A thorough analysis and cross-comparison of recent relevant works was provided, outlining a closed-loop process for EEG data analysis based on deep learning. EEG data were introduced, and the application of deep learning in three key stages: preprocessing, feature extraction, and model generalization was unfolded. The research ideas and solutions provided by deep learning algorithms in the respective stages were delineated, including the challenges and issues encountered at each stage. The main contributions and limitations of different algorithms were comprehensively summarized. The challenges faced and future directions of deep learning technology in handling EEG data at each stage were discussed.

A field investigation was conducted on a heavy haul railway line with axle load of 25 t. There was a significant difference in rail rolling contact fatigue (RCF) on the entering and leaving transition sections on curves with radius of 580?1000 m, and particularly, the RCF on the leaving transition section was severer. A dynamic model of heavy haul train including two locomotives and 108 wagons was established using Simpack on the basis of on-site wheel/rail observation and train parameter investigation, and a damage function model was applied to numerically analyze the mechanism of the rail RCF difference on the entering and leaving transition sections. Result shows that the RCF difference is dominated by the curving behavior of wagons, and the contribution of the leading wheelsets is the most significant, while the contribution of the trailing wheelsets and locomotives is relatively slight. More detailed analysis shows that the RCF is not significant under the condition of standard wheel/rail profile matching. However, after the wagon wheels wear, the wheel/rail creepage and creep force on the leaving transition section are higher than those on the entering transition section, which is the primary reason for the RCF. And the effect of rail worn profile is not significant for the RCF. The frequent interaction between the worn wagons leading wheelsets and the worn rail on the sharp radius curve is the main reason for the rail RCF difference on the entering and leaving transition sections.

To deeply understand the mechanism of flow induced noise excitation source of waterjet pump and to support the low-noise design of waterjet pump, the acoustic performance of a two-stage propulsion pump was researched. Based on the measured radiated noise under multiple rotational speeds, the modulation mechanism of flow induced noise of this waterjet pump and the extraction method of flow induced noise source were studied. The radiated noise of this two-stage waterjet pump was measured in a cavitation tunnel, and the cyclostationary analysis was conducted to demodulate and extract the characteristic frequency of the flow-induced noise source. The unsteady numerical simulation of the internal flow field of the two-stage propulsion pump was carried out to analyze the distribution characteristics of the transient internal flow field and the characteristics of three-dimensional exciting force. Combining the signal processing results with the internal flow field simulation results, the key characteristics and formation principle of the flow induced noise source of the propulsion pump were studied. Results show that the guide vane plays a key role in the radiated noise characteristics of the two-stage waterjet pump, and the modulation intensity is decided by the incoming flow conditions and operating conditions of the waterjet pump. The matching design of the impeller and guide vane takes a decisive position in the noise and vibration control of waterjet pump.

In order to solve the bottleneck problems of sharing, transferring and reusing of enterprise knowledge resources, caused by the diversity of knowledge carriers and the heterogeneity of information for describing knowledge resources, a metadata model of process planning knowledge resource network (PPKRN) was presented. The formal definition of the PPKRN was introduced through analyzing the current situation of process planning knowledge resources of the enterprises. The model can represent knowledge in a unified standard way by the metadata technology, which is helpful to improve the ordering process of the PPKRN and build the dynamic correlation among knowledge resources effectively. A case on the development of a prototype system used in a Chinese engine enterprise was given to further illustrate the feasibility and validity of the model．

A method based on attention mechanism and multi-scale information fusion was proposed to resovle the problem of low accuracy of the traditional single shot multibox detector (SSD) algorithm in detecting small targets. The algorithm was applied to the vehicle detection task. The feature maps of the target detection branch were fused with 5 branches and 2 branches respectively, combining the advantages of the shallow feature map and the deep feature map. The attention mechanism module was added between the basic network layers to make the model pay attention to the channels containing more information. Experimental results showed that the mean average precision of the self-built vehicle data set reached 90.2%, which was 10.0% higher than the traditional SSD algorithm. The detection accuracy of small objects was improved by 17.9%. The mAP on the PASCAL VOC 2012 dataset was 83.1%, which was 6.4% higher than the current mainstream YOLOv5 algorithm. The detection speed of proposed algorithm on the GTX1 660 Ti PC reached 25 frame/s, which satisfied the demand of real-time performance.

