Top Read Articles

Published in last 1 year |

In last 2 years |

In last 3 years |

All

Please wait a minute...


For Selected: Download Citations EndNote Ris BibTeX Toggle Thumbnails

Select

Carbon Emission Prediction in Transportation Industry Based on SD-ISSA-DALSTM WANG Qingrong, WANG Junjie, ZHU Changfeng, HAO Fule Journal of South China University of Technology(Natural Science Edition)    2025, 53 (5): 66-81.   DOI: 10.12141/j.issn.1000-565X.240356 Abstract （3484）   HTML （38）    PDF（pc） （3678KB）（155）       Knowledge map    Save Aiming at the low accuracy of carbon emission prediction caused by the high volatility and nonlinearity of the carbon emission data series in transportation industry, a transportation carbon emission prediction model combining the secondary decomposition, dual attention mechanism, improved sparrow search algorithm (ISSA) and long short-term memory (LSTM) network is proposed. First, complete ensemble empirical mode decomposition with adaptive noise is introduced to decompose the transportation carbon emission data series into modal components with different frequencies, then sample entropy is used to quantify the complexity of each component, and secondary decomposition is performed on the component with the highest entropy value via variational mode decomposition, which further weakens the volatility and nonlinearity of the transportation carbon emission data series. Next, in order to explore the correlation between transportation carbon emission and its influencing factors, a double attention mechanism-optimized LSTM (DALSTM) model is constructed, in which a feature attention mechanism is added to the input side of the LSTM to highlight the key input features. Meanwhile, a temporal attention mechanism is added to the output side to extract the key historical moments. Finally, the SSA algorithm is improved by combining the Circle chaotic mapping, the dynamic inertia weight factor and the mixed variance operator strategies, ISSA-DALSTM models are established for each component separately, and the predicted values of each component are reconstructed. By measuring the carbon emission data of China’s transportation industry from 1990 to 2019, it is found that the root mean square error, mean square error, and mean absolute percentage error of the proposed model are respectively 5.308 8, 3.566 1 and 0.443 9, which are better than those of other comparative models, thus verifying the validity of the proposed model. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Review of Multi-Level Urban Impacts of Shared Autonomous Vehicles ZHONG Shaopeng, LIU Ao, ZHAI Junnuo, FAN Meihan, LI Xiyao, LIN Yuan, LI Zhenhua Journal of South China University of Technology(Natural Science Edition)    2025, 53 (6): 104-118.   DOI: 10.12141/j.issn.1000-565X.240119 Abstract （2948）   HTML （26）    PDF（pc） （1338KB）（125）       Knowledge map    Save To gain a deeper understanding of the potential impacts of Shared Autonomous Vehicles (SAVs) on urban and promote the sustainable development of urban transportation systems, this paper conducted a comprehensive review and systematic analysis of the multi-level impacts of SAVs. The aim is to summarize the main contributions and shortcomings of previous studies and propose possible directions for future research. The review findings indicate that existing studies primarily focus on the short-term impacts of SAVs on the transportation system, including residents’ travel behavior and road traffic flow. However, there is relatively little research on the long-term impacts of SAVs, particularly concerning urban accessibility, environment, and energy. While some studies have revealed potential negative effects of SAVs, such as adverse impacts on the environment or accessibility, few have proposed targeted and effective development strategies. Additionally, in terms of methods, existing studies mainly rely on qualitative analysis or independent transportation demand models for projections and simulations, which have certain limitations regarding the reliability of the results. Future research should focus on developing integrated land use and transportation models combined with data-driven approaches to more precisely, comprehensively, and systematically characterize the long-term (negative) impacts of introducing SAVs on urban land use, the environment, and energy consumption. Additionally, targeted development strategies and responsive measures should be proposed to optimize the effectiveness of SAV deployment, mitigate potential adverse effects, and promote the evolution of urban transportation systems toward greater efficiency, intelligence, and sustainability. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Fusion Transformer Model-Based Segmentation Algorithm for Laser Point Cloud of Distribution Lines DAI Zhou, LIU Yan, MAO Xianyin, GUO Tao, XU Lianggang, CHENG Guixian Journal of South China University of Technology(Natural Science Edition)    2025, 53 (5): 139-146.   