In order to enhance the safety and efficiency of operations in the double-channel U-shaped apron area of large airports, an optimized operational procedure for the double-channel U-shaped area was studied. Firstly, the utilization and partitioning of taxiways in the double-channel U-shaped apron area were designed, and the positions of pushback holding points were optimized. Secondly, based on the partitioning of the double-channel U-shaped apron area and the optimized positions of pushback holding points, different operational procedures for aircraft were designed for various scenarios. Then, evaluation indicators were designed from the perspectives of safety and efficiency, and corresponding evaluation models were established. Finally, simulation experiments were conducted using Wuhan Tianhe Airport as the object. The results show that the proposed optimized operational procedure can reduce the total operation time by 13.3%, total waiting time by 31.4%, and waiting rate by 22.4%. The flight density was gradually increased until reaching the maximum theoretical capacity of the U-shaped apron area, and further verification was conducted. The results indicate that the proposed optimized operational procedure performs better across different indicators under varying flight volumes, verifying its effectiveness and providing theoretical references for current and future operational procedures of double-channel U-shaped apron areas.

The civil engineering industry faces with a vast array of unstructured textual information during its digital transformation. Large language models (LLMs) provide a new opportunity for the intelligent transformation of the industry because of its powerful natural language processing capability. A systematic literature review approach was employed, and based on the technical framework of LLMs and the current state of research in vertical domains, four major application scenarios for LLMs in civil engineering were suggested, along with corresponding technological approaches, challenges faced, and research trends. It is found that exploratory research and application of LLMs in civil engineering have been conducted, primarily focusing on content creation, intelligent Q & A, text summarization, and analytical reasoning, covering the entire lifecycle of civil engineering projects and featuring interdisciplinary and multimodal integration. However, the utilization of LLMs struggles with low specificity of knowledge, poor timeliness of information, and inferior data quality and interactivity. Based on this, a series of future research opportunities were proposed to enhance the breadth and depth of LLMs application in the field of civil engineering by using parametric efficient fine-tuning technology to inject expertise in model optimization. Combined with knowledge graph, LLMs can improve the accuracy, interpretability and timeliness of answers. Multi-modal data types were integrated to expand the application scenarios of LLMs in civil engineering. Appropriate model evaluation methods were developed to quantify the value and performance of LLMs applications in civil engineering. In terms of application scenarios, combined with the characteristics of LLMs and civil engineering fields, the application of LLMs in complex tasks such as document generation, question and answer system, information extraction and compliance review can be expanded, and the interaction efficiency between practitioners and data can be improved. The purpose of the study is to provide reference for the academic and business circles to further apply LLMs in the field of civil engineering.

The alignment monitoring of steel arch bridges constitutes an essential component of bridge health monitoring systems. Three-dimensional laser scanning technology was utilized, and the traditional density-based spatial clustering of applications with noise(DBSCAN) algorithm was improved by integrating the random sample consensus(RANSAC) algorithm to extract the alignment of steel arch bridge ribs. Three-dimensional laser point cloud data, characterized by its comprehensiveness and detailed representation, is capable of fully presenting the structural shape and deformation information of the bridge. The RANSAC-integrated improved DBSCAN algorithm, constrained by the structural features of the steel arch bridge, effectively achieves the removal of discrete points as well as point clouds from the bridge deck, cross bracing, lateral connections, and web members. Point clouds extracted using the RANSAC-integrated improved DBSCAN algorithm are fitted to identify key points, and a comparison is made with results obtained manually. The extraction errors for the key points of the arch ribs are all within the millimeter range, with the maximum error being 9.2 mm and the minimum error being 0.1 mm. This extraction method is demonstrated to more accurately and effectively accomplish the alignment extraction of steel arch bridges, achieving millimeter-level precision in alignment extraction. It significantly reduces labor and time costs, provides better robustness for the complex structures of steel arch bridges, and adapts well to practical production demands.

To mitigate the impact of highway accidents on traffic capacity and driving safety, a coordinated control strategy was proposed involving both service areas and toll stations. Firstly, the proposed coordinated control strategy was described in detail. Secondly, to simulate traffic flow more accurately under highway accident scenarios, the cellular automata model was enhanced by introducing different random deceleration probabilities, acceleration/deceleration rates, and lane-changing conditions for different vehicle types. Finally, the effectiveness of the proposed control strategy was validated through simulation. The results indicate that, compared to scenarios without control measures, implementing service area control can reduce average vehicle delay, fuel consumption, and cumulative carbon emissions by 62.90%, 69.50%, and 69.50% respectively. Moreover, using the coordinated control strategy of service areas and toll stations can further reduce these metrics by 55.76%, 59.58%, and 59.58% respectively. Precise control measures can significantly reduce the impact of accidents.

