The online health of mutual inductor directly affects the reliability of measurement and protection, and it is difficult for traditional offline detection to capture the performance drift. Therefore, this paper develops an intelligent mutual inductor quality monitoring platform that integrates cloud computing and multi-dimensional indicators, carries out testing based on four core parameters and nine characteristic indicators, combines lightweight Long Short-Term Memory to realize health state modeling and trend prediction, and relies on edge clouds to cooperate to complete early warning, life prediction, and evaluation. After three months' on-site verification of 68 10kV mutual inductor, the deviation between the comprehensive score and the laboratory calibration is less than 2.7%, the early detection rate of insulation deterioration is 94%, the overall detection rate is 96.3%, and the average relative error is 3.2%. Therefore, the abnormality can be warned in advance in extreme environment. The platform can provide a paradigm for the state perception of distribution networks.
This study proposes a computationally intensive and repeatable framework for large-scale thematic analysis of China literature texts from 1925 to 2024, aiming at supporting the realization of Qualcomm's text processing under the challenge of language complexity, including word segmentation ambiguity, traditional simplified text variants, and diachronic semantic drift. In order to realize large-scale time-aware topic modeling, the authors integrate Latent Dirichlet Assignment (LDA), Dynamic Topic Model (DTM), and Structural Topic Model (STM) into the high-performance computing pipeline, supplemented by word embedding alignment and loop verification. The workflow adopts closed-loop paradigm, quantitative analysis → close reading → model refining, and uses parallel preprocessing and distributed model training to efficiently process century-old text data. This paper presents an advanced computational text analysis based on domain-aware design and optimized performance, which provides a portable blueprint for large-scale cultural data analysis in a language environment with limited resources.
High-performance computing data centers demand reliable and energy-efficient power delivery to manage rapid workload variations and high power density. Traditional transformer systems often show limited efficiency and weak adaptability under high performance computing (HPC) conditions. This study designed a three-level solid-state transformer with intelligent fault-tolerant control and workload-aware optimization. The architecture employed a neutral-point-clamped topology to reduce switching stress, a residual-based mechanism to sustain operation during device faults, and adaptive voltage regulation that tracked real-time workload changes. Experiments on a 50 kW prototype using HPC-oriented traces achieved 95.8–97.1% median efficiency, a 21% improvement in mean time between failures, and notable reductions in tracking error. The results demonstrate that combining multilevel conversion with intelligent control enhances efficiency, resilience, and workload responsiveness, providing a practical route to more scalable and stable HPC data center power delivery.
This study presents AI-driven real-time load balancing (AI-RTLB), a framework for heterogeneous cloud–edge–Internet of Things (IoT) systems. AI-RTLB combines long short-term memory attention-based workload forecasting with adaptive multi-agent reinforcement learning (decentralized actors, shared critic) and a post-optimization layer that enforces energy, fairness, thermal, and power-budget constraints. This design anticipates demand surges, coordinates distributed decisions, and produces service-level agreement (SLA)-aware schedules. Across large data-center traces, edge workloads, and IoT demand logs, AI-RTLB reduces average latency by 16.3% and improves energy efficiency by 21.7% over strong baselines, while increasing throughput and lowering SLA violations (5.3%) with a lower fairness index (0.041). Convergence is faster and more stable than with single-agent deep reinforcement learning, and robustness is maintained under workload noise and mixed job types. An ablation study confirms complementary gains from prediction, multi-agent control, and fairness- and energy-aware optimization. AI-RTLB offers a practical path toward efficient, equitable, and sustainable computing.
Against the background of China’s double-carbon strategy, the authors put forward a double-loop Transformer–Bi-directional (Bi)–long short-term memory–actor–critic framework that realizes the closed-loop coordination of carbon emission prediction and load control through carbon-load coupling loss and a self-calibration feedback mechanism. The experimental results, which are based on the 15-min resolution dispatching data from six provinces in East China, showed that the carbon emission mean squared error of the framework was 0.038 total carbon dioxide (tCO2), the peak–valley load difference was reduced by 8.7%, and the peak-shaving benefit was increased by 5.4%. In addition, the sensitivity analysis results showed that when the carbon emission weight exceeded 0.6 the system will give priority to reducing carbon emissions. The framework successfully realizes the end-to-end collaborative modeling of carbon emissions and elastic loads and provides an extensible solution for low-carbon scheduling and demand side management.
At present, financial risk early warning systems in colleges and universities lack the ability to process real-time data flow, making it difficult to capture short-term risk fluctuations in a timely manner and limiting their accuracy in short-term forecasting. This study builds a real-time data pipeline based on Apache Kafka and Spark Streaming. Short-term financial index prediction and risk classification are realized by combining a bidirectional long short-term memory network with the XGBoost model. In addition, anomaly detection and dynamic threshold adaptive adjustment are carried out through isolated forests to improve the real-time performance and prediction accuracy of the system. Experiments show that the highest rate of misjudgment is about 2.5% under the robustness test, the cross-school accuracy of migration is over 80%, consistency with auditor hits is over 78.5%, and the average detection rate in real-time stream detection is 83.3%. The results of this study verify the efficiency and adaptability of the system.
