The strong interference in the visible channel and the insufficient error correction performance of the coding techniques for OCC(Optical Camera Communication) systems cause the reliability of OCC systems to be reduced. A QC-LDPC(Quasi-Cyclic Low Density Parity Check Code) compilation code algorithm for OCC system is proposed. According to the structural characteristics of OCC system and channel link characteristics, QC-LDPC code with structured design and low-complexity decoding characteristics is identified as the error correction code type, and fast coding algorithm is used to reduce the computational complexity of the coding process.Combined with the normalized least-sum algorithm and the layered decoding algorithm, an improved L-MS(Layer-Min Sum) decoding algorithm is proposed. The experimental results show that the BER(Bit Error Rate)of the OCC is 3. 16x10^-6. The communication rate is 2 850. 4 bit / s under the communication scenario of indoor environment. The BER of the OCC system is 9.07x10^-6 and the communication rate is 2 797. 9 bit / s under the communication scenario of outdoor environment. The proposed method effectively maintains a low bit error rate while ensuring a satisfactory data transmission rate.

To address the performance distortion caused by a priori information loss in signal transmission processes of communication systems, an information fidelity-based performance bounds evaluation method is proposed for information recovery. In practical communication environments, due to channel noise and system non-ideality, the receiver often can not obtain a complete priori information from the transmitter, which limits the applicability of traditional Fisher-information-based Cramer-Rao bounds under such conditions. By introducing the concept of information fidelity and its moment generating function, an inequality relating the noise-to-signal ratio, the Fisher information, and the fidelity information is established, and a new lower-bound estimation model for NSR(Noise-to-Signal Ratio) is derived, providing more robust theoretical bounds for information recovery performance in communication systems. Theoretical analysis and two numerical examples ( including a binary communication system and a multimode signal transmission channel) demonstrate that, compared with classical Fisher-information-based bounds, the information-fidelity-based bounds, while more relaxed, exhibit stronger robustness and practicality in realistic scenarios with incomplete a priori information, and small sample sizes,offering a new theoretical tool for communication system performance evaluation.

In the detection of communication network vulnerabilities, the accuracy is insufficient due to the lack of cluster structure information. Therefore, a security detection method for communication network node attack vulnerabilities based on ant colony algorithm is studied. Firstly, sliding time window segmentation and spatiotemporal cross-correlation analysis are used to extract low dimensional interpretable node features and enhance anomaly pattern discrimination. Then, convolutional neural networks are used to perform convolution, pooling, and flattening operations on the reduced dimensional features, combined with a Softmax classifier to determine the probability of abnormal nodes and narrow down the detection range. Finally, based on the passive clustering algorithm, the cluster structure is divided using ant colony algorithm, combined with the relationships

and communication modes of nodes within the cluster, to track the location of vulnerabilities through pheromone concentration. The experimental results show that this method can accurately locate communication network attack vulnerabilities. Therefore, this method can comprehensively and accurately evaluate the network security situation, providing an effective solution for communication network vulnerability detection.

In mobile network communication, image data usually needs to be transmitted and shared with multimedia data such as audio and video. Image data often contains personal information, such as photos,videos, etc. In order to ensure the security of network communication data effectively, a double chaotic data encryption algorithm is proposed based on the image data modified in large-scale mobile network communication.A new filling curve is used for global scrambling of image data in large-scale mobile network communication. An improved Joseph ergodic method is used to expand the bit-level scrambling of adjacent pixels, and bidirectional ciphertext feedback is introduced to realize the initial encryption of image data. Double chaos and Arnold transform are effectively combined. The initial state value is formed by using the key, which is used to control the double chaos system to form a series of chaos matrix, the position transformation and value change of image pixels are expanded as an aid, and each component of the image is expanded by Arnold transform. The set scramble-diffusion encryption algorithm is used to process the horizontal connected image, and finally the XOR ( exclusive OR) operation is performed on the image pixels to realize the security encryption of double chaos data in large-scale mobile network communication. Experimental results show that the proposed algorithm can obtain satisfactory encryption effect and has good plaintext sensitivity.

