基于神经网络算法的三维模型预取系统

3D MODEL PREFETCHING SYSTEM BASED ON NEURAL NETWORK

三维漫游已经成为十分普遍的远程游览或监控的一种方式。三维漫游过程中,系统往往需要动态调用数据并生成三维模型,这会影响系统运行的效率,降低用户体验。针对这一问题,提出一种基于神经网络的智能预取系统。历史访问日志中,模型的特征与被访问频率之间的映射被提取出来,并用以训练系统。系统可以根据模型特征判断哪些模型更可能被用户请求到,并把它们提前存入内存中。通过某三维变电站实时监控与装配管理系统对该预取系统进行验证。结果表明,通过短时间(一天)对系统的训练,即可做出长期(三个月)的高准确率(90%)的预测,有效提高了系统运行的效率。

Three-dimensional roaming technology has become a very popular way of remote tour or monitoring. In 3D roaming, the system usually needs to dynamically call the data and generate 3D model, and this may impact the efficiency of system operation and reduce user＇s experience. In light of this issue, we propose a neural network-based intelligent prefetching system~ The maps between model features and model access frequency are sorted out from the historical access logs and are used to train the system. The system can determine which model is more likely to be requested by users according to model features, and then put them into cache in advance. A certain 3D substation real- time monitoring and assembly management system is used to validate the prefetching system. Result shows that the highly accurate （90%） long-term （3 months） predictions can be made by training the system for a short period （1 day）. It effectively improves the efficiency of the system operation.