Xiaoxiang Liang

The abnormal fluctuations in network traffic may indicate potential security threats or system failures. Therefore, efficient network traffic prediction and anomaly detection methods are crucial for network security and traffic management. This paper proposes a novel network traffic prediction and anomaly detection model, MamNet, which integrates time-domain modeling and frequency-domain feature extraction. The model first captures the long-term dependencies of network traffic through the Mamba module (time-domain modeling), and then identifies periodic fluctuations in the traffic using Fourier Transform (frequency-domain feature extraction). In the feature fusion layer, multi-scale information is integrated to enhance the model’s ability to detect network traffic anomalies. Experiments conducted on the UNSW-NB15 and CAIDA datasets demonstrate that MamNet outperforms several recent mainstream models in terms of accuracy, recall, and F1-Score. Specifically, it achieves an improvement of approximately 2\% to 4\% in detection performance for complex traffic patterns and long-term trend detection. The results indicate that MamNet effectively captures anomalies in network traffic across different time scales and is suitable for anomaly detection tasks in network security and traffic management. Future work could further optimize the model structure by incorporating external network event information, thereby improving the model’s adaptability and stability in complex network environments.

Abnormal fluctuations in financial markets may signal significant risks or market manipulation, so efficient time series anomaly detection methods are crucial for risk management. However, traditional statistical methods (e.g., ARIMA, GARCH) are difficult to adapt to the nonlinear and multi-scale characteristics of financial data, while single deep learning models (e.g., LSTM, Transformer) have limitations in capturing long-term trends and short-term fluctuations. In this paper, we propose WaveLST-Trans, a financial time series anomaly detection model based on the combination of wavelet transform (WT), LSTM and Transformer. The model first uses wavelet transform to perform multi-scale decomposition, extracts low-frequency trend and high-frequency fluctuation features, and feeds them into LSTM (to learn the long-term dependence) and Transformer (to capture local mutations) respectively, and finally integrates the multi-scale information through the feature fusion layer, which improves the detection accuracy and robustness. The experiments are conducted on Binance (cryptocurrency market) and S&P 500 (stock market) datasets, and the results show that WaveLST-Trans mostly outperforms the mainstream models in terms of F1-score, recall, and precision, and improves the detection performance by 3% and 10% in high-frequency market and long-term trend market, respectively. This study provides a more accurate and stable anomaly detection method for financial market risk management, which can be widely used in market regulation, quantitative trading and financial risk control, helping to improve the security and stability of the financial system.

With the increasing global focus on renewable energy and the growing proportion of renewable power in the energy mix, accurate forecasting of renewable power demand has become crucial. This study addresses this challenge by proposing a multimodal information fusion approach that integrates time series data and textual data to leverage complementary information from heterogeneous sources. We develop a hybrid predictive model combining CNN and Bi-GRU architectures. First, time series data (e.g., historical power generation) and textual data (e.g., policy documents) are preprocessed through normalization and tokenization. Next, CNNs extract spatial features from both data modalities, which are fused via concatenation. The fused features are then fed into a Bi-GRU network to capture temporal dependencies, ultimately forming a robust CNN-Bi-GRU model. Comparative experiments with ARIMA, standalone GRU, and EEMD-ARIMA (a hybrid model combining ensemble empirical mode decomposition with ARIMA) demonstrate the superiority of our approach in both short- and long-term forecasting tasks on the same dataset. This research offers a potential framework to enhance renewable power demand prediction, supporting the industry’s sustainable growth and practical applications.

Optimizing transportation routes to improve delivery efficiency and resource utilization in dynamic supply chain scenarios is a challenging task. Traditional route optimization methods often struggle with complex supply chain network structures and dynamic changes, which require a more efficient and flexible solution. This study proposes a method that integrates Graph Neural Networks (GNNs), self-attention mechanisms, and meta-reinforcement learning (Meta-RL) in order to address route optimization in supply chains. The goal is to develop a path planning method that excels in both static and dynamic environments. First, GNNs model the supply chain network, converting node and edge features into high-dimensional graph representations in order to capture local and global network information. Next, a Transformer-based strategy network captures global dependencies, optimizing path planning. Finally, Meta-RL enables rapid strategy adaptation to dynamic changes (e.g., new demand points or route disruptions) with minimal sample support. Experiments on multiple supply chain datasets show that our method improves path planning quality by about 7%, compared to traditional methods, achieving a path coverage of 92.29%. Ablation studies reveal that the on-time delivery rate improves by nearly 30% over the baseline model. These results demonstrate that the proposed method not only optimizes routes but also significantly enhances the overall efficiency and robustness of supply chain networks. This research provides an efficient route optimization framework applicable to complex supply chain management and other scheduling fields, offering new insights and technical solutions for future research and applications.

With the rapid development of digital economy, the scale and complexity of public data assets are increasing, and how to efficiently and accurately mine and utilize these data has become a key issue. Existing methods are insufficient in dealing with multimodal data fusion and temporal feature capture, which is difficult to meet the needs of complex public decision-making. To this end, this paper proposes ReLSTM-Opt, a hybrid deep learning model that fuses ResNet and LSTM, to achieve automatic hyperparameter adjustment through Bayesian optimization, and to enhance the model's integrated learning ability for spatial and temporal features. Experimental results show that ReLSTM-Opt significantly outperforms traditional methods in classification accuracy and temporal prediction performance on multiple public datasets, demonstrating good generalization ability and stability. The model provides strong technical support for intelligent analysis and data-driven decision-making of public data assets and promotes the efficient utilization of public data resources in the digital economy.