To solve the problems of insufficient positioning accuracy and high energy consumption of existing mobile anchor node positioning methods in complex network environments, an intelligent dual mobile anchor node positioning algorithm for regional wireless networks is proposed. By calculating network connectivity, it ensures effective information exchange between nodes, providing a fast communication background for positioning operations. In this context, introducing virtual force to push the moving anchor node towards the target position can more accurately estimate the trajectory and position of the moving anchor node. In order to reduce the energy consumption of mobile anchor nodes during the process of virtual force attraction, the DASCAN(Dual mobile Anchor Node Scanning and Coordination Algorithm) algorithm is used for path planning of dual mobile anchor nodes, and the weighted centroid algorithm is used to estimate the approximate position of the target unknown

node. The whale optimization algorithm is further used to improve the positioning accuracy. The experimental results show that the path tracking results of the proposed method are in good agreement with the actual path, and can accurately receive all the position information of unknown nodes, with high positioning accuracy. The energy consumption during the positioning process remains within 0. 25, further proving that the proposed method can effectively improve the accuracy and feasibility of node positioning. This method can ensure the coverage of network node positioning and has high positioning accuracy and efficiency.

To address the degradation of performance in speed control and insufficient robustness of permanent magnet synchronous motors under high dynamic operation and external load disturbances, a continuous-time model predictive control method that integrates Laguerre function-based parameterization with a nonlinear disturbance observer compensation mechanism is proposed. The method employs finite-dimensional parameterization of the control increment using Laguerre orthogonal functions to significantly reduce the optimization dimension and improve computational efficiency. An adaptive smoothing factor is designed to achieve a dynamic balance between control response speed and input smoothness. To enhance the engineering

applicability of the control system, a continuously differentiable soft constraint mechanism is introduced to avoid numerical instability caused by hard constraints. Simulation results verify that the proposed method exhibits superior speed tracking accuracy and disturbance rejection performance compared to conventional model predictive control under the same conditions.

To meet the demand for talent cultivation in the context of the rapid development of drive-by-wire chassis, and to tackle issues such as high implementation difficulty, high costs, and safety risks in experimental teaching of vehicle handling and stability, an experimental teaching platform for handling and stability has been designed and developed. The platform is developed based on an intelligent model vehicle chassis, integrating sensors including cameras, gyroscopes, and wheel speed sensors, which enables experimental tests under extreme vehicle operating conditions. The experiment management software is developed on the Matlab platform, facilitating remote configuration and control algorithm loading through wireless communication, featuring high flexibility and scalability. A number of experiments related to handling stability and chassis control have been conducted based on this platform. The experimental results demonstrate that the intelligent model vehicle differs from real vehicles in terms of response frequency and phase delay characteristics, exhibiting higher dynamic response capability and lower phase delay. Its overall trend is consistent with that of real vehicles, which can effectively verify the relevant content in theoretical instructional.

In response to the impact of the randomness of wind and solar power generation and the high penetration rate of DG ( Distributed Generation ) on the grid, the IGAN ( Improved Generative Adversarial Network) and the IChOA( Improved Chimpanzee Optimization Algorithm) are integrated to construct a source-storage two-layer optimal configuration model for active distribution networks. Firstly, IGAN is used to generate typical wind and solar power generation scenarios to represent their uncertainty. Then, a source-storage two-layer optimization architecture containing voltage quality indicators is constructed. The upper layer uses IChOA to optimize the capacity and location configuration schemes of ESS(Energy Storage Systems) and DG, and the lower layer uses SOCP ( Second-Order Cone Programming) to solve the optimal daily dispatch strategy of energy

storage. Finally, the IEEE 33-node system verification shows that this collaborative strategy can significantly reduce the annualized cost and network loss of the system, improve the voltage level, and effectively avoid voltage limit violations and DG power fluctuations.