DOI: 10.12141/j.issn.1000-565X.240542 Abstract （2043）   HTML （32）    PDF（pc） （1833KB）（233）       Knowledge map    Save As laser point cloud models are crucial for distribution line inspection and management, most distribution channels have constructed laser point cloud models at present. With the increase of the number of models, extracting key component locations (e.g., conductors, insulators) becomes vital. In order to enhance the accuracy and efficiency of segmenting key components such as lines, towers and insulators, this paper presents a segmentation algorithm for laser point cloud of distribution lines based on a fusion Transformer model. Given the need for detailed features in the point clouds of distribution lines, a dual-channel parallel feature extraction module is designed to capture high-frequency and low-frequency features. The low-frequency features are processed via average pooling and a fusion Transformer-based extractor, while the high-frequency features are handled through max pooling and a multi-layer perceptron (MLP) module with convolutional layers. The feature vectors from both channels are then fused to improve the ability of detail feature extraction. Additionally, the fused features are fed back into the MLP module for further refinement, achieving precise point cloud target segmentation. Extensive experiments demonstrate the accuracy and effectiveness of the proposed algorithm. It has potential advantages in many aspects, such as improving the inspection accuracy of unmanned aerial vehicles, enhancing the level of automation, improving the robustness, integrating multi-source data and reducing inspection costs. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Lane Line Detection Algorithm Based on Deep Learning YUE Yongheng, ZHAO Zhihao Journal of South China University of Technology(Natural Science Edition)    2025, 53 (9): 22-30.   DOI: 10.12141/j.issn.1000-565X.240609 Abstract （1998）   HTML （72）    PDF（pc） （2888KB）（252）       Knowledge map    Save Aiming at the problem of lane detection accuracy of intelligent vehicles in complex scenes, this paper proposed a lane line detection algorithm which incorporates a multi-scale spatial attention mechanism and a path aggregation network (PANet). The algorithm first introduced the pre-anchored frame UFLD lane detection model and incorporated a feature pyramid enhancement module PANet with depthwise separable convolution to achieve multi-scale feature extraction of images. Next, a multi-scale spatial attention module was designed in the network framework and a SimAM lightweight attention mechanism was introduced to enhance the focusing ability on target features. Then, an adaptive feature fusion module was designed to perform cross-scale fusion of feature maps output from PANet by intelligently adjusting the fusion weights of feature maps at different scales, so as to effectively enhance the network’s ability to extract complex features. Finally, the application of TuSimple dataset detection proves that the proposed algorithm achieves a detection accuracy of 96.84%, representing a 1.02 percentage point improvement over the original algorithm, and outperforms conventional mainstream algorithms. Experimental results on the CULane dataset demonstrate that the proposed algorithm achieves an F1 score of 72.74%, outperfor-ming conventional mainstream methods with a 4.34 percentage point improvement over the baseline. Notably, it exhibits significant performance gains in extreme scenarios (e.g., strong illumination and shadows), confirming its superior detection capability in complex environments. In addition, the real-time test shows that the model infe-rence speed reaches 118 f/s, which meets the real-time demand of intelligent vehicles. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Dynamic Scheduling of Demand Responsive Transit Based on Model Predictive Control JIN Wenzhou, ZHANG Yong, SUN Jie Journal of South China University of Technology(Natural Science Edition)    2025, 53 (6): 77-90.   DOI: 10.12141/j.issn.1000-565X.240331 Abstract （1967）   HTML （26）    PDF（pc） （3833KB）（327）       Knowledge map    Save As a typical representative of the new mode of shared public transport, demand responsive transit (DRT) systems are facing the challenge of efficiently processing travel demand and real-time planning of vehicle routes. Traditional dynamic scheduling methods for DRT primarily focus on adjusting vehicle routes after demand has been realized, which often limits their ability to effectively respond to dynamic fluctuations in travel demand. Therefore, this study introduced a Model Predictive Control (MPC) approach and develops a dynamic scheduling model for DRT based on a multi-period rolling optimization framework. The model used potential future stage passenger flow information to optimize current stage scheduling decisions and timely re-planning according to the latest disclosed information to cope with the uncertainty and dynamic changes of demand. In terms of solution methods, this study integrated the adaptive large neighborhood search (ALNS) strategy to design the MPC-ALNS algorithm. It iteratively optimized the vehicle scheduling sequence through a two-phase heuristic approach. Numerical experimental results demonstrate that in ideal scenarios without prediction deviation, compared to traditional dynamic scheduling methods, the proposed method significantly reduces the total cost of the system by 14.54%. Even in a pessimistic scenario with a 30% prediction deviation, it still achieves a cost optimization of 5.27%, and various passenger service indicators show superior performance, indicating strong universal applicability in different stochastic environments. At the same time, the experiment further verified the stable optimization performance of the method in dealing with different orders and vehicle scales, and analyzed the sensitivity of the rejection cost and proposed the setting idea of the optimal rejection cost suitable for different operating scenarios. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Research on the Joint Optimization of Shared Bikes and Demand-Responsive Connector XU Hang, LI Xin, YUAN Yun Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 20-28.   