A dynamic 3D reconstruction technology for construction sites was proposed to ensure safety and efficiency in the construction process. A Binocular camera was deployed to scan the reconstruction site in 3D to obtain the model base and target activity trajectory. The YOLOv8 model was enhanced with an attentional scale sequence fusion (ASF) module to form the YOLOv8-ASF framework, which improved the accuracy and performance of the model, to solve the pain points such as target occlusion and target loss. The improved semi-global block matching (SGBM) algorithm was fused, and the YOLOv8-ASF-SGBM algorithm was integrated with the YOLOv8-ASF to achieve near-real-time target recognition and localization based on 2D images. The obtained depth information was used to 3D project the behavior trajectories of dynamic elements into the substrate, to realize the near-real-time and full-view monitoring of the real construction site. Experimental results show that the proposed technology reproduces the movement trajectory of construction dynamic elements in high-precision three-dimensional, and the relative error with the real motion trajectory of dynamic elements is less than 5%, which can realize high-precision full-view three-dimensional monitoring based on two-dimensional image and video information, and has good application scenarios and engineering value.

A multi-distortion type underwater image enhancement algorithm based on improved CycleGAN was proposed, aiming at the difficulties of underwater image blurring, low contrast and image distortion recognition caused by various factors such as scattering, absorption and color deviation. Firstly, in order to improve the image enhancement effect, Auto-Encoder+Skip-connection network structure was used in the generator of CycleGAN, and global color correction structure was added for global enhancement in terms of pixel as well as color, so as to better capture the color information in underwater images. Secondly, a multidimensional perceptual discriminator was designed to learn the global and local features of the image. This discriminator payed more attention to the local details of the image, effectively targeted scattering and color noise, perceived the image from a multidimensional space, and had a stronger ability to extract the features, thereby enhancing the accuracy of image discrimination. Finally, the experimental results on EUVP, UIEB and U45 datasets showed that the proposed method achieved better results, compared with other algorithms. In processing multi-distortion types of underwater images, the algorithm’s SSIM indicator was higher than that of the second place by an average of 1.57%, the PSNR indicator was higher by 1.836%, the UIQM indicator was higher by 1.324%, and the UCIQE indicator was higher by 1.086%. The proposed method performed well in processing color and noise details.

A multimodal feature fusion model based on non-smooth non-negative matrix factorization (nsNMF-PCNN-GRU-MSA) was proposed to address the problems of poor generalisation ability, single feature extraction mode and model uninterpretability in the fatigue driving detection methods. This model detected the level of driver fatigue by analyzing electroencephalogram (EEG) signals. A channel weighting module was designed in the shallow layer of the network, and the non-smooth non-negative matrix factorization (nsNMF) algorithm was introduced to compute the contribution of the electrode channels. A multimodal feature fusion module was designed in the middle layer of the network, where the Gramian angular field imaging method was introduced to map the 1D EEG data into a 2D image, and the spatio-temporal features of different modes were fused in parallel with the PCNN-GRU module. The multi-head self-attention (MSA) mechanism was fused in the deep layer of the network to complete the task of fatigue driving state classification. The experimental results showed that the fatigue detection accuracies of the model on the mixed samples of the SEED-VIG and SAD datasets were 93.37% and 90.78%, respectively, and the lowest accuracies for single-subject data were 86.60% and 85.59%, respectively, which were higher than those of the state-of-the-art models. The analysis method of mapping the feature activation values onto the brain topology map not only improves the interpretability of the model, but also provides a new perspective on fatigue driving detection.

In order to portray the public emotion orientation caused by the public opinion on traffic incidents disseminated in short videos, a physiological feature graph was constructed by the text sentiment analysis and the multimodal physiological signal feature extraction. This work collected 136 highly-liked videos with 38 805 comments on TikTok. Considering all videos as a document set, with each video treated as a document and comments as words, the latent Dirichlet allocation topic model was adopted to obtain the distribution of comments under different topics and the distribution of topics under different videos. Naive Bayes-based SnowNLP was utilized to calculate the sentiment scores of comments and analyze the sentiment tendencies expressed by different opinion topics. Neuroscience experiments were carried out to collect multimodal physiological signals such as EEG, eye movement, ECG, and respiration as well as emotion ratings. Statistical test results show that videos with different sentiment tendencies induce different emotions, and the multimodal physiological features such as the relative spectral power of EEG, blinking frequency, respiration standard deviation, and the very low-frequency power of ECG are specific under different emotions. The emotional semantics embedded in the comments influence public emotion in various ways beyond that evoked by videos.

