Optimization of sensor networks and data processing algorithms in ecological environment monitoringWu, Yonghui
doi: 10.1117/12.3114194pmid: N/A
Ecological environment monitoring has put forward higher requirements for data transmission efficiency and monitoring accuracy. Aiming at the problems such as uneven network energy consumption and insufficient data quality existing in traditional systems, this paper constructs a comprehensive algorithm system that integrates sensor network optimization and intelligent data processing. The method achieves node deployment and topological structure optimization at the network layer by improving the particle swarm optimization algorithm, thereby enhancing the monitoring coverage rate and link stability. Anomaly detection, adaptive filtering and multi-source fusion algorithms are designed at the data layer to achieve the collaborative optimization of noise suppression and feature extraction. The experimental results show that the algorithm proposed in this paper reduces the average energy consumption by 17%, extends the life cycle by 22%, and the overall transmission success rate exceeds 94%. In terms of data processing verification, the noise suppression rate has increased by 12.3%, and the fusion accuracy has improved by 10.6%. This model effectively enhances the transmission reliability and data processing quality of the ecological monitoring system in complex environments, providing technical support for intelligent environmental monitoring and ecological decision-making.
A BERT sentiment analysis model integrating dynamic layer attention and parallel networksHan, Wangyuan
doi: 10.1117/12.3114181pmid: N/A
To address the issues of insufficient semantic utilisation in pre-trained model intermediate layers and the limited feature extraction capabilities of traditional serial architectures, this paper proposes an enhanced BERT model integrating dynamic layer attention with a parallel architecture. The model first introduces a dynamic layer attention mechanism, adaptively fusing the hidden states from BERT's final four layers to balance deep and shallow semantic features. Subsequently, a parallel feature enhancement network is constructed: BiLSTM captures global temporal dependencies while TextCNN extracts local key phrases, ultimately merging multidimensional representations for classification. Experimental results on the SST-2 dataset demonstrate an accuracy of 91.80% and an F1 score of 90.79%. These findings validate the model's superiority in enhancing classification performance and robustness, offering a more effective solution for text sentiment analysis within complex contexts.
Depth modeling and congestion prediction algorithm of large-scale spatio-temporal data of urban traffic flowZou, Jiahe; Wang, Yulin; Liu, Yuhan
doi: 10.1117/12.3113016pmid: N/A
Aiming at the challenges of limited static graph structure, insufficient multi-modal information fusion and poor robustness of long-term prediction in large-scale spatio-temporal data prediction of urban traffic flow, this paper proposes a congestion prediction framework based on Dynamic Spatio-Temporal Graph Network (DSTGN). The core contributions include: (1) designing an adaptive adjacency matrix learning module to dynamically capture the real-time correlation strength between nodes through attention mechanism, breaking through the traditional static graph assumption; (2) A cross-modal attention fusion mechanism is proposed, which uses BERT to encode the semantics of the event text, and realizes the deep alignment between the text and the traffic state through multi-head attention, thus enhancing the ability of emergency perception; (3) Construct a backbone network with the collaboration of space-time graph convolution and Transformer encoder, model local space-time dependence and global long-range correlation respectively, and adopt course learning strategy to realize multi-scale prediction from 15 minutes to 1 hour. Experiments on two real-world city datasets demonstrate that DSTGN achieves Mean Absolute Error (MAE) values of 4.78/5.64/6.85 for 15/30/60-minute forecasting tasks, respectively. This represents a 5%-12% improvement over advanced baselines such as Diffusion Convolutional Recurrent Networks (DCRNN), while exhibiting the smallest performance degradation over extended forecasting horizons. Ablation experiments verify the effectiveness of dynamic graph structure, cross-modal fusion and global coding module, and visual analysis shows that the dynamic association learned by the model conforms to the traffic flow propagation law. This study provides an effective scheme for high-precision and interpretable congestion prediction by fusing multi-source heterogeneous data.
Algorithm improvement and empirical verification of time series data mining in financial index fluctuation predictionJiao, Lei
doi: 10.1117/12.3113495pmid: N/A
In this paper, a hybrid time series mining algorithm based on attention mechanism and graph neural network (GNN) is proposed to solve the challenges of traditional methods in financial index fluctuation prediction, such as difficulty in capturing nonlinear dependence, adapting to concept drift and automatic feature selection. In this method, the dynamic correlation between financial indicators is modeled by graph attention network, and the time series dependence is captured by Transformer encoder. The online learning framework driven by reinforcement learning (RL) is introduced to dynamically adjust the model parameters to cope with the changes in market conditions. The sparse group Lasso is used to realize the adaptive selection of multi-granularity features (technical indicators, macroeconomics, market sentiment) and improve interpretability. This paper empirically selects the data of 300 constituent stocks in Shanghai and Shenzhen from 2010 to 2023 to predict the volatility of return in the next 5 days. The results show that the root mean square error (RMSE) of the proposed model is 1.36×10⁻³ on the test set, and the directional accuracy is 67.4%, which is 8.1% and 2.9% higher than the optimal baseline MTGNN, especially in the turbulent city environment. The ablation experiment verifies the effectiveness of the graph structure, online learning and feature selection module, and the feature importance analysis reveals the leading role of market sentiment and technical indicators. The research provides a more adaptable and interpretable technical framework for financial time series prediction.
