering the power loss and electromagnetic radiation hazards, the optimal operating frequency for implantable brain electrodes varies with the depth of implantation. 2.45 GHz electromagnetic wave is suitable for implantable electrodes in the subcutaneous tissue within 5 mm of the head, and 1.8 GHz and 400 MHz are the optimal frequencies for implantable electrodes located 15 mm and 25 mm beneath the skin, with the power density increasing by 9.6% and 77.4% compared to 2.45 GHz.

Abstract:Metal-magnetic memory detection technology is a new type of non-destructive testing technology, which can identify the early damage of ferromagnetic materials. In order to promote the application of this technology in the field of bridge steel damage detection, the static bending test was carried out for the single U-rib orthotropic steel bridge deck inside the steel box girder, the magnetic signal data of the U-rib web under different load levels were collected, and the stress was obtained by combining with ANSYS finite element software, and the force-magnetic curve of the U-rib web along the height direction and the relationship curve between the stress in the stress concentration area and the magnetic signal and magnetic signal gradient were established, and the relationship between the force characteristic parameters and the magnetic characteristic parameters was analyzed. The results show that along the direction of the detection line, the normal and tangential magnetic signal values of each detection line decrease with the increase of load, and the normal magnetic signal curve fluctuates less than that of the tangential magnetic signal under the same load level, and the distribution is more gentle. The normal magnetic signal of the U-rib web has a “peak-trough” phenomenon along the height direction, which has a good correspondence with the stress distribution, and the stress concentration area can be judged by this phenomenon. The “steep rise” phenomenon of the magnetic gradient K curve in the stress concentration area can be taken as the danger warning signal of the yield strength in the actual engineering component. The μσ-λk curve of the force-magnetic correlation characteristic parameter has a good linear fit, and the stress concentration degree of the stress concentration area can be preliminarily judged according to the curve.

Abstract:For the problem of occlusion interference in robot grasping of occluded targets, an improved YOLO-CA-SD and semantic segmentation occluded target detection model and grasping method are proposed to complete the grasping when multiple targets and non-target objects occlude and interfere with each other. Firstly, the model adds a coordinate attention to YOLOv5l, considers the problem of detection frame matching direction based on the loss function, adds angle information between frames, and detects the original model decoupling is partially performed to reduce information loss caused by coupling. Secondly, an improved DeeplabV3+ target segmentation model was proposed. The original DeeplabV3+ backbone network was replaced by MobileNetV2 to reduce the model complexity. A CA module was added to the Atrous Spatial Pyramid Pooling structure to fuse pixel coordinate information to improve segmentation accuracy and solve the occlusion interference problem. Finally, the end rotation angle of the target poses relative to the template pose and the optimal grasping point are obtained by point cloud registration. The performance test is carried out on the self-built 2 750 commonly used tool occlusion data set. The experimental results show that the improved model improves the detection accuracy by 0.052%, 0.968%, 6.000%, and 7.400% on mAP@0.5, mAP@0.5:0.95, 60% target object occlusion rate and 60% non-target object occlusion rate datasets. The improved semantic segmentation model on this basis improves the segmentation speed and MIOU by 33.45% and 0.625%, and the ABB IRB1200 robotic arm is used to realize the experiments on the grasping of obscured targets, which verified the feasibility and practicality of the method.

Abstract:The accuracy of reservoir porosity prediction is crucial for assessing the storage capacity and quality of underground reservoirs. However, existing methods for porosity prediction face challenges such as limited model algorithms, low accuracy, and poor generalization. To enhance the precision of porosity prediction, this study proposes a Stacking ensemble learning method optimized by Optuna. First, gray relational analysis is used to select input parameters, including acoustic time difference, well depth, rock density, dip angle, and photoelectric absorption index. The input data is then normalized, and Optuna is employed to optimize the model parameters. Based on metrics like root mean square error, mean absolute error, and determination coefficient, random forest, support vector regression, and K-nearest neighbors are chosen as base learners for the Stacking model, with elastic net regression serving as the meta-learner. Comparative results reveal that while RF excels in handling nonlinear data, it shows instability in predictions; the Stacking model reduces RMSE by approximately 10% compared to RF. SVM demonstrates strong generalization ability but requires complex parameter tuning, with the Stacking model achieving about a 39% reduction in RMSE compared to SVM. KNN is insensitive to outliers but performs poorly on high-dimensional data, with the Stacking model lowering the error by about 21% compared to KNN. Additionally, XGBoost effectively avoids overfitting but is sensitive to outliers and requires complex tuning, with the Stacking model reducing the error by approximately 30% compared to XGBoost. Overall, the results indicate that the Optuna-optimized Stacking model significantly improves the accuracy of porosity prediction, providing a valuable reference for evaluating the oil and gas storage capacity of reservoirs.