CFOA(Catch Fish Optimization Algorithm) usually includes two update stages, the exploration stage and the exploitation stage. However, this algorithm is still prone to getting stuck in local optima and has a relatively low convergence rate. To address these issues, the ICFOA ( Improved Catch Fish Optimization Algorithm) is proposed based on personalized fishing strategies. Firstly, an adaptive Gaussian perturbation is introduced in the exploration stage to enhance the global search ability and efficiency while avoiding local optima. Secondly, based on personalized fishing strategies, the positions of fishermen are updated by randomly selecting either the “ bare-handed fishing “factor or the “ using a fishing net ”factor to accelerate the convergence. The CEC2020 test suite is used to conduct comparative experiments to evaluate the performance differences between ICFOA and other excellent meta-heuristic algorithms, and the Wilcoxon rank sum test is used to verify the validity of the statistical results. Finally, the maximum power point tracking in photovoltaic power plants, ICFOA effectively reduces the power tracking deviation, demonstrating its effectiveness in solving practical problems. Experimental results indicate that ICFOA possesses greater competitiveness compared to the original CFOA.

Under the coupling effect of equipment noise and external environmental interference, the real partial discharge signal is masked or distorted, which increases the difficulty of feature extraction. Therefore, a partial discharge signal feature extraction algorithm based on feature pattern decomposition is proposed. Using wavelet transform to analyze the main edges of the signal, calculate the correlation coefficients between each scale and adjacent scales, and eliminate the noise of partial discharge signals. Using the variational mode decomposition method to decompose the discharge signal, calculating the multi-scale entropy of the intrinsic mode components,and further filtering out interference; Combining the kernel principal component analysis method to reduce the feature parameters of the input signal, calculate the energy, modulus, and absolute mean of the frequency band projection sequence to complete feature extraction. Experimental results have shown that the proposed algorithm effectively extracts partial discharge features, provides detailed fault information, and ensures stable operation.

A novel method integrating multi-feature weighted clustering and intelligent optimized neural network is proposed to address the issues of subjective feature weighting and insufficient nonlinear mapping accuracy in low-voltage distribution network line loss analysis. First, an improved k-means clustering algorithm ( MFW-IKCA: Multi-Feature Weighted Improved k-Means Clustering Algorithm) is established by introducing dynamic weight matrices and reference weights, combining entropy weighting and mutual information methods for adaptive feature weighting, and optimizing the objective function via the ADMM ( Alternating Direction Method of Multipliers). Subsequently, an IGA(Improved Genetic Algorithm) is employed to optimize the parameters of the LMBP-NNM ( Levenberg-Marquardt Backpropagation Neural Network ), enhancing convergence speed and prediction accuracy through simulated binary crossover, polynomial mutation, and adaptive damping factors.Finally, decision tree classifiers are utilized to extract clustering rules, while kernel density estimation enabled anomaly detection. Experimental results on 710 transformer district datasets have demonstrated a 14. 5% improvement in clustering purity, a 29. 3% reduction in MSE ( Msemean Squared Error ), and a 0. 892 coefficient of determination (R 2 ). This study provides a new perspective for refined line loss analysis and offers significant engineering value for improving scientific decision-making in loss reduction of distribution network.

The extensive integration of distributed generation complicates power system operation, challenging the computational efficiency and scalability of existing analytical tools. To address this, efficient reduced-order modeling techniques must be developed for representing multi-inverter networks. A promising approach involves coherency-based aggregation to create simplified dynamic models for large-scale distributed generation systems with heterogeneous inverters. Simple and efficient, conventional methods retain network nonlinearities, limiting accuracy in dynamic analyses under disturbances. And prior coherency analyses often overlook virtual EMFs (Electromotive Forces) and parameter coupling effects among inverters. To resolve these limitations, a parameter weight-based coherency equivalence method for VSGs ( Virtual Synchronous Generators) is proposed. First,through small-signal analysis, the voltage-current dual-loop control of inverters is shown to exert minimal influence on virtual rotor motion, enabling the derivation of a virtual synchronous generator excitation mode grounded in physical principles. This achieves precise coherency identification under small perturbations while clarifying the parametric influence mechanisms. Subsequently, virtual rotor motion equations are utilized to quantitatively analyze parameter impacts (e. g. , moment of inertia J, damping coefficient D, and line reactance X), with results assigned as parameter weights to establish a weighted coherency algorithm. This algorithm enables rapid and accurate identification of coherent inverter clusters, followed by an aggregation algorithm that integrates active power conservation and reactive power dynamic equivalence. The correctness and effectiveness of the proposed coherency and aggregation methods are validated via PLECS simulations.

