Time-to-Green predictions for fully-actuated signal control systems with supervised learning

TL;DR

This work tackles the problem of forecasting the next red-phase duration ($T2G$) at fully-actuated intersections to enhance SPaT messaging. It proposes a generic Time-to-Green (T2G) framework that ingests loop detector and signal data, performs cycle-based feature engineering, and evaluates Linear Regression, Random Forest, and LSTM models, with RF delivering the best overall performance on a Zurich dataset (MAE ≈ $2.27$ s, EH ≈ $59.7\%$, NM ≈ $79.5\%$). The study demonstrates substantial improvements over a naive baseline and LR, while highlighting challenges in high-variance, transit-priority signals (signals 11–12) and the impact of post-prediction detections, suggesting online or meta-learning for real-time adaptation. The results imply that accurate T2G predictions can meaningfully improve speed-advisory systems and traffic flow homogeneity, though practical deployment will require handling dynamic re-optimizations and environment shifts.

Abstract

Recently, efforts have been made to standardize signal phase and timing (SPaT) messages. These messages contain signal phase timings of all signalized intersection approaches. This information can thus be used for efficient motion planning, resulting in more homogeneous traffic flows and uniform speed profiles. Despite efforts to provide robust predictions for semi-actuated signal control systems, predicting signal phase timings for fully-actuated controls remains challenging. This paper proposes a time series prediction framework using aggregated traffic signal and loop detector data. We utilize state-of-the-art machine learning models to predict future signal phases' duration. The performance of a Linear Regression (LR), a Random Forest (RF), and a Long-Short-Term-Memory (LSTM) neural network are assessed against a naive baseline model. Results based on an empirical data set from a fully-actuated signal control system in Zurich, Switzerland, show that machine learning models outperform conventional prediction methods. Furthermore, tree-based decision models such as the RF perform best with an accuracy that meets requirements for practical applications.

Figures (12)

• Figure 1: T2G framework. The input data includes by LD and signal data from the traffic operator.
• Figure 2: Feature computation of red and green time based on $s_i(k, c_{i,n})$.
• Figure 3: Feature computation based on traffic signal and detector data $s_i(k, c_{i,n})$ and $d_j(k, c_{i,n})$.
• Figure 4: Test intersection in the city center of Zurich, Switzerland.
• Figure 5: Distributions of red and green times for all signals 1--10. Additionally, the distributions for red, green, and cycle time of signal 4 are shown with corresponding standard deviation thresholds +SD/2 and -SD/2.