Due to differences in the environment of different network channels, delay control is difficult. To solve this problem, a transmission delay consistency control algorithm based on discrete mathematics model is proposed. First, based on the discrete mathematics model, the transmission time matching function in a single channel unit is designed, and the data transmission delay is estimated to establish the transmission delay function. Then, discretization processing is adopted to obtain the limiting spectrum parameters with the greatest correlation with transmission delay parameters. Using this parameter as a reference, the conditions that the maximum allowable delay needs to meet is set, a balanced transmission amplitude response model is established based on the balanced signal-to-noise ratio, and delay consistency equalization control is implemented using adaptive equalization scheduling method. Experiments show that the transmission delay and control overhead are significantly reduced, and the network throughput is increased after the application of the algorithm.

In the process of super-resolution processing, the noise inherent in low resolution images will be amplified, resulting in distortion of super-resolution images. To this end, a super-resolution image noise recognition method based on neural networks has been proposed. The Activation function in the neural network is used to determine the peak signal to noise ratio. By combining the noise data set and hyperparameter coefficients, residual values are obtained, and combined with the noise information distribution density, super- resolution image noise recognition is achieved. The experimental results show that the proposed method has high clarity and good recognition performance in super-resolution images, with a maximum peak signal-to-noise ratio of 50 dB, indicating that the use of the proposed method can improve image quality.

An improved scorpion-simulated method for estimating the position of a vibration source is proposed, to address the issue of limited positioning accuracy due to incomplete scorpion-simulated vibration source positioning models. Based on the different attenuation rates of vibration signal frequency components with increasing distance, the method first improves the existing scorpion vibration source positioning mechanism to address the lack of distance estimation in current models. Then, it extracts the features of the vibration source location based on the improved model and constructs a complete feature database. Next, the K-means method is used to create sub-databases to reduce data matching time. Finally, the KNN(K-Nearest Neighbor)algorithm is used to estimate the position of the vibration source. To verify the effectiveness of the proposed method, tests are conducted on a dataset containing 64 locations of personnel stepping signals in a concentric circle with an inner radius of 2 m and an outer radius of 6 m. The results show that the improved method for simulating scorpion vibration source positioning increases the average positioning accuracy by 0. 439 6 m compared to the existing method and improves the positioning time by 31. 34% compared to the original matching method. Therefore, the improved method has achieved significant enhancements in both positioning accuracy and efficiency.

The current challenges in deep learning for 3D reconstruction are difficulty in extracting comprehensive scene information from images and insufficient consideration of depth consistency between views. A multi-view stereo network with adaptive depth consistency and cross-frequency attention (ADCFA-MVSNet: Multi-View Stereo with Adaptive Depth Consistency and Cross-Frequency Attention) is proposed. The CFA (Cross-Frequency Attention) module integrates high-frequency, low-frequency information within images and global scene information across views, enabling more comprehensive feature extraction. The AD(Adaptive Depth) consistency module precisely captures the geometric structure of the scene and dynamically considers the contribution of different views to depth consistency, enhancing it across various scales. The innovation of this method lies in utilizing comprehensive image information to ensure geometric consistency, achieving excellent performance in 3D reconstruction tasks. On the DTU(Technical University of Denmark) dataset, it achievs an accuracy of 0. 319, completeness of 0.285, and an overall score of 0.302, surpassing other methods. On the BlendedMVS dataset, the EPE(End-Point-Error) score is 0.27, e1 score is 5.28, and e3 score is 1.84, outperforming other methods. These results demonstrate the effectiveness of ADCFA-MVSNet in improving the completeness and accuracy of multi-view 3D reconstruction. Experimental results show that this method enhances the quality of multi-view reconstruction and achieves good reconstruction effects.

