神经网络测试在自主船舶中的应用与未来展望  

Application and Future Prospects of Neural Network Testing in Autonomous Ships

作　　者：张永洋 卢煜腾 伍江华[2] ZHANG Yongyang;LU Yuteng;WU Jianghua(Naval Equipment Department Project Management Center,Beijing 100071;China Ship Research and Development Academy,Beijing 100101)

机构地区：[1]海军装备部装备项目管理中心,北京100071 [2]中国舰船研究院,北京100101

出　　处：《舰船电子工程》2024年第8期22-26,共5页Ship Electronic Engineering

摘　　要：神经网络在自动驾驶、医疗诊断等安全攸关领域有着广泛应用,但其复杂性和黑箱特性带来的可靠性和安全性问题也不容忽视。神经网络测试是应对这些挑战的关键手段。论文探讨了神经网络测试的重要性,梳理了主要的测试方法,并分析了当前面临的关键挑战及未来的发展方向。同时,论文结合自主船舶这一具体应用场景,深入讨论了深度学习在自主船舶导航和防碰撞中的应用,特别是系统智能和分布式智能策略。论文提出了一种综合方法,通过将神经网络测试与自主船舶技术相结合,以保障自主船舶系统的安全性和可靠性。通过对抗性测试、模糊测试、覆盖率测试和变异测试等方法的应用,提升自主船舶在复杂海洋环境中的鲁棒性和适应性,以期为未来智能交通系统的发展奠定坚实基础。Neural networks have widespread applications in safety-critical domains such as autonomous driving and medical diagnosis.However,the complexity and black-box nature of these systems present significant challenges in terms of reliability and safety.Neural network testing is a crucial means to address these challenges.This paper explores the importance of neural network testing,outlines the main testing methods,and analyzes the key challenges and future directions in this field.Additionally,this pa-per delves into the specific application scenario of autonomous ships,discussing the use of deep learning in autonomous ship naviga-tion and collision avoidance,particularly focusing on system intelligence and distributed intelligence strategies.This paper proposes a comprehensive approach by integrating neural network testing with autonomous ship technology to ensure the safety and reliability of autonomous ship systems.Through the application of adversarial testing,fuzz testing,coverage testing,and mutation testing,the robustness and adaptability of autonomous ships in complex marine environments are enhanced,laying a solid foundation for the fu-ture development of intelligent transportation systems.

关 键 词：神经网络安全 测试 自主船舶 可解释人工智能 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]