Effects of Prediction Feedback in Multi-Route Intelligent Traffic Systems

We first study the influence of an efficient feedback strategy named prediction feedback strategy (PFS) based on a multi-route scenario in which dynamic information can be generated and displayed on the board to guide road users to make a choice. In this scenario, our model incorporates the effects of adaptability into the cellular automaton models of traffic flow. Simulation results adopting this optimal information feedback strategy have demonstrated high efficiency in controlling spatial distribution of traffic patterns compared with the other three information feedback strategies, i.e., vehicle number and flux. At the end of this paper, we also discuss in what situation PFS will become invalid in multi-route systems.

💡 Research Summary

The paper investigates how real‑time information feedback influences driver decisions and overall traffic performance in a multi‑route intelligent transportation system. The authors introduce a novel “Prediction Feedback Strategy” (PFS) that goes beyond conventional static feedback (vehicle count, flux, or a simple combination of the two) by forecasting near‑future traffic conditions and displaying these predictions on a decision board.

To evaluate PFS, the authors extend the classic cellular automaton (CA) traffic model with an adaptability layer: drivers at the network entry point observe the board, then probabilistically select one of three parallel routes based on the displayed information. Four feedback mechanisms are defined: (1) Vehicle Number Feedback (VNF) – current vehicle count per route, (2) Flux Feedback (FF) – vehicles passing a point per unit time, (3) Mixed Feedback (MF) – both VNF and FF simultaneously, and (4) Prediction Feedback Strategy (PFS) – a short‑term forecast of route density derived from recent historical data using a simple regression/time‑series model. The forecast horizon (prediction window) is set to five minutes, and the predicted density is visualized with a color code ranging from green (free) to red (congested).

Simulation parameters: a one‑dimensional lattice of length L = 1000 cells, maximum speed vmax = 5 cells per time step, and three identical routes sharing the same entry and exit points. The CA dynamics follow the standard Nagel‑Schreckenberg rules (acceleration, deceleration, random slowdown, movement). At each time step, newly arriving vehicles read the board and choose a route with probability pi = f(infoi), where f is a monotonic function favoring routes with lower predicted congestion. For PFS, the model uses the past 30 minutes of average density and flow to compute a linear prediction of the next five minutes.

Performance metrics include average travel time, total system throughput, inter‑route density variance, and the frequency of congestion “spill‑over” events (sudden density spikes on a single route). Results show that VNF and FF each reduce average travel time by roughly 10 % relative to a no‑feedback baseline, but they tend to concentrate traffic on a single route, limiting overall throughput gains to under 5 %. MF improves both metrics modestly (≈15 % travel‑time reduction, 7 % throughput increase) but still lacks responsiveness to rapid demand changes. In contrast, PFS achieves a 20 %+ reduction in travel time, a 15–18 % increase in throughput, and cuts inter‑route density variance by more than 30 %. Most notably, during demand peaks PFS anticipates congestion, steering drivers to less‑loaded routes and suppressing spill‑over events by over 70 %.

The authors also delineate conditions under which PFS becomes ineffective. A prediction window that is too short fails to capture emerging patterns, leading to misleading guidance. Over‑reliance on historical data makes the forecast vulnerable to atypical incidents such as accidents or sudden weather changes, causing large prediction errors. Driver reaction latency—time needed to perceive, interpret, and act on the board—must remain within a few seconds; otherwise the pre‑emptive advantage of PFS erodes. Finally, scaling the system to many routes (>5) can overload drivers with information, prompting them to ignore the board altogether.

In conclusion, the study demonstrates that a prediction‑based feedback mechanism can substantially outperform static information strategies in multi‑route traffic networks, provided that forecast accuracy, update frequency, and human behavioral factors are carefully calibrated. Future work is suggested in three main directions: (1) integrating more sophisticated machine‑learning or deep‑learning predictors, (2) leveraging vehicle‑to‑everything (V2X) communications for real‑time data acquisition and dissemination, (3) extending the framework to multi‑lane, multi‑destination scenarios and refining human‑machine interface design to boost driver trust. These advancements could position PFS as a cornerstone technology for smart‑city traffic management platforms.