 Financial announcement is the carrier for enterprises to publicly inform the society of major financial events, and its information is of great significance to financial practitioners. However, financial events have the characteristics of strong argument specialization and high dispersion, and traditional event extraction methods are difficult to achieve accurate extraction. Therefore an event extraction method combining the local features of paragraphs and the global features of documents is proposed. This method first segments the financial announcement document, and then uses all the paragraphs in parallel Fin-BERT(Financial Bidirectional Encoder Representation from Transformers ) Pre training model, convolutional neural network and self attention mechanism to obtain local features of documents. Then Bi LSTM(Bi directional Long Short Term Memory) is used to learn the semantic information of the whole document to obtain the global features of the document. Finally, the local features of the paragraph and the global features of the document are fused to output event arguments and event types. A series of experiments are carried out on the financial open data set chfinann. The experimental results show that the method achieves an average F1 value of 80. 2%, which is better than the baseline model, and proves the effectiveness of the method. 

In order to solve the problem that the accuracy of deep neural network model learning is affected by the small sample of rice disease image dataset, an improved adversarial generative network model ViT-WGAN-GP (The Fusion of Vision Transformer and Wasserstein Generative Adversarial Networks with Gradient Penalty) is proposed for enhancing the image dataset. Firstly, the Vision Transformer structure is introduced in the generation model to enhance the learning of global features. Secondly, the WGAN-GP structure is used in the discrimination model to ensure the stability of the model training and improved the effect of the generated images through the Wasserstein measure function and the gradient penalty term. Finally, the enhanced sample set is used to train the deep neural network model. The experimental results show that the ViT-WGAN-GP model generates images with significant improvement compared with GAN and WGAN-GP. The average accuracy of rice disease recognition is 94. 3%,96. 2%, and 97. 5% for VGG16, ResNet34, which are improved by 9. 7%, 2. 8%, and 4.8%, respectively. The proposed ViT-WGAN-GP model can generate more realistic rice disease images and can improve the recognition accuracy of deep neural network models significantly with small sample sets.

Aiming at the problem of low accuracy of methods in Chinese word segmentation in professional domains due to the mismatch of cross-domain data distribution and the limitation of a large number of unregistered professional words, a professional domain word segmentation method based on XLBMC(XLNet- BiGRU-Multi-head Self-attention-Conditional Random Field) is proposed. Firs, the dynamic word vectors containing contextual semantic information is generated through an improved XLNet pre-training model, enabling the model to better utilize the boundary and semantic knowledge. Then the acquired word vectors are input into BiGRU for feature extraction to obtain the hidden state representation of each character. On the basis of BiGRU coding, a sparsified MHSA(Multi-head Self-Attention) mechanism is introduced to weight the representation of each character, which improves the prediction accuracy of the model for fine-grained and strongly long-term dependent time series under restricted memory budget. Finally, the CRF(Conditional Random Field) decodes the dependencies between neighboring tags and outputs the optimal segmentation sequence. Segmentation experiments are conducted on a self鄄constructed control engineering corpus. The results show that the accuracy of the proposed segmentation model is 94. 27%, the recall is 93. 24%, and the F1 value is 95.52%, which proves the reliability of the model in Chinese segmentation tasks in the professional domain.

 Deep neural network techniques based on supervised learning have been extensively employed in the field of welding. However, in real industrial scenarios, obtaining labeled data samples is challenging, which limits the performance of deep network models. For unlabeled or partially labeled datasets, transfer learning algorithms offer a novel solution. Transfer learning algorithms in terms of domain adaptation and pre-training-fine- tuning are introduced and a summary of current research on their development over recent years and their applications in weld penetration identification and defect detection is provided. And transfer learning issues in welding that require further attention and exploration in the futureis highlighted. Transfer learning methods can enhance the effectiveness of deep learning models by better utilizing existing data and knowledge in the welding field, specifically in weld penetration recognition and welding defect detection accuracy. This promotes the development of intelligent welding manufacturing technology. 

In community emergency management during scenarios such as pandemic disasters, traditional methods can not quickly and accurately capture community dynamics. Therefore, an intelligent perception system combining MLLMs(Multimodal Large Language Models) and YOLOv8(You Only Look Once version 8) is proposed. The system comprehensively analyzes textual data from social media and community video surveillance streams to identify changes in community public service needs in real-time. Experimental results demonstrate high accuracy and responsiveness in demand recognition and anomaly detection. This enhances the responsiveness of public services in emergency management and provides strong technical support for smart city development.