DOI: 10.12141/j.issn.1000-565X.240455 Abstract （1961）   HTML （30）    PDF（pc） （2642KB）（124）       Knowledge map    Save Demand-Responsive Connector (DRC), as a flexible public transportation mode, can provide personalized bus services according to passengers’ needs and has been widely applied in urban areas both domestically and internationally. However, in actual operation, it faces the dilemma of balancing service efficiency and operational costs, as well as the challenge of achieving “door-to-door” services. To address these issues, a joint travel mode combining shared-bike transfer and DRC was proposed. By integrating the advantages of shared bikes and DRC, the coupling optimization of the two transportation modes can be realized, thereby improving the overall service efficiency and service level of public transportation. To this end, based on the continuous approximation method, discrete demand points and shared bicycle deployment locations were continuousized. The operating costs of the transit system, shared bicycle costs, and passenger travel time costs were derived and calculated. By minimizing the total system cost, the joint mobility system was optimized. With the goal of minimizing the total system cost, the coupling optimization of shared bicycles and demand-responsive buses was realized. To verify the effectiveness of the proposed joint travel system, an empirical study was conducted using the university town area of Chongqing as a case. The operation of the joint travel system under different scenarios was simulated and compared with the traditional DRC system without shared-bikes. The results show that the joint travel system can effectively address the operational problems of DRC. Compared with the traditional DRC system, the joint travel system can reduce the total system cost by up to 14.8%, the travel time saving by 15.2%, and the detouring saving of DRC vehicles by 29%. It is demonstrated that introducing shared bicycles as a first- and last-mile connection tool in demand-responsive transit systems can significantly reduce transit operating costs and passenger travel times. At the same time, it minimizes unnecessary detours by transit vehicles, optimizes transit routes, and greatly improves the efficiency and quality of public transportation services. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

A Study on the Optimization of Modular Autonomous Public Transit Services ZHANG Jiyu, TANG Chunyan Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 42-49.   DOI: 10.12141/j.issn.1000-565X.240355 Abstract （1668）   HTML （29）    PDF（pc） （1672KB）（90）       Knowledge map    Save With the rapid development of intelligent connected technologies and autonomous driving, the emerging autonomous modular buses have attracted significant attention in the public transportation field. The autonomous modular buses can achieve flexible design of bus capacity through freely coupling/decoupling vehicles adapting to the uneven distribution of passenger demand in terms of space and time. However, the existing full-route service mode fails to fully leverage the flexible operational characteristics of modular buses to efficiently meet passengers' differentiated needs. Therefore, this study proposed a new service mode for autonomous modular buses, which combined the coupling/decoupling characteristics of modular buses with the differentiated service advantages of the skip-stop strategy to realize efficient and differentiated bus line supply. This paper first employed a discrete-time modeling approach combined with the extended Newell’s theory to develop an optimization model for autonomous modular bus skip-stop services, aiming to minimize passenger travel costs and operator costs. The model simultaneously optimized departure intervals, vehicle groupings, and skip-stop schedules.Firstly, this study adopts a discrete-time modeling method and an extended Newell theory to develop an optimization model for autonomous modular bus skip-stop service mode, with the optimization objectives of minimizing passenger travel costs and operational costs for the agency. It can simultaneously optimize bus headways, vehicle formulation, and skip-stop plans. By extending the Newell theory, the model expands from calculating passenger waiting and travel times at individual bus stops to efficiently calculating these times from the entire bus line system perspective, significantly reducing the modeling complexity. Secondly, taking Bus Route 110 in Dandong as a case study, an optimized operational scheme was proposed and compared under both off-peak and peak periods with the traditional fixed-capacity bus service model and the full-route modular bus service model. The results show that the proposed modular bus skip-stop service mode can greatly reduces the total system cost, saving 3.34% to 24.65%. Specifically, passenger waiting time costs and travel time costs are reduced by 7.49% to 48.52% and 2.31% to 6.28%, respectively. Moreover, during peak hours, modular bus dispatching is more frequent than during off-peak periods, with a tendency to adopt low-capacity vehicle groupings and skip-stop service strategies. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Modeling Methodologies for Unmanned Aerial Vehicle Path Planning in Emergency Rescue: A Comprehensive Review and Prospect PEI Mingyang, SHAO Kangshun, LI Linqing, XU Fengjuan Journal of South China University of Technology(Natural Science Edition)    2025, 53 (12): 17-33.   