Abstract:Aiming at the problem that the positioning accuracy of satellite navigation system in indoor closed places is too low due to insufficient positioning penetration and the traditional inertial guidance has a large trajectory offset in indoor pedestrian positioning, through the in-depth analysis of pedestrian trajectory projection algorithm (PDR), an improved PDR algorithm is proposed, which aims to improve the positioning accuracy in indoor positioning. The algorithm firstly designs Kalman filter and FIR filter to preprocess the sensor data to improve the data smoothing and anti-noise performance; secondly, it improves the traditional Weinberg step calculation model by adding new variables as the joint reference of step frequency detection and step calculation, which helps to reduce the cumulative error of the step estimation; and then it takes the appropriate threshold value as the pedestrian zero-speed judgment to correct the step number as well as the pedestrian’s position; finally, an extended Kalman filter (EKF) is designed to optimize the pedestrian’s position to achieve the dynamic optimization of the actual pedestrian trajectory. The simulation results show that the improved PDR algorithm significantly improves the positioning accuracy, and the average error of the pedestrian trajectory is reduced from 5.5 m to 1.2 m. Overall, the improved PDR algorithm can effectively reduce the trajectory deviation and the cumulative error, and improve the accuracy of the pedestrian positioning, which is of wide application prospect.

Abstract:In response to the problem of falls caused by incoordination between the walking speed of the elderly and the designated speed of the walking rehabilitation training robot during rehabilitation, this paper proposes a human-robot speed coordination anti-falling method, consisting of two parts: a falling prediction model and an anti-falling control method. First, the walking posture signals of the subject are collected by an inertial measurement unit (IMU), and a falling prediction model for the elderly is constructed using long short-term memory (LSTM) network and attention mechanism. Second, based on the falling prediction, a multi-dimensional phase space reconstruction (PSR) speed prediction model is designed for the anti-falling controller. Finally, the predicted speed of the subject is used as the target speed, and the PSR theory and model predictive control (MPC) technology are used to design an anti-falling controller for the walking rehabilitation training robot, achieving precise tracking of the subject’s walking speed and preventing falls caused by incoordination between the subject’s walking speed and the robot’s designated speed during rehabilitation training. Simulation and experimental results show that the prediction accuracy of the falling prediction model can reach 95.2%, and the lead time for falling prediction can reach 1.82 s. The human-robot speed coordination anti-falling method can effectively prevent falls caused by incoordination between the subject’s walking speed and the robot’s designated speed, enabling the subject to complete walking rehabilitation training safely.

Abstract:The double-exponential asymptotic sliding mode control is applied to the single-phase three-level inverter system, which exhibits a multitude of operating modes and intricate nonlinear dynamic behavior. The operational principles of the system across various modes are thoroughly analyzed, and a discrete model of the system is established utilizing the flash mapping method. The impact of system parameters on its nonlinear behavior is investigated through bifurcation diagrams and folding diagrams, leading to the identification of a two-dimensional stable operation domain for both control parameters and main circuit parameters. The stability of the system under double-exponential sliding mode control is examined using the fast-varying stability theorem, with comparisons made against bifurcation and folding diagrams for validation, as well as against other control strategies. Finally, the nonlinear behavior of the system is corroborated through time-domain waveform diagrams and their corresponding spectra under varying control parameters. The study reveals that the three-level inverter topology demonstrates more complex nonlinear dynamics; furthermore, it establishes that double-exponential asymptotic rate sliding mode control offers an expanded parameter stable domain. Specifically, the stable operational range for control parameters has been extended from 0.15~0.95 to 0.05~1.65, while shifting the unstable starting point from 1.3 (under improved exponential asymptotic rate sliding mode control) to 1.65. These research findings provide theoretical support for parameter design in implementing double-exponential asymptotic rate sliding mode control within three-level inverters.

Abstract:Partial discharge ultrasonic signal monitoring is one of the commonly used methods to determine the insulation status of oil-immersed transformers. However, the on-site noise interference is difficult to avoid and often accompanied by white noise. Therefore, a denoising method based on improved variational mode decomposition and wavelet transform is proposed. Firstly, taking the kurtosis-permutation entropy criterion as the objective function, the ant colony optimization is used to determine the optimal decomposition level and penalty factor of the variational mode decomposition, and the noisy partial discharge ultrasound signal is decomposed into multiple intrinsic mode function. Then, the correlation coefficient method is used to divide the multiple intrinsic mode function into noise free function, noise containing function, and noise function. The maximum-minimum permutation entropy criterion is used as the objective function, and the ant colony optimization is used to determine the optimal wavelet threshold and propose an improved wavelet threshold function for wavelet denoising of the noisy function. Finally, the noise free function and the denoised wavelet function are reconstructed to complete the denoising of the partial discharge ultrasound signal. By denoising simulated and measured partial discharge ultrasound signals and comparing with four other denoising methods, the results show that the proposed denoising method has excellent performance. The signal-to-noise ratio and normalized correlation coefficient are average improved by 43.62% and 2.39% respectively compared with other methods, and root mean square error is average reduced by 35.46%.