A new multimodal sentiment analysis model (MTSA) was proposed on the basis of cross-modal Transformer, aiming at the difficult retention of the modal feature heterogeneity for single-modal feature extraction and feature redundancy for cross-modal feature fusion. Long short-term memory (LSTM) and multi-task learning framework were used to extract single-modal contextual semantic information, the noise was removed and the modal feature heterogeneity was preserved by adding up auxiliary modal task losses. Multi-tasking gating mechanism was used to adjust cross-modal feature fusion. Text, audio and visual modal features were fused in a stacked cross-modal Transformer structure to improve fusion depth and avoid feature redundancy. MTSA was evaluated in the MOSEI and SIMS data sets, results show that compared with other advanced models, MTSA has better overall performance, the accuracy of binary classification reached 83.51% and 84.18% respectively.

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.

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 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 traffic flow prediction models based on graph neural networks and attention mechanisms have shortcomings in capturing complex spatiotemporal dependencies, overcoming the constraints of predefined graph structures, and modeling periodic patterns. Thus, a multi-scale adaptive graph attention Transformer (MSAGAFormer) was proposed. Short-, medium-, and long-term historical traffic data were divided into low-, medium-, and high-scale temporal sequences, and a compression mechanism was employed to reduce redundant information and enhance the efficiency of temporal feature representation. A spatiotemporal embedding method was designed to encode node positions and temporal attributes, thereby strengthening the model’s capability to interpret spatiotemporal data. A GAT-based multi-head attention mechanism was utilized in the spatial layer to model dynamic spatial correlations, while a multi-scale temporal attention structure was incorporated in the temporal layer to capture dynamic variations across different temporal granularities. Experimental results on the PEMS datasets demonstrated that MSAGAFormer outperformed state-of-the-art models such as Trendformer, ATST-GCN, and STTN in prediction accuracy.

An aerial small target detection algorithm that balanced performance and resource consumption was proposed to address the issues of low detection accuracy and large model parameter size in small target detection of aerial images. On the basis of YOLOv8s, an adaptive detail-enhanced module (ADEM) was proposed by reducing the channel dimension and enhancing the focus on the high-frequency features to capture the fine-grained features of small targets while discarding the redundant information. A feature fusion network was optimized based on the PAN-FPN architecture to enhance the attention on shallow features. Multi-scale convolutional kernels were introduced to enhance the focus on the target contextual information, thereby adapting to the small object detection scenario. A parameter-adjustable Nin-IoU was constructed to overcome the limitations of traditional IoU in flexibility and generalization, and this adjustment achieved by introducing adjustable parameters allowed the Nin-IoU to be tailored to different detection tasks. A lightweight detection head was proposed to enhance the integration of multi-scale feature information while reducing redundant information transmission. Experimental results on the VisDrone2019 dataset indicated that the proposed algorithm achieved an mAP_0.5 of 50.3% with only 8.08×10⁶ parameters, representing a 27.4% reduction in parameters and an improvement of 11.5 percentage points in accuracy compared to the YOLOv8s benchmark algorithm. Experimental results on the DOTA and DIOR datasets further demonstrated the strong generalization capabilities of the proposed algorithm.

A GW-XGBoost model considering spatial heterogeneity was constructed, and the SHAP model was used to explain the extent and spatial differences in the role of built environment factors, in order to explore the influence of spatial heterogeneity on the nonlinear relationship between the built environment and bike-sharing trips. Compared with the geographically weighted regression and extreme gradient boosting tree models, the GW-XGBoost model significantly improved the explanatory and predictive power of the model by introducing geospatial weighting and adaptive bandwidth, with the overall goodness-of-fit increased by 15.59% on average, and it could reveal the intensity, direction and local differences of the built environment on the nonlinear impact of bike-sharing trips. The results showed that the built environment factors had a nonlinear impact on bike-sharing trips. When the population density reached 20000 persons per km², the impact turned from negative to positive. When the distance from CBD factor was between 15 and 20 km, its effect shifted from positive to negative, and then became stabilized when moving outward from the city center. When the floor area ratio reached 1.8, the impact effect turned from negative to positive. The research results provide a scientific basis and methodological support for the resource optimization of the urban bike-sharing system.