A multi-scale parallel MRI reconstruction model based on a variational model and Transformer (VNTM) was proposed, to enhance the quality of reconstructed MR images from undersampled multi-coil MR data. First, undersampled multi-coil k-space data were used to estimate sensitivity maps, with an intermediate-stage enhancement strategy applied to improve the accuracy of these maps. Next, the undersampled multi-coil k-space data and estimated sensitivity maps were input into a variational model for reconstruction. In the variational model, resolution was reduced through a pre-processing module to reduce computational load; multi-scale features were then effectively fused through a multi-scale U-shaped network with the Transformer. Finally, a post-processing module was applied to restore resolution, and data consistency operations were performed on the output to ensure fidelity. Extensive quantitative and qualitative experiments were conducted on publicly available datasets to validate the effectiveness of the proposed method. The experimental results indicate that the proposed reconstruction model achieves superior reconstruction quality and more stable performance in terms of peak signal-to-noise ratio, structural similarity, and visual effects. In addition, a series of ablation studies and robustness evaluations with varying auto-calibration signal (ACS) region sizes were carried out, confirming that VNTM maintained consistently high reconstruction performance under diverse conditions.

A control strategy integrating braking force redistribution and path tracking was proposed to address the problem that instability and yawing were prone to occur when vehicles equipped with electromechanical brake (EMB) system experience braking failures during cornering. Gaussian perturbation and staged optimization were introduced to improve the algorithm in order to mitigate the deficiencies of the slime mould algorithm (SMA). The enhanced SMA was employed to optimize the weight matrix of the linear quadratic regulator (LQR). The improved LQR algorithm was utilized to compute the vehicle’s yaw moment upon detection of a single-wheel failure in the EMB system, followed by braking force redistribution to maintain vehicle stability. The pure pursuit algorithm was modified by shifting the tracking control point to enhance the response speed of the algorithm. An adaptive fuzzy control algorithm was incorporated to accommodate dynamic factors such as road conditions and vehicle speed, thus improving its adaptability. Path tracking was implemented to guide the vehicle along a predefined trajectory until a safe stop when a double-wheel failure was detected in the EMB system. The experimental results demonstrated that the maximum lateral deviation was reduced by 59.15% for single-wheel failure and by 41.95% for double-wheel failure compared with conventional methods. The proposed control strategy can more effectively ensure driving safety during cornering braking failure.

Existing anomaly detection methods of time series data focus on extracting the temporal variation features, while the spatial dependency features between multiple variables are ignored. To address this problem, a detection method based on a spatiotemporal graph attention network was proposed. The original multivariate time series data were transformed into a time-series graph with spatiotemporal dependencies, and a spatiotemporal graph attention network was designed to separately extract the temporal variation features and spatial dependency features. The periodic patterns of fused spatiotemporal features were learned by a multilayer perceptron, and an anomaly detection was performed based on the anomaly scores between prediction values and observation values. Experimental results on public datasets showed that the proposed method significantly outperformed state-of-the-art baseline methods in terms of anomaly detection accuracy and robustness.

A finite element numerical model of the segregated foundation pit was established based on the case of a deep foundation pit in Hangzhou adjacent to an operating underground shield tunnel in order to analyze the influence of the construction sequence, the separation wall location and other factors on the deformation of deep and large foundation pits and adjacent facilities caused by the segregated-pit construction. The reasonableness of the parameters of the HSS model was verified by combining with the measured data. The influence of the construction sequence of the "platform" type segregated pit on the displacements of out-of-pit strata and existing adjacent tunnels were analyzed by combining with a simplified model based on the case. Results show that the displacements of strata and tunnels caused by the excavation of the segregated pit in Hangzhou soft soil are related to the construction sequence, the location of the separation wall, the thickness of the soft clay, and the relative position of the tunnel and the pit. The deformation of the close pit retaining wall, the surface settlement and the tunnel displacement will be greater with a wider far sub-pit when the close sub-pit is firstly constructed. An opposite finding is observed if the far sub-pit is firstly excavated, and the optimal control effect on the deformation of the retaining wall and adjacent tunnels is achieved by dividing the ratio of the far sub-pit width to the close one by 3.0 to 4.0 and the width of the close sub-pit by 15 m to 20 m. The deformation of the close pit retaining wall, the surface settlement and the tunnel displacement caused by the two sub-pit construction sequences will increase as the thickness of the soft clay layer increases. The concept of the displacement impact zone resulting from different sub-pit construction sequences was proposed, and the demarcation line of the zone can be simplified to be a straight line with an angle of 45° to the wall of the pit. The range of the displacement impact zone which is defined as the strata displacement caused by the close-first-then-far construction sequence is smaller than that of the far-first-then-close construction sequence gradually decreases with the increase of the width of the far sub-pit and the thickness of the soft clay layer. A parametric analysis was conducted to propose formula for fitting the demarcation line of the impact zones related to the location of the separation wall and the thickness of the soft soil layer.