In SSPML( Semi-Supervised Partial Multi-Label Learning), complex correlations exist between the feature and label, and within them. Models leveraging label correlations are susceptible to interference from noisy labels in candidate label sets. T-SSPML ( Transformer Encoder-Based Semi-Supervised Partial Multi-Label Learning), a novel semi-supervised partial multi-label image classification algorithm is proposed based on Transformer. A backbone network is constructed using Transformer encoders, taking both image features and label features as joint inputs, and a new loss function is defined. The multi-head self-attention mechanism is employed to capture holistic structural information and label interdependencies. A label masking mechanism is

introduced to enhance the model’s capability in learning label correlation information. Extensive experiments on multiple datasets demonstrate that the T-SSPML algorithm achieves superior classification performance compared to other representative methods in most cases.

To address the issues of multi-source heterogeneity in production logging data, strong multi-solution interpretability, the inefficiency and expert-dependency of traditional methods, a question-answering system enhanced reasoning method based on CoT(Chain of Thought) is proposed. This method is built on the LLaMA2 large language model and the parameter-efficient fine-tuning LoRA ( Low-Rank Adaptation ) technique is introduced, which inserts low-rank matrices into the self-attention and feed-forward network layers to achieve a balance between model performance maintenance and computational resource optimization. A DCoT(Divergent Chain of Thought)reasoning framework is constructed, enabling the model to generate multiple reasoning paths and enhancing the system’s multi-step interpretation ability for complex production logging problems.

Experimental results show that this method improves F1 score, precision, and recall by 4. 6% , 7. 0% , and 2. 1% , respectively, compared to the baseline model, demonstrating strong reasoning performance and application potential. The research proves that the LoRA+DCoT method has good practicality and scalability in improving the intelligent interpretation effect of logging data.

To tackle the challenge of synergistically optimizing space utilization and center-of-gravity stability in loading highly heterogeneous goods, a novel 3D boxing algorithm is presented combining adaptive clustering and improved swarm intelligence optimization. It overcomes the limitations of traditional K-means and ant colony algorithm combinations by innovatively building a “ volume-weight" dynamic clustering mechanism, which optimizes cluster numbers via silhouette coefficients to enable collaborative goods grouping and initial loading sequence generation, resolving stacking instability from weight differences in similar goods. Further, an improved ant colony strategy with center-of-gravity constraints is designed, converting six-direction attitude selection into a pheromone-guided multi-dimensional optimization. By introducing a center-of-gravity compensation factor and spatial fitness heuristic function, collaborative optimization of loading attitudes and placement sequences is

achieved. Experimental results show the algorithm effectively breaks the “ seesaw effect "between space utilization and center-of-gravity stability, offering a new technical path for intelligent loading of highly heterogeneous goods in logistics and a valuable collaborative optimization framework for complex combinatorial problems.

To address the challenges associated with complex nonlinear characteristics and limited detection accuracy in long-distance pipeline leak detection, the application value and development potential of deep learning technologies are systematically investigated. Firstly, the current research status of deep learning in long-distance pipeline leakage detection both domestically and internationally are reviewed. It analyzed key technologies in deep learning-based pipeline leak detection and summarized existing challenges. Finally, this study discussed the advantages and limitations of deep learning in the field and outlined future research directions for long-distance pipeline leakage detection technology.

Medical terminology is characterized by strong heterogeneity and scarce labeled data, leading to weak semantic feature representation and low retrieval accuracy. To address these issues, a semi-supervised learning-based fuzzy retrieval algorithm for medical term information keywords is proposed. Word embeddings are utilized to standardize and map original medical term information. Based on core metadata information nodes and user query information nodes, the confidence distance between retrieval samples is determined, and a vector space model is employed to construct a metadata feature space for medical term information. Within this metadata feature space, data are merged according to their temporal attributes, thereby synchronizing data indexes.Semantic features are enhanced through multi-level index clustering and contextual encoding. The MixMatch semi-supervised learning algorithm, based on consistency regularization and entropy minimization principles, is introduced. A small amount of labeled data is used for data augmentation of unlabeled data, generating low-entropy pseudo-labels for unlabeled data to enhance semantic feature representation. The similarity between retrieval terms and candidate information is then calculated, enabling efficient matching across categories or synonymous terms. Experimental results demonstrate that applying the proposed method for medical term information keyword retrieval yields a retrieval result relevance exceeding 95% , effectively improving information retrieval accuracy.