Intelligent vehicle is a strategic focus of the global automotive industry. To meet new demands for experimental teaching in this field, we designed and constructed a camera-injection experimental platform. First, a virtual-physical integration scheme is implemented by combining simulation environments with an in-vehicle domain controller to define the overall architecture. Second, a camera simulation model is developed and virtually calibrated , on which a data link is built using GMSL2(Gigabit Multimedia Serial Link 2) and CSI-2 (Camera Serial Interface 2) protocols to enable seamless interaction between simulated video streams and controller hardware. Next, we quantized offline-trained deep-learning models, converted them into a universal format, and deployed them on the domain controller for real-time interaction with the virtual environment. Finally, this platform is used to provide students with an integrated theory-and-practice learning environment, deepening their understanding of camera principles and mastery of multi-channel video signal generation, transmission, and perception. This system effectively enhances students‘ practical skills and innovation capacity in intelligent vehicle technology.

In order to improve the real-time performance and security of the traditional A* algorithm during path planning, a path planning method incorporating improved A* and DWA(Dynamic Window Approach) is proposed. Firstly, the search neighborhood of the A* algorithm is optimized to reduce the search direction of nodes. Secondly, the search mechanism is optimized by introducing bidirectional search and dynamically defining the target node strategy to carry out bidirectional path searching from the start node and the target node. Dynamic weighting coefficients are introduced to reduce the generation of redundant nodes in the process of path searching, and the paths are smoothed by the Bezier curves. And lastly, the improved A* algorithm is fused with the DWA algorithm to realize dynamic obstacle avoidance. Simulation is carried out using PyCharm, and the results show that, compared with the other two algorithms, the search nodes of the improved A* algorithm are reduced by more than 46.25%, and the search time is reduced by more than 24.06%. The integrated algorithm is able to realize dynamic obstacle avoidance, and the smoothness and safety of the planned paths have been improved.

In order to improve the accuracy and convenience of obtaining medical information for patients with chronic diseases, a chronic disease health knowledge Q&A(Question and Answer) system based on knowledge graph and WeChat mini program has been developed. The system includes question and answer library, server- side, and client-side. The question and answer library is built based on knowledge graph, text generation and matching technology, and the server-side is written in Python language, using WeChat mini program as the client. In terms of functionality, the system is capable of intelligent question answering, health data recording, health tweets, and knowledge base updates for disease knowledge. The application results indicate that the system can effectively complete the health knowledge question and answer task for chronic patients.

When detecting false data injection attacks, if the detection accuracy of the detection model is poor, it will directly affect the detection effect of false data injection attacks. In order to effectively improve the detection accuracy of the detection model, the incremental detection of false data injection attacks based on the grey wolf algorithm is proposed to optimize the kernel extreme learning machine. The state of power system is estimated, and the attack behavior of false data injection is analyzed. On this basis, an incremental detection model of false data injection attack is established based on kernel extreme learning machine, and the model is optimized by grey wolf algorithm. Finally, the normalized results of the collected power system state data are used as the model input data, and the accurate detection of false data injection attacks in power system under incremental changes is realized through the optimized model. The experimental results show that using this method to detect false data injection attacks can get better detection effect and high precision results.

The network nodes are fragile, with many potential attack behaviors and redundant intersection situations, resulting in poor feature recognition accuracy and classification performance and low monitoring stability and efficiency. Therefore, a network potential attack monitoring based on fuzzy Markov game algorithm was studied. Using the fusion degree compressed sensing method and the feature recognition degree parameter analysis method, the random discrete distribution sequence of network potential attack characteristics is analyzed, the characteristics of network potential attack spectrum is also extracted and analyzed. The random forest algorithm is adopted to distinguish the types of potential network attacks, and the fuzzy Markov game analysis of potential network attack risk is carried out. According to the risk state set and the principle of minimum and maximum, the potential network attack risk is monitored. The test results of the example show that after the proposed method is applied, potential attack behavior parameters are set, and the fluctuation of potential attack recognition rate is small. The fuzzy Markov game analysis results are closest to the actual risk value, and have high recognition accuracy, monitoring efficiency, and monitoring stability. 