The mechanical and electrochemical characteristics of LiFePO₄ battery under different temperature and electric field were analyzed in order to introduce the in-situ surface expansion force as an additional input variable for the estimation of state of charge (SOC) and thus improve the estimation accuracy. A multi-physics signal acquisition platform was designed and constructed. Open-circuit voltage (OCV) tests, hybrid pulse power characterization (HPPC) tests, and in-situ surface expansion force measurements were conducted at different temperature. The mechanical and electrochemical characteristics of battery and its multi-physics responses under various operating conditions were analyzed. Results show that the in-situ surface expansion force first increases, then decreases, and then increases again as SOC rises, and it is more sensitive to SOC than OCV. The extrema of the expansion force curves are slightly affected by temperature, showing small delays with increasing temperature. They are strongly affected by current, occurring earlier and gradually disappearing as the current increases. The internal resistance decreases significantly with increasing temperature. The OCV curves exhibit high consistency across different temperature. The experimental results demonstrate that the expansion force signal has potential in SOC estimation and provide theoretical foundation and data support for SOC estimation methods based on expansion force signals.

A LLC resonant three port DC-DC converter with integrated photovoltaic and storage design and its advanced control strategy were proposed, for the application requirements of solar powered UAV’s energy manager. Firstly, time-domain analysis was used to analyze the multiple operating modes of the resonant tank of the three-port converter under different power transmission modes. Phase shift control was used to achieve the flexible power control among the three ports. Secondly, polynomial approximation was used to fit the gain surface obtained from time-domain analysis to obtain an accurate mathematical expression for the gain characteristics of the converter. On this basis, a decoupling control strategy was proposed. The design of the decoupling loop could effectively reduce the power coupling degree between multiple control loops of the three-port converter and optimize its dynamic performance. Finally, a 500 W experimental prototype was built, to verify the steady-state operating characteristics, dynamic mode switching process, and decoupling loop design of the three-port topology. The experimental results verified that the time-domain analysis method could accurately describe the circuit characteristics, and the decoupling loop could effectively reduce the degree of power coupling between control loops and improve the dynamic response performance of the system.

A prediction method based on the multi-channel graph aggregated attention mechanism was proposed, to address the challenges of limited spatial scope, insufficient spatiotemporal information capture, and low accuracy in short-term bike-sharing demand prediction. Firstly, the city was divided into multiple bike-sharing virtual stations using a flow-adjusted virtual station partitioning method according to bike flows in different areas. A dynamic adjacency matrix was constructed using the origin-destination (OD) matrix between stations to form a bike-sharing graph network structure. Next, spatial information of stations across different time periods was captured via a multi-channel graph aggregation module, which was combined with a multi-head self-attention module to capture temporal correlations. Finally, a cross-attention mechanism, along with exogenous variables, was introduced to uncover potential relationships among various variables. Experiments conducted in Shenzhen and New York demonstrated that the model significantly outperformed other deep learning methods across various time periods and regions, maintaining stable and low prediction errors. The results confirmed that the dynamic adjacency matrix and the cross-attention mechanism integrating external features could effectively enhance the prediction accuracy of shared bike usage.