Heating units can be modified to operate in multiple modes such as high backpressure, extraction-condensing, and pure condensing, which significantly enhances the operational flexibility and economic performance of thermal power plants during the heating season. To select the operating mode of each heating unit in a thermal power plant before the heating season to maximize the overall operational benefits during the heating season, the operating characteristics of heating units under different modes were analyzed. A plant-wide comprehensive benefit comparison model for the heating season considering the capacity revenue, deep peaking market revenue, fuel cost, and the cost of replacing rotors in high backpressure units was constructed, to evaluate and compare the different operating schemes of the plant. A case study of an actual thermal power plant in Northeast China was conducted, where the proposed model was applied to calculate and compare the benefits of two typical operational schemes. The results verify the effectiveness and practicality of the model, demonstrating that the high backpressure mode and the extraction-condensing mode have good complementary characteristics in operation. The selection of the optimal operational mode of heating units heavily depends on the heating load level of the thermal power plant.

Utilizing idle computing resources distributed worldwide to train large language models provides a novel paradigm for supplying computing power while helping to reduce training costs. A blockchain-based large model fine-tuning compute-power sharing platform, CompuDEX, was proposed to address the challenges of privacy risks, malicious attacks and lack of trust associated with this method. The trustless and anonymous nature of blockchain technology was leveraged to create a transaction platform that eliminated the need for trusted intermediaries while protecting user privacy. The smart contracts and cryptographic tools were employed to encourage fair competition among compute power providers, further reducing training costs. A security scheme combined with a “Red Balloon” incentive mechanism was designed based on zero-knowledge proofs to identify and penalize the malicious behaviors of compute power providers. The structure of the fine-tuning method LoRA was split to ensure data privacy during the training process without introducing additional computational overhead. Experimental results showed that during the forward propagation phase of fine-tuning, the computational cost was only 8% to 14% of that charged by major domestic cloud service providers. By increasing the number of parallel nodes, the execution time was significantly reduced from 175% of the local training time to 20% or even lower.

To study the dynamic competitive relationship between ride-sourcing and cruise taxis in urban environments and their impact on traffic carbon emissions, the Herfindahl-Hirschman index (HHI) was introduced to quantify the competitiveness of urban taxis. A four-player game model encompassing the government and enterprises was constructed. Considering two distinct decision-making sequences, 16 potential game scenarios were proposed. Equilibrium points were computed using the Stackelberg model, leading to the identification of 8 stable game scenarios. Based on the game outcomes, carbon emissions were calculated by integrating the COPERT model and a modified Michaelis-Menten (T-M-M) equation. Taking Xi’an as a case study, relationships between vehicle numbers (including both ride-sourcing vehicles and cruise taxis) and vehicle kilometers traveled, as well as between vehicle fleet size and carbon emissions, were established. The theoretical optimal carbon emissions under the proposed game scenarios were calculated. Results show that maintaining an optimal ratio of ride-sourcing vehicles and cruise taxis within the range of 29∶35 to 1∶1 in Xi’an can effectively balance market stability and low-carbon objectives. The government can implement traffic restriction policies (e.g., driving bans based on license plates) to maintain a slightly larger fleet of cruise taxis relative to ride-sourcing vehicles, alongside introducing incentive policies to accelerate the electrification of the cruise taxi fleet.

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.

A new composite bent cap consisting of a shell made of steel plate and ultra-high-performance concrete (UHPC) and cast-in-place core normal concrete (NC) was proposed in order to realize the assembly and rapid construction of ultra-large-scale bent cap for urban viaducts or highway reconstruction and expansion projects. Parametric analysis of different UHPC and steel plate thickness was conducted in order to analyze the influence of the thickness of UHPC and steel mold plate on its stress performance. Results showed that the stiffness of the shell was affected by the thickness of UHPC and steel plate and their ratio together under the action of self-weight. The thicker the UHPC and steel plate are, the better the stress performance of the shell is, but the economy will be reduced when tensioning prestress and casting concrete. It is recommended to use UHPC thickness of 70 mm and steel plate thickness of 6 mm. A piece of 1∶2.5 scaled-down model was designed and static loading test was conducted in order to verify the feasibility and safety of this scheme. Results show that the new UHPC-NC composite bent cap has good force performance and high safety reserve, which can provide reference for the assembly construction of bent cap.

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.