Learning When to See for Long-term Traffic Data Collection on Power-constrained Devices

TL;DR

This work tackles long-term traffic data collection on power-constrained devices by scheduling observations over a horizon of $K$ time steps using a predictor–controller–estimator pipeline. The predictor (RNN encoder–decoder) forecasts future profiles, the DRQN controller selects when to observe next, and the estimator reconstructs the full data from sparse samples. On PeMS-derived occupancy data, the approach yields about a 10% improvement in estimation accuracy over a Kalman-filter baseline with uniform sampling and extends device lifetime under a fixed observation budget. The results highlight the advantages of active, data-driven sampling and post-estimation refinement for energy-efficient traffic sensing, with code to be released.

Abstract

Collecting traffic data is crucial for transportation systems and urban planning, and is often more desirable through easy-to-deploy but power-constrained devices, due to the unavailability or high cost of power and network infrastructure. The limited power means an inevitable trade-off between data collection duration and accuracy/resolution. We introduce a novel learning-based framework that strategically decides observation timings for battery-powered devices and reconstructs the full data stream from sparsely sampled observations, resulting in minimal performance loss and a significantly prolonged system lifetime. Our framework comprises a predictor, a controller, and an estimator. The predictor utilizes historical data to forecast future trends within a fixed time horizon. The controller uses the forecasts to determine the next optimal timing for data collection. Finally, the estimator reconstructs the complete data profile from the sampled observations. We evaluate the performance of the proposed method on PeMS data by an RNN (Recurrent Neural Network) predictor and estimator, and a DRQN (Deep Recurrent Q-Network) controller, and compare it against the baseline that uses Kalman filter and uniform sampling. The results indicate that our method outperforms the baseline, primarily due to the inclusion of more representative data points in the profile, resulting in an overall 10\% improvement in estimation accuracy. Source code will be publicly available.

Figures (5)

• Figure 1: Workflow of the proposed two-stage framework.
• Figure 2: Prediction of AR(4)$_{kal}$ and LSTM$_{pred}$ following Uniform observation policy. LSTM$_{pred}$ can work with historical data that will fail AR(4)$_{kal}$.
• Figure 3: The proposed LSTM$_{pred}$+DRQN+LSTM$_{est}$ configuration predicts the start and end of the peak hours. It can generate smooth predictions even without new observations.
• Figure 4: Observations are strategically taken either when there is a significant prediction error or at representative points, such as peaks.
• Figure 5: The distribution of observations allocated in each day by the DRQN controller. The y-axis represents the ratio of the number of observations assigned at a specific time to the total number of test instances. Each day starts at 0:00 and ends at 23:00.