In order to solve the problem of feature point extraction in quantum images, the implementation scheme of Harris corner detection algorithm on quantum computers is studied. Firstly, to address the efficient encoding and computation of rational numbers in Harris corner detection quantum circuits, a quantum image representation model based on complement floating-point representation is proposed. In this model, pixel values are represented in the form of complement-based floating-point numbers, and the position information of pixels is represented by unsigned numbers. Each data structure is based on the basic state of quantum bit sequences for information representation. Secondly, based on the relevant theories of the classical Harris corner detection algorithm, a specific quantum circuit for Harris corner detection is designed, and its complexity analysis is conducted to verify its exponential acceleration compared to classical schemes. Finally, through simulation experiments on classical computers, it is demonstrated that the overlap rate between the classical Harris corner detection algorithm and the quantum Harris corner detection scheme exceeds 85% , which to some extent verifies the feasibility and effectiveness of the proposed scheme.

To address the limitations of traditional heuristic algorithms in solving the ship ocean weather routing problem, namely slow convergence, susceptibility to local optima, and high dependence on parameter settings,an improved route planning framework is constructed based on intelligent optimization algorithms. By integrating the advantages of genetic algorithm, ant colony algorithm and simulated annealing algorithm, a reversal mutation and pheromone fusion mutation strategy is proposed, and a horizontal dynamic adjustment mechanism is designed based on fitness and a vertical dynamic adjustment mechanism based on simulated annealing for genetic probabilities. These strategies enhance the global optimization capability and convergence efficiency of the algorithm. Comparative simulation experiments show that, compared to traditional algorithms, the proposed

fusion algorithm shortens the path length by 11. 7% ~ 36. 7% and reduces the navigation time by 12. 2%~37. 4% , verifying its optimization performance in complex meteorological environments.

A planning method based on artificial bee colony optimization algorithm ABC(Artificial Bee Colony) is proposed to optimize the cross regional allocation path of emergency disinfection materials after natural disasters. Using the entropy weight method to comprehensively evaluate multidimensional indicators such as disaster area, population density, and medical resource gaps in designated hospitals, the urgency of demand for each disaster site and related hospital is quantified. Based on the quantification results, the emergency area is divided into four levels: red, orange, yellow, and blue, in order to achieve priority scheduling in different zones. A fitness function for the artificial bee colony optimization algorithm is designed, which includes total delivery time and unmet urgent needs. Artificial bee colony optimization algorithm is used to solve and find the optimal planning scheme for the allocation path. The experimental results show that under two typical operating conditions, the maximum fitness functions reach 0. 97 and 0. 95, respectively, which is 3. 19% to 13. 10% higher than the artificial fish swarm algorithm and genetic algorithm. It significantly improves the timeliness and emergency priority of material distribution for key institutions such as hospitals.

In the process of material supply and demand allocation in the sterilization supply room, conventional particle swarm optimization algorithms are commonly used to solve the allocation scheme, which is prone to falling into local optima, resulting in poor resource utilization efficiency. Therefore, a supply and demand allocation algorithm for material supply chain in the sterilization supply room based on an improved MOPSO (Multi Objective Particle Swarm Optimization) algorithm is proposed. Relying on supply chain algorithms, a rapid scheduling system for materials in the sterilization supply room is constructed. Dynamic medical monitoring data is obtained and is inputted into a spatiotemporal attention network model for learning, to derive the demand for sterilization materials. With meeting material demand as the premise, a supply and demand allocation objective function for the supply chain is set, focusing on minimizing cost and maximizing resource utilization. By

utilizing an improved multi-objective particle swarm optimization algorithm combined with a linear differential decrement strategy, the objective function is comprehensively solved to generate the optimal material allocation scheme. Experimental results show that after applying this algorithm, the utilization rate of allocated materials stabilizes at over 85% , achieving full utilization of materials in the sterilization supply room.