This article takes the Xitiaoxi River Basin as an example and proposes an ecological water conservancy engineering planning and design method based on a fusion combination genetic algorithm. By dividing the entire ecological water conservancy project into terrain analysis, water resource allocation, and an irrigation system design, a fusion combination genetic algorithm was used to solve and simulate the evolutionary process of nature and gradually find water resource allocation schemes and irrigation methods that meet the conditions. The research results indicated that fusion optimization using combinatorial genetic algorithms can achieve optimal utilization of resources and maximum protection of the ecological environment. Through instance verification, the algorithm had high reliability, was easy to understand and operate, and had high computational efficiency. The research results provided theoretical data support for using a fusion combination genetic algorithm for ecological water conservancy engineering planning and design.
Power transformers are crucial equipment in the power grid because they are essential for ensuring stable grid operation. Sequential machine learning and artificial intelligence diagnostic algorithms often face issues of low efficiency and prolonged processing times with respect to handling large volumes of oil-immersed transformer fault data. In this article, the authors propose a new transformer fault diagnosis method that is based on the parallel AdaBoost-Naive Bayes algorithm. This method allows for resampling and reweighting, making the model pay more attention to samples that are difficult to classify and thereby improving performance on imbalanced datasets. The Spark platform is used for parallel processing of massive data, utilizing the cluster's multiple nodes for efficient fault diagnosis. Experimental results show that compared with traditional diagnostic methods, the proposed method achieves a significant improvement in diagnostic accuracy, with an accuracy rate of 93.38%. The significant speedup ratio achieved by parallel processing technology underscores its effectiveness and advantages in handling large-scale transformer fault data.
With the development of information technology, accounting big data plays an increasingly important role in supporting enterprise decision-making. Aiming to address problems of data security and privacy protection in the existing cloud data storage environment, this article proposes a data integrity verification algorithm based on bilinear pairing and hash functions. By analyzing the characteristics and task processing flow of modern accounting information systems and reviewing current research, an efficient and safe data integrity verification scheme is designed and implemented. This scheme not only verifies the integrity of data indefinitely but also reduces the consumption of resources. The experimental results show that, compared with traditional file transfer protocols and provable data possessions, the new method reduces the communication overhead and has higher verification efficiency when dealing with large files. The research in this article provides new ideas and technical means for improving the security of accounting information systems and has important practical application value.
This study establishes a coupling model between Narrowband Internet of Things (NB-IoT) wireless communication technology and the degree of industrial water pollution, evaluating the influence of NB-IoT on industrial water pollution monitoring precision. Results show that NB-IoT wireless communication technology significantly increases the precision of industrial water pollution monitoring to 0.96 compared to traditional wire communication technology at 0.65. Additionally, the precision of industrial water pollution monitoring also changes with the economic scale, with an economic scale of 800,000,000,000 making the precision reach 0.89. Overall, this study suggests that wireless communication technology based on NB-IoT is more conducive to ecological environment protection through improved industrial water pollution monitoring.
The stability of rock and soil masses has become increasingly critical due to large-scale expansion and landfilling, resulting in frequent landslides that pose significant threats to safety and property. Consequently, soil slope stability monitoring is essential. To mitigate slope instability risks, this study investigates soil slope stability monitoring using big data technology within the context of the internet of things. The research examines slope monitoring techniques and summarizes various methods for detecting slope deformation. By monitoring displacement and deformation, the operational status of slopes can be assessed, safety evaluated, disasters prevented, and adverse social impacts avoided. Collected geological data support the development of slope models, enabling analysis under different damage conditions. The findings indicate that 50% damage corresponds to a warning threshold, while 80% damage triggers an alarm. Simulation results show that slope stability increases with higher internal friction angle and cohesion but decreases as the slope angle increases.
While networks bring convenience to people, more attention must be paid to the security of the network platform. This study combines big-data technology and machine learning (ML) to investigate the application of big-data analysis and cloud-computing technology in network security. First, the data-collection technology of abnormal network behavior is introduced, and the Flume data-collection component and Kafka distributed technology are discussed. Second, the data-processing process of abnormal network behavior and the corresponding algorithm processing are analyzed. Finally, an abnormal network behavior detection model based on big data is constructed and compared with related models based on decision-tree and random-forest (RF) algorithms. The experimental results show that the big-data anomalous network behavior detection technology based on the ML framework can effectively improve the type and efficiency of anomalous network behavior detection, which is of some significance for improving the network security control capability.
Artificial intelligence (AI) has an important significance in the construction of smart grids. There is a problem of low accuracy of behavior detection due to a large number of attacks and long data fragments in the power grid data. In this paper, a behavior detection method based on a generative adversarial network (GAN) is proposed. The method focuses on optimizing the learning strategies of the generator and discriminator based on a GAN. In this regard, the loss function is converged by maximizing the root mean square error of the attack in the discriminator, and the linear activation function is used as the modulation function before the output in the generator. Therefore, the improved GAN can utilize the data enhancement features in the original mode and ensure the stability of attack behavior recognition, and then, effectively detect the attack information in the grid. The experimental results show that the method in this paper can accurately and effectively monitor abnormal attack behavior in the embedded terminals of the smart grid, and its effective detection rate is 97.6%.