A multi-goal multi-agent path planning algorithm was proposed to realize the efficient assignment of tasks to each agent and plan the shortest possible paths for the agents without collision with other agents. The definition of conflict between agents in continuous time and the way of conflict resolution were defined, and the concepts of safety interval and labeling were introduced based on A* algorithm aiming at the problem of low success rate due to the use of discrete time in traditional path planning algorithms. Then the A* algorithm can plan optimal paths that satisfy continuous time constraints. A conflict hierarchical strategy was proposed to reduce the number of nodes extended in the algorithm solving process aiming at the large amount of computation caused by collision detection and conflict avoidance in the multi-agent path planning problem. The experimental results show that the proposed algorithm can solve a better solution and has better applicability, with lower total path cost and higher success rate in the scenario of densely distributed agents.

A visual induced motion sickness (VIMS) estimation model based on attention mechanism was proposed to accurately assess the degree of VIMS experienced by users when interacting with virtual products. The model was constructed upon Transformer architecture, incorporating the self-attention mechanism within temporal and spatial sequences to capture the complex interactions between temporal and spatial features. By utilizing the optical flow information and user attention information, two sub-networks of motion flow and attention flow were designed to form a dual-flow network structure. The motion flow sub-network was responsible for capturing the motion features in the visual content, and the attention flow sub-network focused on extracting critical information, such as objects, textures, and other key elements within the user’s attention area. A late fusion strategy was employed to effectively combine the outputs of the dual-flow network. Experimental validation conducted on public video datasets demonstrated that the synergistic interaction between the attention flow sub-network and the Transformer architecture significantly enhanced the model accuracy. The VIMS model achieved optimal results in terms of the F1 score, accuracy and precision with values of 0.8468, 89.19% and 92.28%, respectively, representing a notable advancement over existing approaches.

An improved YOLOv5s sleeper crack target detection algorithm was proposed, in response to the safety hazards posed by the increasing number of crack defects in high-speed rail sleepers due to extended service life, as well as the issues of missed and false detections of surface fine cracks in high-speed rail sleepers. In the backbone network of the YOLOv5s algorithm, the full-dimensional dynamic convolution based on the multi-dimensional attention mechanism was used instead of the traditional convolution to enhance the overall feature extraction ability of the network and improve the detection accuracy of fine cracks. An improved lightweight C3 structure was proposed based on the ConvNeXt module and depth-separable convolution to compress the model volume and accelerate the convergence of the network to improve the detection efficiency. The scale-optimized weighted GFPN feature fusion network was used to solve the problem of detail feature loss in the sampling process of small targets at multiple scales. The improved YOLOv5s sleeper crack target detection algorithm could solve the problem of missed detection of fine cracks on the sleeper surface effectively. The experimental results showed that the parameter count of the improved algorithm model was decreased by 19.7%, the accuracy rate, recall rate and mean average precision were increased by 1.8, 2.4 and 4.2 percentage points respectively, and the detection speed was up to 96 frames per second. The results verify that the proposed lightweight YOLOv5s-OCG algorithm model provides an effective solution for the real-time detection of surface cracks on sleepers.

A dual-channel graph neural network method was proposed for user-level fraud detection tasks on E-commerce platforms to address the limitations of existing approaches that overemphasized global modeling of user behavior while insufficiently exploiting comment information. Multi-dimensional user behavior was modeled through the construction of two complementary graphs: an entity interaction graph and a comment semantic graph. The entity interaction graph was designed to capture global interaction patterns based on purchase and rating behaviors, while the comment semantic graph was built to model time-sensitive semantic relations between comments for characterizing fine-grained behavioral features. Parallel modeling of the dual graphs was performed using graph neural networks. Dynamic interaction optimization between dual-channel features was achieved through an attention mechanism, and higher-order node features containing multi-hop neighborhood information were generated. A comprehensive user-level behavior representation was produced by adaptively fusing different neighborhood ranges and feature spaces with a multi-head additive attention mechanism. Experimental evaluations were conducted on public datasets to validate the proposed method, and significant improvements were observed in multiple evaluation metrics compared to traditional approaches. Results show that the proposed method effectively enhances fraud detection performance at the user level.