Drug information data has the characteristic of imbalanced categories, with poor interpretability and a large number of sensitive data. The application effect and mining accuracy of sensitive data are low. Therefore, an improved Apriori based sensitive data mining algorithm for drug information is proposed. The drug data is decomposed into several band limited intrinsic mode functions, and is updated and denoised, the feature subset of the sensitive data is extracted according to the information gain of the feature subset of the drug sensitive data and the Monte Carlo sampling strategy. The relationship between the hidden layer output function and the feature subset is analyzed. The extreme learning machine is introduced to improve the Apriori algorithm. And the drug combinations with significant relevance are screened out and solved. The sensitive data features are matched corresponding to the candidate feature subset and a sensitive data mining function is constructed. The experimental results show that the data signal fluctuation amplitude is small, and sensitive data can be clearly distinguished. The number of erroneous data mined does not exceed 2, improving the interpretability of sensitive data.

Since network simulation consumes a lot of system resources and has complex parameter configuration, these factors will affect the test accuracy when performing network testing. Therefore, a lightweight network simulation method based on Linux namespace is designed. Based on Linux kernel virtualization technology, a simulation network is built. In order to solve the problem of complex parameter configuration caused by dynamic adjustment of topology during network testing and high resource consumption when running network simulation system, human-readable data serialization standard file is used to flexibly define the network, and completes the virtual network creation by calling system commands in user state with the help of Linux network namespace mechanism. An automated script is written to perform network testing using iperf3 tool. Experimental results show that the network throughput of the proposed method is close to the test theoretical bandwidth, the system shows high startup efficiency and low running resource overhead, can meet common network testing application scenarios, has good performance, and has certain research value.

When there is a large amount of text data, it is necessary to extract effective features from the text data to capture important information of the text to facilitate the storage and querying of the text. Therefore a machine learning based text classification and label prediction algorithm research is proposed. Conditional random field method is used to annotate and segment the part of speech of the processed text, and obtain the features of the texrt. Text features are inputted into a self attention mechanism recurrent convolutional neural network, and after model training, the classification results and label prediction results of the text outputted. After experimental verification, the proposed algorithm can effectively complete text classification and label prediction, with an average false rate of 95. 2% in text classification and an average loss of 0. 4% in text prediction ranking.

Due to the fact that resource filtering and deduplication are essential steps in ensuring the efficient operation of digital libraries, the process is susceptible to interference from redundant data, resource types, and differences in customer groups. Therefore, a digital information resource filtering and deduplication method based on GRNN algorithm is proposed. Firstly, the GRNN(General Regression Neural Network) algorithm is used to detect outliers in digital information resources, and the outliers are filtered through PSO-LSSVM(Purticle Swarm Optimization-Least Squares Support Vector Machine) to avoid interference from outlier data in the deduplication process. Then, a locally sensitive hash algorithm is used to convert the resource data into binary hash codes, and the filtering and deduplication of digital information resources are completed by detecting the Hamming distance similarity between hash codes. The experimental results show that this method takes short time and has high precision and rate of deduplication.

The leakage of well site pumping units is an important issue that affects the safety production and stable operation of oil fields. The current object detection methods often overlook the special requirements of well site leakage detection, and there are some limitations in the process of feature recognition of well site leakage targets. An attention mechanism leak detection algorithm for small targets in well sites is proposed based on the YOLOv5(You Only Look Once 5) network, introducing channels and spatiotemporal attention modules into the backbone network, to obtain more feature discrimination information, to enhance the model’s attention to important features. An additional small object detection scale has been introduced in the backbone network, which enables the network to integrate more feature information of small target objects and enhance the detection ability of small targets. The effectiveness of the proposed algorithm is validated on a dataset of well site leaks. The experimental results showed that compared to similar algorithms, the proposed method has higher recognition accuracy and can provide reference for the practical application of automatic detection of oil field well site leaks. 