DOI: 10.12141/j.issn.1000-565X.250037 Abstract （1558）   HTML （28）    PDF（pc） （2019KB）（91）       Knowledge map    Save With the rise of the low-altitude economy, the application scenarios of unmanned aerial vehicle (UAV) continue to expand, particularly playing a significant role in emergency response. Leveraging advantages such as high mobility and remote control, UAV has proven to be powerful tools in disaster monitoring, communication restoration, personnel search and rescue, material delivery, and post-disaster assessment during emergencies such as natural and human-made disasters. This paper aims to provoide a comprehensive review of modeling methods and the latest research progress in UAV path planning for emergency rescue, offering thorough theoretical references and technical guidance for researchers in related fields. It begins by outlining typical emergency rescue scenarios such as earthquakes, fires, and floods, summarizing the application requirements of UAV in different contexts. Then it systematically reviews UAV path planning modeling methods, including dynamic models and task models, with task models further categorized into hierarchical, collaborative, fault-tolerant, real-time, and adaptative dimensions. Subsequently, it comprehensively analyzes path planning optimization algorithms based on three core elements: constraints, optimization objectives, and solution algorithms. Finally, the paper discusses the challenges and opportunities of UAV path planning in emergency rescue, highlighting that technological development, multi-UAV collaboration, and interdisciplinary integration represent future development opportunities. This study provides theoretical support and practical reference for the further development and application of UAV path planning modeling. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Collaborative Control for Urban Expressway Mainline and On-Ramp Metering in Connected-Vehicle Environment WU Haodu, SHI Yang, ZHAO Junteng, SUN Jian Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 73-86.   DOI: 10.12141/j.issn.1000-565X.240403 Abstract （1529）   HTML （29）    PDF（pc） （3829KB）（133）       Knowledge map    Save With the increasing application of Connected and Autonomous Vehicle (CAV) technologies in active traffic management, Variable Speed Limit (VSL) strategies have become crucial for improving traffic flow efficiency and safety. To address the issues of decreased traffic capacity and abrupt speed variations caused by traffic conflicts in urban expressway merging areas, a cooperative variable speed limit (VSL) control strategy was proposed for the mainline and on-ramp under a connected vehicle environment. Firstly, a mainline traffic flow prediction model based on Motorway Traffic Flow Network Modle (METANET) was adopted, constructing a bi-objective function to minimize the total travel time and distance, using Model Predictive Control (MPC). Then, the variable speed limit control problem was modelled as a Markov decision process, with a composite reward function based on average speed, throughput, and vehicle delay. By introducing Deep Q-network (DQN), the optimal on-ramp speed limits under different traffic flow conditions were calculated and disseminated to CAVs through Vehicle-to-Infrastructure (V2I) communication. Finally, the proposed coordinated control strategy was simulated and tested using the North Third Ring Expressway in Xuzhou, China as a case study. The empirical results based on SUMO microsimulation demonstrate that the proposed strategy, compared to the scenario with speed control only on the mainline, reduces the total travel time of network vehicles by 8.51%, increases the average speed by 14.49%, and reduces traffic density fluctuations by 14.81%. These results demonstrate that the proposed method can effectively improve traffic flow efficiency in merging areas under a connected vehicle environment, reduce speed differences between mainline and ramp vehicles, and shrink the spatiotemporal scope of congestion, thereby enhancing the stability of urban expressway traffic flow. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Research on Collaborative Transfer Under the Condition of Urban Rail Transit Passenger Flow Control WANG Bao, LUO Xia, QIAO Xuan, SU Qiming Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 11-19.   DOI: 10.12141/j.issn.1000-565X.240362 Abstract （1313）   HTML （27）    PDF（pc） （2705KB）（67）       Knowledge map    Save To address the current lack of attention to the transfer of restricted passenger flows under urban rail transit network flow control scenarios, this study investigated the routing and capacity allocation of transfer vehicles under specific flow control conditions. Firstly, the utility and selection probability of passengers opting for transfer vehicles were analyzed and quantified across various route conditions. Then, a model for the design of transfer bus routes and capacity planning under flow control scenarios was proposed, aiming to minimize total expected travel time and the operational costs of transfer vehicles, and maximize the alleviation of passenger congestion in the rail transit network. To enhance model-solving efficiency, the model was divided into two subproblems: route optimization and service optimization. The first subproblem was transformed into a traveling salesman problem, with the resulting alternative route paths serving as input for solving the second subproblem. Based on Chengdu’s urban rail transit network and passenger flow data during the morning peak period, the effectiveness of the proposed model under different levels of flow restrictions was verified, and the preferences for the number of stops and the selection of transfer station locations were discussed. Results indicate that routes with 2 to 3 stops generally perform well in terms of the objective function, and the selection of stopping stations is highly concentrated, with a strong preference for 3 to 4 specific routes. As flow control intensity increases, there is a clear tendency to choose routes with fewer stops and shorter travel distances to meet rapid transfer demands. The number of scheduled trips increases approximately linearly overall; however, when the flow restriction intensity exceeds 0.8, a single route can no longer meet the transfer demand, and the linear growth trend no longer holds. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