The offshore heavy oil thermal recovery platform has the characteristics of small space, high steam injection temperature, and high steam injection pressure, with temperatures up to 300 ℃. Once high-temperature and high-pressure steam leaks, it will cause serious consequences and pose a huge threat to equipment and inspection personnel. An effective steam leakage monitoring method was urgently needed. In order to solve these problems, the influence of thermodynamics, fluid mechanics and other factors were considered comprehensively to study the mechanism of steam leakage monitoring in offshore heavy oil thermal recovery. A virtual sensing monitoring method based on mechanism and inference was proposed, and for the first time, the indirect measurement method of steam leakage was applied to steam leakage monitoring in offshore heavy oil thermal recovery. A steam leakage monitoring model was built, and a hybrid sensing technology suitable for steam leakage monitoring in offshore heavy oil thermal recovery was formed for real-time online monitoring of steam leakage. The results show that this method can achieve leak discrimination and leak estimation based on operational data, and directly characterize the failure state of steam leaks online. The minimum detectable leak rate can reach 0.5%, and the accuracy of leak discrimination is above 96.49%. Compared with traditional methods, the minimum detectable leakage rate has increased by 90%, and the leakage discrimination rate has increased by at least 1.6%. This method solves the problems of limited installation of physical sensors on site, difficulty in obtaining effective monitoring data, and limited accuracy due to personnel experience, making up for the shortcomings of on-site monitoring methods for thermal recovery platforms and providing safety guarantees for offshore heavy oil development.

Although the multi-task convolutional neural networks (MTCNN) face detection algorithm has achieved good results in some face recognition tasks, the accuracy of face detection needs to be improved in the face of some complex small-scale and multi-person face detection tasks. An improved MTCNN algorithm was proposed. Firstly, the intersection over union (IoU) threshold parameter was fine-tuned when creating the data set to classify face samples more accurately. Secondly, replacing the max pooling layer of the network with convolutional layers can improve network performance. Finally, the squeeze-excitation(SE) attention mechanism was introduced into the O-Net network to improve the feature expression ability of the network. The test results show that compared with the original MTCNN algorithm, the detection accuracy of the P-Net network and R-Net network of the improved algorithm has increased by 1%, and the detection accuracy of the O-Net network has increased by 0.5%. Moreover, the improved algorithm performs better in the actual face detection task.

Image segmentation is a fundamental problem in medical image analysis, the typical deep learning based UNet architecture (UNet) and its variants are widely used in retinal vessel segmentation. However, the UNet network extracts feature information from images through local convolution modules, which makes the global information of the images difficult to be correlated and the long-distance dependencies between pixels difficult to be effectively captured. Considering the problems with the UNet network model and the characteristics of retinal vascular images, an attention module was added to the skip connections of UNet to capture long-distance dependencies between blood vessels. In addition, to enhance the segmentation ability of the network, the group normalization(GN) was used instead of the original batch normalization (BN) of the UNet network model, and the corresponding groups were selected for different channels. To update parameters and optimize the network, the final cross entropy loss function was designed using the side output layer and the final output layer. Experiments are implemented on the DRIVE dataset and CHASEDB1 dataset, and the experimental results show that the proposed model has better image segmentation performance.

Seepage analysis is the key research content of dam safety and stability, and it is of great significance for dam disaster risk control by constructing a high-precision prediction model of seepage quantity for earth-rock dam. In order to further improve the seepage prediction capability of earth-rock dam, a prediction model combining long short-term memory neural(LSTM) networks, convolutional neural(CNN) networks, and attention mechanism (Attention) was proposed. Firstly, CNN was used to mine the deep features of the data, then the time series features of the seepage flow monitoring data was extracted through LSTM, and finally the attention mechanism to the pooling layer and the fully connected layer was added to determine the importance of different time features and assign weights. Through the application analysis of engineering examples, compared with CNN, LSTM and CNN-LSTM models, the CNN-LSTM-Attention model has better prediction effect, and its coefficient of determination R² is as high as more than 0.98, and it can capture the spatial characteristics and temporal dependence of seepage data at the same time, which shows strong reliability and stability in the prediction of seepage flow of earth-rock dam.

The generalized finite difference method for seismic wavefields numerical simulation is capable of adapting to undulating stratigraphic interfaces, eliminating the staircase scattering effect caused by such interfaces, and enhancing the accuracy of forward modeling. However, when the second-order generalized finite difference method is used to solve the wave equation, low temporal accuracy can lead to temporal dispersion at larger time intervals, affecting the precision of forward simulation. A fourth-order generalized finite-difference forward modeling algorithm for the scalar wave equation was studied, along with its stability conditions and dispersion characteristics. By transferring the fourth-order time derivative to the spatial derivative term, fourth-order accuracy in time was achieved, effectively suppressing temporal dispersion. Compared to the second-order generalized finite-difference method, the fourth-order approach allows for larger time intervals, thereby reducing computational costs to some extent. Experimental results demonstrate that the proposed algorithm effectively mitigates both staircase scattering and temporal dispersion, yielding higher computational accuracy. When applied to reverse time migration, it produces high-quality imaging profiles.

In order to achieve accurate segmentation of surgical instruments, a dual-encoding network surgical instrument segmentation method was proposed based on improved Swin Transformer. By taking advantage of different coding advantages of Swin Transformer and convolutional neural network(CNN), the global semantic information and local details of image features can be effectively captured to improve the expression ability of the model. To compensate for the loss of feature details during the downsampling process as much as possible, the multi-resolution feature pyramid pooling(MFPP) block was constructed to obtain more scale context information by combining different dimensional features and enhance the expression of local detail information. Finally, a coordinate attention block was added in the skip connection to fuse target position information with feature information for precise perception of the surgical instrument targets. The experimental results show that the proposed method achieves more accurate segmentation results in both binary and parts segmentation of surgical instruments, further verifying the effectiveness and accuracy of the proposed method.