Due to the fact that most of the violations are instantaneous actions, they are easily obscured by redundant information from a large number of compliant actions, resulting in poor feature discrimination at different distances and low recognition accuracy. Therefore, a mixed attention mechanism is proposed to conduct research on the recognition algorithm of illegal movements in competitive sports. Background subtraction is used to locate the athlete's area, and Gaussian filtering and downsampling techniques are combined to construct a scale space to adapt to motion features at different distances. Hessian determinant is used to extract spatiotemporal and multi-scale features, and high-level features are fused through bidirectional transmission to improve feature discrimination. Spatial attention focuses on key parts of the limbs, temporal attention locks in keyframes of illegal actions, channel attention enhances effective feature channels, and the three work together to output enhanced

features. Finally, the end-to-end training optimization model is used, combined with sequence distance judgment to distinguish similar actions and achieve the recognition of illegal actions. The experimental results show that the algorithm can effectively separate similar action features, with a recognition accuracy of over 99% for illegal actions, an average recognition accuracy of 99. 81% , which are better than the comparative algorithms and have greater application value.

Due to the high heterogeneity of hospital patients and the spatiotemporal fluctuations of external intervention measures, the transmission of nosocomial infections presents non-linear and stochastic characteristics, which increases the difficulty of risk prediction. Therefore, a risk prediction algorithm for nosocomial infection transmission based on Bayesian networks and Monte Carlo simulations is proposed. The factors influencing the risk of nosocomial infection transmission are screened from four dimensions: patients,medical operations, environment and management. The categorical variables are converted into continuous variables to unify the variable dimensions, which is convenient for subsequent calculations. The objective weighting method is introduced to obtain the weight of a single factor, and the subjective weights are integrated to obtain the comprehensive weight of the influencing factors, so as to rank the importance of the influencing factors. An infection transmission risk prediction model is constructed by using Bayesian networks to calculate the joint probability distribution of high-weight influence factors and maximum likelihood estimation. The Monte Carlo simulation algorithm is utilized to randomly sample the uncertain variables in the model, generating simulation scenarios. Based on the statistical simulation results, the probability of infection transmission risk is calculated to determine the risk situation. The experimental results show that when the proposed method is applied to predict the risk of nosocomial infection transmission, the specificity of the prediction results is higher than 90% , false alarms are reduced, and the prediction accuracy is higher.

In the process of intelligent data retrieval, the limitation of a single semantic similarity may affect the judgment of data relevance, resulting in a low ranking sensitivity of the retrieval results. To alleviate this problem, an intelligent fuzzy retrieval algorithm is proposed for high-dimensional overlapping data under hybrid semantic similarity extraction. The information granularity space of high-dimensional overlapping data is established. Through density clustering and fuzzy measurement, the regions where the overlapping data is located in the space are identified, thereby achieving the dimension reduction of high-dimensional data without changing the data structure. Three semantic similarity calculation methods, namely the mixed mean function, approximate inearization statistics, and rule method, are used to comprehensively analyze the correlation of dimensionality reduction data. Under the traversal of the nearest neighbor algorithm, the data information with higher semantic

similarity is extracted in sequence to generate the data retrieval list. The results of the case study show that the average NDCG(Normalized Discounted Cumulative Gain) index value exhibited by this algorithm is 0. 867. The retrieval results have a high ranking sensitivity and higher retrieval quality, and have a good practical application prospect.