History, Present Situation and Prospect of Research on Leg Configuration of Humanoid Robot DING Hongyu, SHI Zhaoyao, ZHANG Pan, FU Chunjiang Journal of South China University of Technology(Natural Science Edition)    2025, 53 (10): 131-144.   DOI: 10.12141/j.issn.1000-565X.240228 Abstract （1307）   HTML （31）    PDF（pc） （8120KB）（132）       Knowledge map    Save The motion performance of humanoid robots has not yet fully reached the level of human beings, which is one of the factors hindering their large-scale industrial application. This limitation stems not only from constraints in control algorithms but also from mechanical structure design, particularly the leg configuration, which largely determines a robot’s dynamic balance, load capacity, and energy efficiency. The study examined the origins and evolution of leg configurations in humanoid robots, both domestically and internationally. Currently, the leg configurations of humanoid robots are primarily categorized into three types: serial, parallel, and hybrid serial-parallel. Their structural characteristics directly influence locomotion performance. The study compared the serial, the parallel and the series-parallel configurations and their performance characteristics. The serial configuration offers a large workspace and high flexibility, but its relatively lower stiffness—due to the extended joint transmission chain—compromises its load capacity. The parallel configuration provides high rigidity and fast dynamic response, yet its range of motion is limited. The hybrid serial-parallel design combines the strengths of both, achieving ba-lanced stiffness and flexibility, which has increasingly made it a key research focus in recent years. Finally, this paper also discussed technical difficulties and hot spots in the study of leg configuration and pointed out the development trend: the leg configuration is developing from single series configuration to parallel and series-parallel configuration, from rigid actuator to elastic actuator and quasi direct drive actuator, from torque control to hybrid force-position control. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Study of Dye Adsorption Properties of Cellulose-Based Nanocellulose Aerogels CHEN Gang, AO Jie, HE Yingying, WANG Chunyu, ZHANG Cheng Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 149-157.   DOI: 10.12141/j.issn.1000-565X.240587 Abstract （1076）   HTML （24）    PDF（pc） （3346KB）（670）       Knowledge map    Save The issue of water pollution is becoming increasingly severe, highlighting the urgent need to develop efficient and sustainable methods for pollutant removal. This study proposed the use of ambient pressure drying to prepare a nanocellulose-based aerogel with high adsorption capacity for both anionic and cationic pollutants. Polyethyleneimine (PEI) was first attached to a carboxymethylated cellulose nanofiber (CNF) framework through electrostatic interactions. Then, γ-aminopropyltriethoxysilane (APTES) and glutaraldehyde (GA) were used for chemical crosslinking to form a hydrogel. Finally, through solvent exchange and ambient pressure drying, a low-density (18.80 mg/cm³) and high-porosity (92.06%) CNF/PEI composite aerogel (CPA) was obtained. This aerogel demonstrated excellent structural stability in water. Owing to the coexistence of anionic carboxymethyl and cationic amino groups, the aerogel exhibited strong adsorption capacity of aerogel per gram for both cationic and anionic dyes in complex wastewater environments. The maximum adsorption capacities of aerogel per gram for methylene blaue (MB) and Congo red (CR) were 516 mg and 2 090 mg, respectively,with removal rates of over 98% for both anionic and cationic dyes. In addition, the aerogel exhibited good structural stability and fatigue resistance. After soaking in an alkaline solution for a week, it remained intact, and after 10 cycles of compression in the wet state, its elasticity recovery rate remained at 60%. Compared to similar adsorption materials, CPA shows significant advantages in terms of adsorption capacity, amphoteric adsorption ability, and reusability. The preparation method proposed in this study is time-efficient and highly effective, making it suitable for large-scale production, with promising potential for application in industrial wastewater treatment. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Machining Feature Recognition Method of B-Rep Model Based on Graph Neural Network HU Guanghua, DAI Zhigang, WANG Qinghui Journal of South China University of Technology(Natural Science Edition)    2025, 53 (5): 20-31.   DOI: 10.12141/j.issn.1000-565X.240329 Abstract （1039）   HTML （42）    PDF（pc） （2520KB）（406）       Knowledge map    Save Automatic feature recognition is one of the key technologies of intelligent manufacturing. Traditional rule-based recognition algorithms have poor scalability, and the methods based on deep convolutional networks are of low accuracy because they use discrete models as input and the recognition results are difficult to accurately map back to the original CAD model, causing inconvenience in application. In view of these shortcomings, a feature recognition method based on graph neural network, which can directly analyze B-Rep models, is proposed. The method extracts effective characteristic information and geometric information from the B-Rep structures to form a feature descriptor, and then establishes an adjacency graph with high-level semantic information based on the topological structure of the CAD model. By taking the adjacency graph as the input, an efficient graph neural network model is