基于深度过程神经网络的车速预测
Vehicle speed prediction based on deep process neural network
现代智能交通系统中具有来源广泛、结构复杂的交通数据，对交通数据进行高效分析并合理利用具有现实意义。传统的神经网络可以对交通数据进行预测分析，但是面对时序输入更有效的方法是过程神经网络。本文采用的深度过程神经网络是一种在传统过程神经网络上优化得出的模型，该模型曾在交通流量预测上取得了较好的结果，其中后积分深度过程神经网络的效果最好。本文类比了关于交通量的研究，将日期属性看作输入数据的属性，将具体时间看做时间序列属性，希望得到包含了人类周期性运动对车速影响的规律的预测结果，在后积分深度过程神经网络上采用了自编码算法将网络简化为先积分深度过程神经网络，最终得出预测结果。深度过程神经网络虽然精度更高，时间特性更好，在面对大量数据预测的情况下效果非常好，但是结构稍微复杂对内存要求比较高，在少量数据且受硬件限制的条件下，结果并不是特别理想。
Intelligent Transport System has traffic data of complex structure from wide sources, which makes the analysis of traffic data difficult. It's important to make efficient and rational use of rich resources. Traditional neural networks can predict traffic data ,but process neural networks performance more efficiently in handling time series data. This paper use deep process neural network as an optimization model derived in the traditional process neural network, which performance well in traffic flow forecasting. Among all the deep neural network accumulation last neural network does best. Accumulation last neural network use a auto-encoder to simplify the deep process network to accumulation first neural network. Although deep process neural network has higher accuracy ,better time characteristics, and performance really well to large scale of data ,the structure is more complex and require much more memory. Under the conditions of a small amount of data and hardware limitations, the results are not particularly desirable
胡怡红、袁月
公路运输工程自动化技术、自动化技术设备计算技术、计算机技术
深度过程神经网络，自编码器，深度学习
eep process neural networks auto-encoder deep learning
胡怡红,袁月.基于深度过程神经网络的车速预测[EB/OL].(2014-12-01)[2025-08-18].http://www.paper.edu.cn/releasepaper/content/201412-1.点此复制
评论