Optimization of computer algorithm for solving power balance constraints with high proportion of new energyXiao, Qiang; Zhang, Mi; Ma, Jingbo; Zhang, Jian; Chen, Yuguo
doi: 10.1117/12.3113001pmid: N/A
To address the power balance challenges caused by high renewable energy penetration, this study focuses on optimizing constraint-solving computer algorithms. By analyzing the limitations of traditional algorithms like linear programming under fluctuating scenarios, we develop a multi-temporal scale balancing model and design a distributed solving algorithm with parallel framework to enhance computational efficiency and robustness. Simulation experiments demonstrate that the optimized algorithm achieves a 40% speed improvement and maintains accuracy within 5% in high renewable energy penetration scenarios. The research provides technical support for the safe and stable operation of high-renewable energy power systems, facilitating renewable energy integration and grid dispatch optimization.
Design of unified control platform architecture for smart campus based on Internet of Things and edge computingShi, Naixin
doi: 10.1117/12.3112989pmid: N/A
In current smart campus development, challenges such as data silos from isolated systems, delayed equipment responses, and coarse energy management hinder management efficiency. This paper proposes an IoT and edge computing-based unified control platform architecture for smart campuses, comprising four layers: sensing, edge computing, networking, and application. The sensing layer collects multi-scenario data, the edge layer processes data locally to reduce latency, the networking layer ensures secure transmission, and the application layer delivers multi-scenario services. The platform achieves over 99.7% device access success rate and 31.2% energy consumption reduction, effectively meeting campus management needs.
Application of large language models in automated grammar correction for vocational college students' EnglishLuo, Wenli
doi: 10.1117/12.3113781pmid: N/A
Leveraging encoder-decoder large language models, this study presents an automated grammar correction system designed specifically for English writing by vocational college students. By using a sequence-to-sequence neural architecture, domain-adaptive fine-tuning, and targeted data augmentation, the system is able to effectively handle complex grammatical errors commonly found in vocational education. The training and evaluation of the model are based on a strictly annotated real student composition corpus. The framework proposed in this paper has achieved significant improvements in sentence-level accuracy, recall and F1 value through experimental comparison with traditional rule and statistical error correction methods. It consistently maintains high performance across various proficiency levels and technical fields. Integrated into the education platform through a modular API, it demonstrates high throughput and low latency deployment, and validates system robustness under stress conditions involving a large number of real-time user interactions. The results show that the advanced LLM-based solution can provide reliable, scalable and context-related grammar correction, which directly supports the needs of middle school students and educators for dynamic learning environment. The technical approach described in this paper provides a replicable paradigm for the engineering design of AI-assisted writing support systems in vocational English education.
Implementation of real-time rhythm synchronization detection and optimization algorithm for percussion orchestrasXue, Fei
doi: 10.1117/12.3113011pmid: N/A
To address the problems of accumulated rhythm deviations and the difficulty of real-time synchronization in percussion ensemble performances, this paper proposes a real-time rhythm synchronization detection and optimization algorithm that integrates signal processing and optimization control. The algorithm collects rhythm signals through a multi-channel microphone array, combines short-time Fourier transform (STFT) to extract beat features, and introduces the rhythm consistency mechanism theorem to ensure the convergence and stability of the feature sequence. In the dynamic correction stage, an adaptive feedback regulation model is constructed, and in the collaborative optimization stage, global synchronous control is achieved based on the Lyapunov stability principle. The experimental results show that the synchronization error of this algorithm in multi-channel rhythmic tasks is reduced from 0.084 to 0.052, which is 38.1% lower than that of the traditional time alignment algorithm. The response delay is shortened from 0.95 s to 0.68 s, the consistency coefficient is increased to 0.94, and the overall performance score is improved by 6.7%. In addition, the optimization mechanism has increased the convergence speed of rhythm errors by 32.6%. The system remains stable and synchronized under fast rhythms and complex beats, verifying the algorithm's high precision, rapid response and strong adaptability.
An improved PPO reinforcement learning algorithm for robotic arm trajectory planningWang, Zenghong
doi: 10.1117/12.3114251pmid: N/A
To address the slow convergence and low optimization efficiency of traditional algorithms for robotic arm trajectory planning in dynamic and complex environments, this paper proposes a trajectory planning method based on an improved Proximal Policy Optimization (PPO) reinforcement learning algorithm. By introducing an adaptive learning mechanism and optimizing the reward function, the proposed method establishes a control framework tailored to the continuous action space of robotic arms. Experiments were conducted on a 5-DOF(5-degree-of-freedom) KUKA robotic arm in the PyBullet simulation environment, accumulating 600000 training steps. The results indicate that the reward value rapidly exceeds 1000 after approximately 250 episodes, with the smoothed reward stabilizing between 1200 and 1300. Furthermore, the value function loss converges quickly and remains near zero, enabling stable grasping within a 10 cm distance threshold. This method effectively enhances convergence speed and control precision, offering a viable solution for real-time and robust trajectory planning in dynamic scenarios.
New flower color simulation and breeding target-oriented algorithm based on generative adversarial networksMang, Guangwei; Tan, Yufeng; Ren, Jie; Kang, Liang; Gong, Yu
doi: 10.1117/12.3113520pmid: N/A
To enhance the diversity of flower species and achieve the simulation of flower colors, research is conducted on the simulation of new flower colors and breeding goal-oriented algorithms based on the application of Generative Adversarial Networks (GANs). A residual network is employed to construct a generator for training flower color feature samples. A color mixing channel is introduced to simulate new flower colors by adjusting the weights of different color channels. Genetic algorithms are utilized for flower color design, considering the fitness of flower combinations, to achieve research on breeding goal-oriented algorithms for new flower color combinations. Experimental results show that after applying this method, standardized generation of simulated samples for new flower colors can be achieved, avoiding color confusion. Meanwhile, comparative tests demonstrate that the design method results in less loss in the simulation of new flower colors and achieves optimal breeding goal-oriented results for new flower colors.