摘要
面对建筑行业安全事故频发的问题,开发创新的培训工具变得尤为重要。设计旨在构建一个基于Unity3D和神经网络的建筑安全虚拟仿真系统,以提高建筑安全培训的效率和效果。通过利用Unity3D强大的图形渲染能力创建仿真环境,并结合神经网络进行行为分析和风险评估。结果表明,在此框架下的仿真系统经过1400个训练周期后,损失值稳定在约0.043的水平,随后的训练并未显著改善损失值,表明模型逐渐收敛。结果显示出所设计的仿真系统能够有效达到训练饱和,为建筑安全培训提供了一种稳定且可靠的虚拟环境。
In the face of frequent safety accidents in the construction industry,developing innovative training tools has become particularly important.The design aims to build a building safety virtual simulation system based on Unity3D and neural networks to improve the efficiency and effectiveness of building safety training.By utilizing the powerful graphics rendering capabilities of Unity3D to create a simulation environment,and combining it with neural networks for behavior analysis and risk assessment.The results showed that after 1400 training cycles,the simulation system under this framework had a stable loss value of about 0.043,and subsequent training did not significantly improve the loss value,indicating model convergence.The conclusion shows that the designed simulation system can effectively achieve training saturation,providing a stable and reliable virtual environment for building safety training.
作者
李超
张馨予
陈秋帆
罗国辉
LI Chao;ZHANG Xinyu;CHEN Qiufan;LUO Guohui(Guangzhou Huali College,Guangzhou 511325,China)
出处
《自动化与仪器仪表》
2024年第10期137-140,144,共5页
Automation & Instrumentation
基金
广东省普通高校青年创新人才项目《“双碳”时代下岭南装配式新农居设计路径研究》(2022WQNCX128)
广东省高等教育学会“十四五”规划2023年度高等教育研究课题《建筑评论视角下应用型高校建筑学教学模式与方法研究》(23GQN66)。