To address the challenges of high randomness on both the supply and demand sides, difficulties in integrating renewable resources, and low flexibility in industrial parks, a study on the optimal configuration of a steam heating system coupled with green electricity in industrial parks was conducted, considering time-of-use steam pricing. The aim was to enhance system economic efficiency and green electricity utilization level. A system demand response management strategy was developed based on the modeling of the steam heating system in a green electricity-coupled industrial park. By introducing time-of-use steam pricing, users were encouraged to adjust their steam usage behavior, thereby regulating user-side loads and improving system flexibility. Based on the optimized time-of-use pricing results, a bi-level optimal configuration model considering both planning and operation stages was established for the steam heating system in industrial parks. A case study based on a power plant in Zhejiang Province was conducted for validation. The comparative results showed that the proposed optimization method reduced the peak-valley load difference by 58.32%. Additionally, the optimized configuration scheme decreased the total system cost by 98200, 253700, and 182500 yuan under different scenarios with combined heat and power unit output limits of 70%, 60%, and 50%. This time-of-use steam pricing-based optimization approach provided valuable guidance for the low-carbon transition of steam heating systems in industrial parks.

A reinforcement learning-based cloud-edge collaborative computing resource scheduling algorithm, KNCS, was proposed aiming at the problem of insufficient resource utilization in cloud-edge collaborative computing scenarios due to imbalances in network and computational resources, as well as uncertainties in task types and arrival times. This algorithm achieved shorter transmission time, processing time, and turnaround time by comprehensively considering the state of network resource and computational resource. A unified information transmission platform was designed to aggregate information from computational nodes and various tasks, facilitating the definition of task dependencies, dynamically adjusting subsequent tasks based on the type of running tasks, and providing a more realistic task scheduling scenario. The experimental results show that the performance of the KNCS algorithm surpasses that of the default Kubernetes scheduling algorithm in cloud-edge collaborative computing scenario.

A linguistic failure mode and effects analysis (FMEA) method for risk evaluation based on personalized individual semantics was proposed, as the existing FMEA method has deficiencies in linguistic information expression, linguistic modeling, and factor assignment when assessing risk problems in complex situations. A group of evaluation experts used a distributed linguistic preference relation to evaluate failure patterns for each risk factor, while another group also used the distributed linguistic preference relation to evaluate the relative importance of evaluation experts and risk factors. The obtained linguistic preference relation was converted into a corresponding numerical preference relationship through the numerical scale model. A first-stage personalized individual semantic model was constructed to obtain the weight values of risk factors and evaluation experts, and a second-stage personalized individual semantic model was further constructed. Based on the solution results of the two-stage personalized individual semantic model, the final score of the failure mode was calculated, and the prioritization was completed. The risk assessment problems in the development process of a certain aero engine were selected for verification, and the results showed the feasibility and effectiveness of the proposed linguistic FMEA method. Experimental comparison with the uniformly distributed method shows that the personalized individual semantics obtained by the proposed method yield highly consistent risk-assessment outcomes, supporting reliability and accuracy in complex product development.

A desensitization design method for parallel robots considering multi-source uncertain hybrid perturbation was proposed aiming at the problem of optimal design of high-performance parallel robots. A probabilistic error model was established by using the first-order perturbation method for error modeling. The optimal dimension design parameters were obtained by using multi-target subregion meta-heuristic iterations after analyzing the high-value targets corresponding to the working subregion. A performance sensitivity index was constructed to optimally allocate the design tolerances. The sensitivity of maintenance to parameters was calculated by establishing an in-service accuracy performance sensitivity model, and a low-sensitivity preventive maintenance strategy was obtained. An additive manufacturing parallel robot was used as an example for validation. Results show that the static performance and dynamic in-service accuracy maintenance can be effectively improved via desensitization design.

Existing blockchain-based crowdsourcing methods for fingerprint data lack effective utilization of fingerprint distribution characteristics in data quality control consensus and incentive mechanism reward distribution, affecting data collection quality. To address this issue, a decentralized indoor positioning crowdsourcing data quality control method considering fingerprint distribution features was proposed. A data quality control consensus algorithm was designed based on fingerprint distribution characteristics. Characteristic parameters were estimated by retrieving historical data within identical label classes. User-submitted fingerprint vectors were chained only when their weighted mean square error with historical mean vectors was below a given threshold. Duplicate data were rejected to resist replay attacks. To protect user identity privacy, data were anonymously uploaded with the block ownership verified using the Schnorr protocol. An appropriate incentive function was determined according to the data quality errors between user-submitted and on-chain fingerprint distributions. Experiments on three fingerprint datasets (UJIIndoorLoc, MALL and WiFi-RSS) demonstrated that compared with the original datasets, fingerprint data filtered by the proposed method improved training accuracy of indoor positioning models by approximately 10, 3 and 10 percentage points respectively, effectively enhancing fingerprint data acquisition quality.