Scholarlink

<p>Aiming at the problem of insufficient real-time performance in the logistics supply chain security risk warning system due to the high computational complexity of deep learning models, this paper proposes a solution based on lightweight Convolutional Neural Network-Particle Swarm Optimization (CNN-PSO) encryption algorithm and builds a distributed security risk warning system. Depthwise-separable convolution is used to reduce the computational workload of convolutional neural network (CNN), and knowledge distillation is combined to train small models to improve inference efficiency. An adaptive particle swarm optimization algorithm is applied to dynamically adjust CNN hyperparameters, thereby improving model convergence speed and generalization ability. At the encryption level, Advanced Encryption Standard (AES) is used to quickly encrypt logistics transaction data; Elliptic Curve Cryptography (ECC) is combined to achieve key distribution; a hash index is used to optimize the storage structure to reduce encryption computational costs. A federated learning architecture is constructed, and each logistics node trains the CNN model locally. The global model is updated by encryption parameter aggregation to avoid data leakage and single-point attack risks. Experimental results show that the inference time of the optimized Depthwise-Separable Convolutional Neural Network with Knowledge Distillation model is reduced to 95&#xa0;ms, which is 62% less than the traditional CNN, and the test accuracy is increased to 97.8%. The encryption time of AES combined with ECC and hash index optimization is 40% lower than that of traditional AES, and federated learning reduces the risk of data leakage to 10%. The accuracy of CNN-PSO combined with federated learning in five types of risk detection exceeds 94%, with the highest reaching 98.3%. The proposed method improves the risk warning capability of the logistics supply chain, ensures data security, and meets the real-time requirements in large-scale logistics environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Logistics supply chain security risk warning system based on CNN-PSO encryption algorithm

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Lifan Feng,
  • Guibin Zhang,
  • Hongxia Gu,
  • Min Feng,
  • Cheng Wang

摘要

Aiming at the problem of insufficient real-time performance in the logistics supply chain security risk warning system due to the high computational complexity of deep learning models, this paper proposes a solution based on lightweight Convolutional Neural Network-Particle Swarm Optimization (CNN-PSO) encryption algorithm and builds a distributed security risk warning system. Depthwise-separable convolution is used to reduce the computational workload of convolutional neural network (CNN), and knowledge distillation is combined to train small models to improve inference efficiency. An adaptive particle swarm optimization algorithm is applied to dynamically adjust CNN hyperparameters, thereby improving model convergence speed and generalization ability. At the encryption level, Advanced Encryption Standard (AES) is used to quickly encrypt logistics transaction data; Elliptic Curve Cryptography (ECC) is combined to achieve key distribution; a hash index is used to optimize the storage structure to reduce encryption computational costs. A federated learning architecture is constructed, and each logistics node trains the CNN model locally. The global model is updated by encryption parameter aggregation to avoid data leakage and single-point attack risks. Experimental results show that the inference time of the optimized Depthwise-Separable Convolutional Neural Network with Knowledge Distillation model is reduced to 95 ms, which is 62% less than the traditional CNN, and the test accuracy is increased to 97.8%. The encryption time of AES combined with ECC and hash index optimization is 40% lower than that of traditional AES, and federated learning reduces the risk of data leakage to 10%. The accuracy of CNN-PSO combined with federated learning in five types of risk detection exceeds 94%, with the highest reaching 98.3%. The proposed method improves the risk warning capability of the logistics supply chain, ensures data security, and meets the real-time requirements in large-scale logistics environments.