In the design of personalized recommendation algorithms for vocational education resources, chaotic mapping is commonly used to construct recommendation learning models. However, such models are susceptible to the curse of dimensionality, leading to suboptimal NDCG ( Normalized Discounted Cumulative Gain). To address this, an optimized design of a personalized recommendation algorithm is proposed for vocational education resources based on dynamic knowledge graph and reinforcement learning. Specifically, a dynamic knowledge graph representation is constructed from multi-source heterogeneous vocational education resource data, and a temporal-aware embedding representation vector is introduced to derive dynamic embedding functions. A graph attention network is employed to aggregate neighborhood information of entities, updating node representations to deeply integrate structural information of the dynamic knowledge graph. An Actor-Critic framework is adopted to maximize cumulative expected rewards, and entropy regularization is introduced to achieve a composite reward

function that balances recommendation objectives. By combining a cost-sensitive search algorithm with the reinforcement learning strategy, a heuristic evaluation function is generated. A resampling mechanism is applied to locate effective resources within the dynamic knowledge graph for path insertion, enabling adaptive recommendation path planning and achieving personalized recommendation of vocational education resources.Experimental results demonstrate that the proposed algorithm consistently achieves NDCG values above 0. 8 across different types of resources. It effectively accounts for item relevance and ranking positions, placing items of higher user interest in more prominent positions. The recommendation quality remains high without being affected by browsing and filtering delays or demand mismatches, exhibiting significant recommendation consistency.

During the high-speed dribbling action in basketball, when a player breaks through, displacement blurring occurs, resulting in a motion blurring effect. It is difficult to obtain the dribbling trajectory map of a basketball player, leading to a deviation between the dribbling trajectory and the actual trajectory. Therefore, a real-time recognition method for basketball players' dribbling trajectories based on the graph optimization DWA(Dynamic Window Approach) algorithm is proposed. Obtaining the point cloud data of the target athlete, and for the basketball court and obstacle events ( other athletes), establishing a raster map, a kinematic model is constructed and a dynamic window is established to simulate multiple dribbling prediction trajectories of the target athlete. The trajectory safety is determined through the dynamic window and the best dribbling prediction trajectory from multiple dimensions is evaluated and determined. Based on the predicted trajectory, the turning points and path points during the dribbling process are obtained. Combined with the A* algorithm, the dynamic obstacle avoidance route is acquired, and the trajectory graph model is derived through the pruning algorithm.The final evaluation function is constructed by designing the heuristic function. According to the evaluation results, the optimal trajectory points in the trajectory graph model are determined, and the obtained trajectories are smoothed using Bezier curves. The priority level of the trajectory points is determined by traversing the graph model, and the trajectories with better smoothness are obtained by comprehensively processing the coordinates and visualizing the output results. The experimental results show that the deviation between the dribbling trajectories of the athletes identified by this method and the actual trajectories is less than 5cm, indicating that the dribbling trajectories of basketball players identified by this method have relatively high accuracy.

The traditional particle swarm optimization algorithm exhibits strong randomness during evolutionary process. In the stage of searching for the optimal solution, some particles in the population that deviate from the global optimum can interfere with the convergence direction of the evolutionary process, leading the algorithm to easily fall into a local optimal state. To effectively address this issue, an emergency resource dynamic scheduling method based on improved particle swarm optimization algorithm is proposed. The objective function aims to minimize the total cost of disaster relief scheduling and maximize the fulfillment of actual material demands, and corresponding constraint conditions are established to construct a dynamic scheduling model for emergency resources. Chaos motion theory is used to improve the traditional particle swarm algorithm, obtain the global optimal solution, and the improved particle swarm algorithm is used to solve and calculate the scheduling model to obtain the optimal scheduling plan, achieving precise scheduling of emergency resources in disaster stricken areas. The experimental results show that when using this method for dynamic scheduling of emergency resources, the quantity of material scheduling highly matches the actual demand, and the scheduling path length is relatively short, which verifies its efficiency and reliability in practical applications.

In the process of data migration, it is necessary to ensure the consistency and integrity of data, avoid data loss or damage, and reduce the amount of duplicate data. Therefore, a hash graph based batch lossless database information migration method is proposed. The similarity of data between database is obtained based on the similarity of database data attributes and the optimal matching algorithm for bipartite graphs, the detection and deletion of duplicate data is completed based on the similarity. The hash values of each data after deduplication is calculated using a locally sensitive hash function, and a hash graph based on the hash values of all data is established. Conditional deep convolution is utitized to generate adversarial networks for batch migration of data, and the hash values of the migrated data is compared with the corresponding values in the hash graph. The migration is re-expanding for data with different hash values to ensure lossless migration of database

information. The experimental results show that this method has a fast information transfer speed, low data duplication rate, and can ensure data consistency, indicating that it has achieved the research objectives and is feasible.

