How fast is the traffic?
Problem being addressed
Traffic speed is usually difficult to predict since traffic tends to move in waves, and can be interrupted by street lights, tolls and intersections. How can traffic data be used to predict traffic speed?
A deep belief network model is proposed for the estimation of short term traffic speeds. The model is first trained through greedy unsupervised learning, and then this is fine tuned by labelled traffic data obtained from a main roadway in Beijing. The model is used to predict the traffic speed at 2 minute, 10 minute and 30 minute horizons. The results are alos compared to other statistical predictors such as the ARIMA method and bacpropagated neural networks (BPNN). It is found that the deep belief network is excellent at making short term predictions, but is less accurate over the 30 minute horizon prediction, but still better than ARIMA and BPNN.
Advantages of this solution
Traffic data is important for the understanding of city management and congestion control. These methods can be used to map out and model traffic across a city, and assist with easing congestion.
Solution originally applied in these industries
Transport & Logistics Industry
Possible New Application of the Work
Energy usage has a traffic component to it in the sense that energy systems work under different loads at different times. Perhaps this research could be used in the energy sector to build load anticipation methods.
In the hospitality industry traffic is the equivalent of people. A similar methodology could be used to predict the load on hospitality and services during peak times and holiday seasons. This could allow for better planning and staffing.
There are some similarities between traffic flow patterns and sales rates. A possibility is that businesses could model their sales as traffic flow systems, and adjust their supply chains to match the outflow via sales. This would lead to cost savings when it comes to inventory management.
Source DOI: #############
Source URL: #############