Due to the pandemic, there has been a drastic change in the advancement of online learning platforms. This article will help us understand the reasons for the increase and decrease of using online learning platforms. Based on the research conducted, it was observed that almost the majority of the students (48.4%) have not completed the enrolled course. Few of the students have come at least halfway (14.5%) to the completion of the course. And the rest of the students (37.1%) responsibly completed the enrolled course; almost half the students who haven't completed the course indicated that the main barrier faced among the students is the lack of interaction (36.7%).
Outlier detection is an important data mining technique. In this article, the triangle inequality of distances is leveraged to design a pre-cutoff value (PCV) algorithm that calculates the outlier degree pre-threshold without additional distance computations. This algorithm is suitable for accelerating various metric space outlier detection algorithms. Experimental results on multiple real datasets demonstrate that the PCV algorithm reduces the runtime and number of distance computations for the iORCA algorithm by 14.59% and 15.73%, respectively. Even compared to the new high-performance algorithm ADPOD, the PCV algorithm achieves 1.41% and 0.45% reductions. Notably, the non-outlier exclusion for the first data block in the dataset is significantly improved, with an exclusion rate of up to 36.5%, leading to a 23.54% reduction in detection time for that data block. While demonstrating excellent results, the PCV algorithm maintains the data type generality of metric space algorithms.
The density local outlier factor algorithm (LOF) needs to calculate the distance matrix for k-nearest neighbor search. The algorithm has high time complexity and is not suitable for the detection of large-scale data sets. A local outlier detection algorithm is proposed based on grid query (LOGD). In the algorithm, the k other data points closest to the data point in the target grid must be in the target grid or in the nearest neighboring grid of the target grid, it is used to improve the neighborhood query operation of the LOF algorithm, the calculation amount of the LOF algorithm is reduced in the neighborhood query. Experimental results show that the proposed LODG algorithm can effectively reduce the time of outlier detection under the condition, the detection accuracy of the original LOF algorithm is basically the same.
In order to solve the problems of high distortion rate and low decoding efficiency of the decoded video when the current coding and decoding methods are used to encode and decode the remote video monitoring system, considering the local area network, research on the optimization method of the coding and decoding of the remote video monitoring system is proposed. The local area network is used to collect image information, to process, and to output the image information. By preprocessing the remote video monitoring system, the low frame rate remote video monitoring system is decoded in parallel. The motion information of the lost frame is estimated to realize the fast coding and decoding of the remote video monitoring system. The experimental results show that the proposed method has low distortion rate and high decoding efficiency and has high practical value.
Earlier detection and classification of squamous cell carcinoma (OSCC) is a widespread issue for efficient treatment, enhancing survival rate, and reducing the death rate. Thus, it becomes necessary to design effective diagnosis models for assisting pathologists in the OSCC examination process. In recent times, deep learning (DL) models have exhibited considerable improvement in the design of effective computer-aided diagnosis models for OSCC using histopathological images. In this view, this paper develops a novel duck pack optimization with deep transfer learning enabled oral squamous cell carcinoma classification (DPODTL-OSC3) model using histopathological images. The goal of the DPODTL-OSC3 model is to improve the classifier outcomes of OSCC using histopathological images into normal and cancerous class labels. Finally, the variational autoencoder (VAE) model is utilized for the detection and classification of OSCC. The performance validation and comparative result analysis for the DPODTL-OSC3 model are tested using a histopathological imaging database.
For the problems of low decoding accuracy, long decoding time, and low quality of decoded video image in the traditional lossless decoding method of compressed coded video, a lossless decoding method of compressed coded video based on inter frame difference background model is proposed. At the coding end, the inter frame difference background model is used to extract the single frame video image, and the mixed coding method is used to compress the video losslessly. At the decoding side, the CS-SOMP (compressive sensing-synchronous orthogonal matching pursuit algorithm) joint reconstruction algorithm is composed of synchronous orthogonal matching pursuit algorithm (SOMP) and K-SVD (kernel singular value decomposition) algorithm to losslessly decode the compressed encoded video. The simulation results show that the lossless decoding method based on the inter frame difference background model has higher accuracy, shorter decoding time, and ensures the quality of the decoded video image.
This article discusses the importance of designing an efficient medium access control (MAC) protocol for wireless sensor networks (WSNs) to optimize energy consumption at the data link layer while transmitting high traffic applications. The proposed protocol, EE-MMAC, is an energy-efficient multichannel MAC that reduces energy consumption by minimizing idle listening, collisions, overhearing, and control packet overhead. EE-MMAC utilizes a directional antenna and periodically sleep technique in a multi-channel environment. Nodes exchange control packets on the control channel to choose a data channel and decide the beam direction of the flow. Simulation results show that EE-MMAC achieves significant energy gains (30% to 45% less than comparable IEEE 802.11 and MMAC) based on energy efficiency, packet delivery ratio, and throughput.