To address the issue of hesitation degree redistribution in similarity measurement of intuitionistic fuzzy sets, researchers typically rely on fixed formulas yielding constant results, which deviate from practical situations. To resolve this, a novel similarity measurement method for intuitionistic fuzzy sets is proposed. This method innovatively introduces a random function to perform fine-grained redistribution of hesitation degrees into membership and non-membership degrees, while preserving the remaining hesitation degree. This approach significantly enhances the discrimination ability and computational efficiency of the measurement while maintaining information integrity, fulfilling the core requirements of similarity measurement. Furthermore, the possible value range of the similarity between two intuitionistic fuzzy numbers is thoroughly discussed, enhancing the applicability of the similarity measure. The application of this method in the selection of cutting slope schemes verifies its rationality, practicality, and innovation, demonstrating promising application prospects.

n response to the issue that photovoltaic energy storage charging stations have high operating cost and load fluctuations in the distribution grid caused by the shortcomings of the operating strategy, a multi-timescale optimal scheduling strategy based on time-varying boundary constraints for photovoltaic energy storage charging stations is proposed. Firstly, the goal is to minimize the daily operating cost, and the electric vehicle charging load and photovoltaic generation power intervals are predicted. Next, is to minimize the mean square deviation of load fluctuation on the distribution network. Constraints with time-varying boundaries are constructed based on the boundaries of the predicted intervals, grid feed-in pricing of photovoltaic and time-of-use pricing. Finally, setting multiple scenarios and utilizing the CPLEX solver in Matlab for optimization, The simulation results reveal that the scheduling strategy reduces daily operational costs and decreases the mean square deviation of load fluctuation.

The original user side net load sequence of the power system is chaotic. In order to accurately predict the changes in user side load data of the power system, a heterogeneous computing based user side net load prediction algorithm is proposed. The user side net load data of the power system is analyzed with noise, the binary wavelet transform is expanded, and the user side net load data of the power system is preprocessed by setting threshold values and determining estimated signals. The empirical mode decomposition method is applied to decompose the user side net load of the power system. Two different algorithms, EKF(Extended Kalman Filter) and KELM(Kernel Extreme Learning Machine) are used to establish a power system user side net load prediction function based on EKF-KELM. The optimal parameters for IMF(Intrinsic Mode Function) components are calculated isomerically, and a kernel function is introduced to overlay all predicted values. The user side net load prediction results of the power system are obtained under heterogeneous computing. The experimental results show that the predicted value of the power system user side net load obtained by the proposed algorithm is basically consistent with the true value, with low root mean square error and average absolute error. This effectively reduces the time required for power system user side net load prediction and can obtain high-precision power system user side net load prediction results. 

 The environmental factors such as temperature, humidity, and air quality in the office area of power enterprises have dynamic changes and complex relationships with each other. The impact of these factors on energy consumption is difficult to accurately quantify, which increases the complexity of control algorithms. Therefore, an energy-saving power control algorithm for office energy consumption in power enterprises under the dual carbon background is proposed. Using environmental sensors to deploy sensor networks, collecting information data from office environments, and establishing a linear relationship between environmental factors and energy consumption and using regression models to quantify the impact of environmental factors on energy consumption. Based on the linear relationship between environmental factors and energy consumption, the power loss and annual electricity consumption of energy consuming equipment in the office environment are calculated, and the energy consumption changes of the enterprise‘s office during a specific period in the future are predicted. By setting the connection instructions between sensors and energy consumption data with the control center, a device energy consumption regulation constraint model is constructed, and an energy-saving control function is designed based on the energy consumption intensity and modulation power of the device, thereby achieving office  energy consumption energy-saving power control. The experimental results show that under the application of research methods, the energy-saving rate and emission reduction of the office environment in power enterprises are as high as 35% and 35 tons respectively, with good energy-saving and emission reduction effects, which can meet the requirements of the dual carbon target. 