constructed. By introducing a differentiable generalized message aggregation function and a residual connection mechanism, the model possesses stronger information aggregation performance and multi-level feature capture capabilities. What is more, message normalization strategy is used to ensure the stability of the training process and to accelerate the convergence of the model. After the training, the network can directly classify and annotate all faces in the B-Rep model, thereby realizing feature recognition. Experimental results on the public dataset MFCAD++ demonstrate that the proposed method achieves an accuracy of 99.53% and an average intersection-over-union ratio of 99.15%, which outperforms other similar studies. Further evaluations using more complex testing cases and typical CAD cases from real engineering applications show that the proposed method is of better generalization ability and adaptability. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Design and Optimization of Fast Fourier Transform Algorithm Based on Ascend NPU LU Lu, WANG Yuanfei, LIANG Zhihong, SUO Siliang Journal of South China University of Technology(Natural Science Edition)    2025, 53 (11): 9-17.   DOI: 10.12141/j.issn.1000-565X.240524 Abstract （842）   HTML （30）    PDF（pc） （1040KB）（73）       Knowledge map    Save As a fundamental algorithm in scientific computing and signal processing, fast Fourier transform (FFT) has been widely applied to such fields as digital signal processing, image processing, deep learning. With the growth of data scale and the increasing demand for processing power, optimizing FFT algorithms on emerging hardware platforms has become particularly crucial. This paper conducts an in-depth analysis of the architectural cha-racteristics of Ascend NPU and their impacts on FFT algorithm optimization. Based on the matrix-computation-based Stockham FFT algorithm, a series of innovative optimization strategies are proposed: (1) A heuristic radix selection algorithm is designed to provide effective radix sequence combinations for different input sizes; (2) An efficient computation flow for single-iteration FFT without real-imaginary separation is developed, significantly reducing the global memory access overhead; (3) An on-chip cache-based data reading optimization strategy is proposed, greatly improving data access speed; (4) A data layout optimization method for multiple iterations is designed, effectively enhancing overall memory access efficiency. Experimental results on Ascend Atlas 800 platform equipped with Ascend 910 AI processor demonstrate that the proposed optimization strategies achieve an average speedup of 4.61 compared to non-optimized implementations. Independent performance analysis and validation of each optimization strategy demonstrate that the individual average speedup ratio ranges from 1.42 to 3.52. This research provides a technical references for implementing efficient FFT algorithms on emerging NPU architectures. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

CODS: An Audio-Text Aligned Dataset for Cantonese Opera Vocal Synthesis LI Yue, HUANG Yihan, PENG Zhengwei, XIE Jixuan, DU Yuye Journal of South China University of Technology(Natural Science Edition)    2025, 53 (9): 1-10.   DOI: 10.12141/j.issn.1000-565X.250134 Abstract （775）   HTML （55）    PDF（pc） （3373KB）（132）       Knowledge map    Save As one of the traditional Chinese arts, Chinese opera culture has unique musical expressiveness. Cantonese opera, as one of the main Chinese opera genres and an important carrier of Lingnan culture, has been indexed in the World Intangible Cultural Heritage List. In recent years, generative artificial intelligence technology has demonstrated its powerful capabilities in the field of content creation. For example, singing synthesis techno-logy can synthesize natural singing based on specified music scores. This provides a new idea for the digital protection and innovation of Cantonese opera. However, the collection and organization of opera data faces problems such as poor audio quality and complex dialect annotation, resulting in an extreme shortage of high-quality opera data sets. Based on this, this paper applied the singing synthesis technology in the field of pop music to the field of Cantonese opera vocal synthesis, and proposed the first Cantonese opera vocal synthesis dataset with phoneme-level annotation and audio-text alignment. Firstly, this paper constructed the CODS dataset through a systematic process. This dataset was derived from 29 original works by four famous performers with a total length of 3.81 hours, which provides important support for the research and digitization of Cantonese opera. Using this dataset, this paper conducted experiments with a deep learning-based method for Cantonese opera voice synthesis, realizing controllable generation in terms of lyrics, timbre, and melody. Finally, this paper established a comprehensive evaluation framework for Cantonese opera synthesis. Both objective and subjective evaluations reached a satisfactory level within the domain, further validating the usability of the proposed dataset. The CODS dataset constructed in this paper successfully filled the gap in artificial intelligence in the field of Cantonese opera vocal synthesis, and strongly promoted the inheritance and innovation of this traditional art. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Traffic Congestion Prediction Based on Dynamic Adaptive Gated Graph Convolutional Networks WANG Qingrong, GAO Huanyi, ZHU Changfeng, HE Runtian, MU Zhuangzhuang Journal of South China University of Technology(Natural Science Edition)    2025, 53 (9): 31-47.   