To effectively mine and utilize web data, a correlation data adaptive mining algorithm is proposed with the support of web crawling technology. Web crawling technology is used to crawl e-commerce web pages, breaking through data acquisition limitations, storing the required data in a structured format in a database, and providing an orderly data foundation for subsequent processing. Based on sparse data stored in the database, the Apriori algorithm is applied according to the minimum support degree. Through pruning strategy, this algorithm can accurately find frequent itemsets in sparse data while reducing data processing, successfully solving the problem of traditional methods being difficult to obtain frequent itemsets due to data sparsity. Using a tree structure as a framework, the frequent itemsets obtained by the Apriori algorithm are used as nodes.

By adaptively combining old and new nodes, a mining tree is constructed until there are no nodes to combine.The generated mining tree comprehensively presents data association relationships, thereby obtaining sufficient and accurate e-commerce webpage association data mining results. After verification on the web pages of large e-commerce websites, it is found that the proposed algorithm can ensure the integrity of web crawling, accurately obtain frequent itemsets targeting purchasing behavior, and precisely mine associated data in e-commerce web pages. While providing useful references for e-commerce platform operation, it also provides better decision support for merchants and consumers.

This study aims to improve the accuracy of high-resolution water body extraction in the Sanjiangyuan region. Using sub-meter Jilin-1 satellite imagery, a self-supervised pretraining approach based on the ViT(Vision Transformer) backbone is developed, followed by fine-tuning a water body extraction model with ViT-B as the encoder and UperNet as the decoder. The results demonstrate that the proposed method achieves an IoU( Intersection over Union) of 92. 93% , outperforming non-pretrained models by 11. 87% and surpassing other open-source pretrained weights. A total of 25 621 water body patches with an area of 1 544 km² are extracted. Comparative analysis with WorldCover 2021 reveals that the method effectively detects small water

bodies missed by previous methods, providing a more accurate reference for water resource analysis in plateau wetlands.

Due to the flexible reconstruction requirements of multi-source heterogeneous data, which require handling of data details and abnormal situations, it is difficult to classify multi-source heterogeneous data by capturing its hidden features and integrating similar data, resulting in low classification accuracy. Therefore, a multi-source heterogeneous data automatic classification method that integrates HHO(Haris Hawks Optimization) algorithm and hybrid clustering algorithm is proposed. Using low rank representation strategy to capture hidden features of multi-source heterogeneous data, its global structural information is obtained, and data quality and consistency are improved. A hybrid clustering algorithm is designed by combining ant colony clustering algorithm and K-means algorithm. By using K-means algorithm to correct the ant nest reduction error in the combined ant colony clustering algorithm, similar data grouping of multi-source heterogeneous data is achieved. Combining

hybrid clustering algorithm with HHO algorithm, by updating individual fitness values, the optimal class center of HHO algorithm is determined to achieve automatic classification of multi-source heterogeneous data after similar grouping. The experimental results show that the proposed method can automatically classify heterogeneous data from multiple sources and has clear category boundaries and compact data point distribution, resulting in higher true case values on the diagonal of the confusion matrix. The average cohesion is as high as 98. 86°,demonstrating high classification accuracy.

To address the limitations of existing unimodal recognition methods for depression ( e. g. , EEG(Electroencephalogram), facial expressions, text analysis), which are susceptible to environmental noise and exhibit limited accuracy, an innovative MFF-DR-DWA(Multimodal Fusion Framework is proposed for Depression Recognition with Dynamic Weight Adjustment). The framework dynamically integrates three heterogeneous data modalities-electroencephalography ( EEG ), linguistic features, and facial expressions-while employing an adaptive weighting algorithm to optimize the contribution of each modality. Experimental results on benchmark datasets demonstrate that the proposed model achieves an accuracy of 99. 09% , an F1 -score of 99. 12% , and an AUC(Area Under Curve) of 99. 97% , significantly outperforming conventional unimodal approaches and static

fusion methods. This multimodal fusion strategy effectively enhances the accuracy and robustness of depression recognition.