In order to solve the problem of “peak-on-peak” load in distribution networks caused by the superposition of EV ( Electric Vehicle) users’ base load and unordered EV charging load in residential communities, the following solution is proposed Firstly, based on cloud-edge collaboration theory and big data technology, a cloud-edge collaborative optimization scheduling framework is established for comprehensive interconnection of distribution networks, charging station operators, intelligent charging piles, and EV user information. Secondly, an EV user charging scheduling mechanism considering the minimum profit or maximum cost acceptable to users is proposed. Then, a two-layer multi-objective orderly charge and discharge optimization regulation model is established from the perspectives of both the grid side and the user side. Finally, taking EV load data in residential areas as an example, the COA(Coati Optimization Algorithm) is proposed to solve the model. The simulation results verify the effectiveness and superiority of the proposed model and method. It can achieve better peak cutting and valley filling, and improve the user’s charging experience. 

To tackle the inefficiencies, high miss-detection rates, and poor mobility of conventional security check methods in identifying contraband hidden on vehicle undercarriage surfaces, a cloud-based mobile intelligent underbody concealment detection system has been devised. This system uses robotics, network communication, image processing, and object recognition technologies to enable intelligent inspection. Employing the SIFT(Scale-Invariant Feature Transform) algorithm for feature information extraction from images, the system attains a panoramic view of the vehicle underside. Four categories of undercarriage concealments are detected using the YOLOv5(You Only Look Once version 5) deep neural network model. A cloud database is constructed to archive vehicle data, with TCP/ IP(Transmission Control Protocol/ Internet Protocol) protocol facilitating seamless interactions among the wheeled robot, the supervisory computer, and the cloud database. Preliminary testing confirms the system’s capability to conduct underbody concealment inspections within designated contexts, achieving 76.7% success rate in image stitching and 87.2% precision in target detection. Hence, it has considerable practical value for vehicle underbody security check applications.

 Aspect-level sentiment analysis is a prominent research task in the field of natural language processing. Aiming to analyze the sentiment tendencies of different aspects of texts, to address the issues of insufficient interaction between aspect words and context, and to deal with low classification accuracy of existing aspect-level sentiment classification models, an ASMFF(Aspect-level Sentiment classification method is proposed based on Multi-interaction Feature Fusion). Firstly, the context and aspect words are distinctly labeled and fed into the BERT(Bidirectional Encoder Representations from Transformers) coding layer for text feature vector extraction. Secondly, the text feature vectors are fed into AOA (Attention Over Attention) and IAN (Interactive Attention Networks) networks to extract the interactive attention feature vectors. Finally, the two interactive feature vectors obtained are fused and learned, and probability calculation, loss back propagation, and parameter updating are carried out using the cross-entropy loss function. Experimental results on three publicly available datasets, Laptop, Restaurant, and Twitter, show that the classification accuracy of the ASMFF model is 80. 25%,84.38%, and 75.29%, respectively, which is a significant improvement over the baseline model. 

Traditional methods of classroom quality evaluation mainly rely on manual observation, which suffers from low efficiency and poor accuracy. To establish a more comprehensive evaluation system, a lightweight classroom evaluation model based on an improved YOLOv5s(You Only Look Once version 5 small) is proposed. By adopting this model and the AHP(Analytic Hierarchy Process), a comprehensive classroom evaluation system is established. The model integrates the CBAM(Convolutional Block Attention Module) attention mechanism into the neck network, enhancing the model’s recognition accuracy. incorporates the Ghost module into the backbone network, significantly reducing the model’s complexity. and utilizes the Focal Loss function to effectively mitigate the problem of class imbalance. Experimental results show that, compared to the YOLOv5s model, the improved model increases average precision by 7. 3%, reduces the number of parameters by 42. 0%, decreases computation by 33. 1%, and improves detection speed by 4%. Finally, a classroom quality evaluation system is established by combining the AHP and the entropy weight method, dynamically displaying the current classroom quality score, which meets the actual needs of the classroom.