DOI: 10.12141/j.issn.1000-565X.250003 Abstract （726）   HTML （29）    PDF（pc） （7533KB）（69）       Knowledge map    Save With the continual rise in the number of motor vehicles in urban areas, traffic congestion has become increasingly severe, adversely affecting environmental protection and urban operational efficiency. Consequently, it is of critical importance to accurately predict traffic congestion for traffic management and optimization. However, existing research still faces limitations in modeling the dynamic, time-varying characteristics of traffic flow and the complex interactions among road segments. To address these challenges, a gated spatiotemporal convolutional network model based on graph neural networks was proposed to more effectively capture and predict traffic congestion. Firstly, an improved K-means clustering algorithm was employed to divide the raw data into multiple congestion-state categories, which are then incorporated as auxiliary features to enhance feature representation. Next, a gated temporal convolutional network was introduced to capture the temporal properties and dynamic dependencies in traffic data, and a dynamic adaptive gated graph convolutional network was constructed to achieve feature fusion and dynamic weight allocation through a signal generation module and a dual-modulation mechanism, thereby facilitating effective extraction of spatiotemporal features. Finally, residual connections were incorporated to improve training stability, and skip connections were utilized to integrate multi-level and multi-scale features. Experimental results on real-world PeMS08 and PeMS04 datasets demonstrate that the proposed model achieves superior prediction accuracy compared with other baseline methods. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Design and Optimization of Small-Batch Matrix Multiplication Based on Matrix Core LU Lu, ZHAO Rong, LIANG Zhihong, SUO Siliang Journal of South China University of Technology(Natural Science Edition)    2025, 53 (9): 48-58.   DOI: 10.12141/j.issn.1000-565X.240498 Abstract （687）   HTML （15）    PDF（pc） （3143KB）（43）       Knowledge map    Save General Matrix Multiplication (GEMM) is one of the most important operations in linear algebra, serving as the backbone for numerous applications in machine learning, scientific computing, and signal processing. In particular, FP16 batch GEMM has become a core operation in deep learning frameworks due to its efficiency in training and inference. However, current implementations on AMD GPUs (e.g., CDNA/MI200 architectures with Matrix Cores) suffer from suboptimal memory access and low compute utilization, limiting performance in high-throughput scenarios. Therefore, this paper proposed a GPU optimization scheme for half-precision batch GEMM (HGEMM). In terms of blocking strategy, it allocates equal memory access and computational loads to threads based on input matrix sizes, while enabling each thread to compute multiple matrix multiplications to improve arithmetic unit utilization. For memory access optimization, it trades redundant data reads for uniform memory access patterns per thread to facilitate compiler optimization, ensuring overlapping of memory and computation time. For extremely small-batch HGEMM with matrix dimensions smaller than 16, the proposed method employs a 4 × 4 × 4 Matrix Core and its corresponding tiling scheme to enhance memory performance while reducing computational resource wastage, and provides the option of whether to use shared memory to achieve the highest performance. This paper compares the performance of this scheme with two operators of rocBLAS on the AMD GPU MI210 platform. The results show that the ave-rage performance of this scheme on AMD GPU MI210 is 4.14 times that of rocBLASHGEMMBatched and 4.96 times that of rocBLASGEMMExBatched. For extremely small-batch HGEMM, the average performance is 18.60 times that of rocBLASHGEMMBatched and 14.02 times that of rocBLASGEMMExBatched. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Analysis of Freeway Accident Factors Integrating Short-Term Traffic Flow WEN Huiying, HUANG Junda, HUANG Kunhuo, ZHAO Sheng, CHEN Zhe, HU Yuqing Journal of South China University of Technology(Natural Science Edition)    2025, 53 (10): 1-13.   DOI: 10.12141/j.issn.1000-565X.240535 Abstract （664）   HTML （11）    PDF（pc） （1691KB）（42）       Knowledge map    Save The severity of freeway traffic accidents is collectively influenced by multiple factors, among which short-term traffic flow characteristics immediately preceding the incident play a particularly critical role. To syste-matically analyze the impact of short-term traffic flow states on injury severity, this study constructed a random parameter logit model accounting for mean heterogeneity, utilizing historical traffic accident data, ETC gantry transaction records, and meteorological data from Guangdong Province’s South 2nd Ring Expressway, Jiguang Expressway, and Western Coastal Expressway (2021—2022). The model was developed to investigate heterogeneous characteristics of accident contributing factors. A total of 29 potential variables were identified across four domains: road cha-racteristics, environmental conditions, traffic flow features, and crash attributes. Three discrete model specifications were employed to model injury severity: a standard multinomial logit model, a random parameter logit model, and a random parameter logit model that accounts for mean heterogeneity. Comparative analysis of model goodness-of-fit using pseudo-R², akaike information criterion (AIC), and Bayesian information criterion (BIC) demonstrated that the random parameter logit model accounting for mean heterogeneity exhibits superior performance in goodness-of-fit. This specification more accurately captures the heterogeneous characteristics of accident contributing factors. Further analysis based on the average elasticity of variables reveals that, at the 99% confidence level, 22 parameters significantly affect injury severity. Specifically, features such as six-lane bidirectional roads and improved visibility significantly reduce injury severity, whereas longer road rescue handling time, higher average speed and proportion of large trucks, and greater speed differentials between large and small vehicles are associated with increased injury severity. The findings of this study offer valuable insights for improving freeway accident prevention and management strategies. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Active Tooth Surface Design and Performance Optimization of High Reduction Ratio Hypoid Gears JI Shuting, LI Jiahao, ZHANG Yueming Journal of South China University of Technology(Natural Science Edition)    2025, 53 (9): 106-116.   DOI: 10.12141/j.issn.1000-565X.240586 Abstract （659）   HTML （17）    PDF（pc） （4605KB）（55）       Knowledge map    Save To enhance the comprehensive transmission performance of hypoid gears with high reduction ratios, this paper proposed a design method for significantly inclined contact lines based on active tooth surface design techno-logy. Firstly, multiple tooth surface imprints with varying degrees of contact line inclination were preset, with specified values for the semi-major axis of the contact ellipse and the length of the contact trace. The pinion conjugate tooth surface was then modified with a parabolic shape to achieve a tooth surface that meets the preset parameters. Subsequently, by integrating Tooth Contact Analysis (TCA) and Load Tooth Contact Analysis (LTCA) techniques, the amplitude of transmission error (ATE), amplitude of loaded transmission error (ALTE), tooth surface load distribution, root bending stress amplitude, and tooth surface flash temperature amplitude were obtained for each tooth surface. The influence of variations in contact trace length on these performance parameters was then analyzed. Finally, a target modified tooth surface was selected, and its comprehensive performance was analyzed and compared with that of the original tooth surface. A case study demonstrates that for a hypoid gear pair with a gear ratio of 5∶75, under conditions of highly inclined contact trace on the tooth surface, a longer contact trace length leads to lower contact stress, as well as reduced root bending stress and flash temperature on the tooth surface. The target tooth surface exhibits weakened edge contact, a 12.0% reduction in maximum root bending stress, more uniform contact stress distribution, and a 6.3% decrease in peak flash temperature. As a result, the scuffing load-carrying capacity is enhanced. Overall, the modified target tooth surface exhibits superior contact performance, better load-carrying capacity, and significantly enhanced comprehensive transmission performance. Table and Figures | Reference | Related Articles | Metrics | Comments（0）

Select

Mix Design and Performance Optimization of Stone Powder-Rich Manufactured Sand Concrete ZHANG Tongsheng, LI Kai, TAN Kanghao, CHANG Zezhou, TAN Yanchen, TANG Liang, YANG Donglai, ZENG Siqing Journal of South China University of Technology(Natural Science Edition)    2025, 53 (8): 123-136.   DOI: 10.12141/j.issn.1000-565X.250042 Abstract （653）   HTML （1）    PDF（pc） （7142KB）（51）       Knowledge map    Save During the production of manufactured sand, a large amount of stone powder was sieved and buried, leading to resource waste and environmental pollution. To improve the utilization rate of manufactured sand stone powder, this study explores the high-value application of waste stone powder in concrete. By treating the stone powder in manufactured sand as a cementitious component to partially replace cement, the effects of granite stone powder on the microstructural evolution of hardened cement paste were investigated using XRD, TG, SEM, and other characterization methods, leading to the identification of the optimal cement replacement range. Furthermore, by adjusting the stone powder content in manufactured sand, coarse aggregate gradation, sand ratio, and water-to-binder ratio, the workability and mechanical properties of concrete were optimized. The study reveals the mechanism by which paste volume fraction influences concrete’s workability and mechanical performance, and successfully produced low-cost concrete with acceptable workability and mechanical strength using manufactured sand with a high stone powder content. The results show that cement paste with 10% stone powder retained a denser microstructure, as the amount of hydration products showed negligible reduction compared to that of pure cement paste after 7-day and 28-day curing. However, when the substitution of cement with stone powder exceeded 20%, the amount of hydration products decreased significantly by more than 20%, leading to a porous microstructure and lower compressive strength compared to that of pure cement paste. When manufactured sand (MS) with high stone powder content is used in concrete production, the dosage of superplasticizer needs to be increased slightly under the same slump requirement. Additionally, the optimal workability and mechanical properties of MS concrete were achieved when the volume fraction of paste lay in the range of 31~32%. Consequently, C30, C40, and C50 concretes meeting target property requirements were prepared using MS with 15.1%, 16.5%, and 18.7% stone powder content, respectively, resulting in cement consumption reductions of 54, 63, and 92 kg/m³, and thereby significant reductions in cost and carbon emissions. Table and Figures | Reference | Related Articles | Metrics | Comments（0）