AI-optimized BPNN model for port safety risk prediction and management

In the field of international port safety management, the traditional Backpropagation Neural Network (BPNN) model is confronted with bottlenecks including limited data processing capability and low optimization efficiency. This study proposes an intelligent prediction framework for port safety risks that integrates the Bald Eagle Search (BES) algorithm, Convolutional Neural Network (CNN), and Fuzzy Logic System (FLS), aiming to improve the scientific nature of risk early warning and management efficiency. The framework employs CNN for feature extraction, uses the BES algorithm to conduct global parameter optimization for BPNN, and introduces FLS to handle uncertainty and fuzzy information. The results show that the proposed ensemble model achieves outstanding performance in the three-classification task of port safety risks, with the Root Mean Square Error (RMSE) reduced to 0.012, the accuracy improved to 98.5%, and the Macro-F1 score reaching 0.982. Furthermore, the model inference latency is only 12.5 milliseconds, which fully demonstrates its application potential in real-time monitoring scenarios. Compared with traditional optimization algorithms, the proposed model has advantages in computational performance, and provides scientific decision support for port administrators by lowering the false alarm rate and enhancing emergency response speed. This study opens up a new approach for the practical application of Artificial Intelligence (AI) in the port safety domain, and carries important practical value for promoting the intelligent and precise transformation of port safety management.