Strategic Sensor Placement in Expansive Highway Networks: A Novel Framework for Maximizing Information Gain

Abstract

Traffic sensors play a pivotal role in monitoring and assessing network-wide traffic conditions. However, the substantial costs associated with deploying an extensive sensor network across real-world highway systems can often prove prohibitive. Thus, the strategic selection of optimal sensor locations within budget and resource constraints becomes imperative, leading to the well-known Traffic Sensor Location Problem (TSLP). In this study, we introduce a novel framework to address the TSLP for large-scale highway networks, focusing on maximizing information gain in a joint vector space that comprehensively captures both network topology and segment-level features. To solve this optimization problem, we devised a genetic algorithm (GA) with penalty handling. Additionally, we developed a physics-guided random walk algorithm, which not only significantly reduces the search space but offers remarkable flexibility in striking a practical balance between computational load and the confidence of achieving global optimality. For illustration purposes, the proposed framework was applied to the Savannah highway network in Georgia. The results from our GA method align well with those from exhaustive research, but with significantly reduced computational time. By leveraging information theory and maximizing information gain in a low-dimensional vector space, the proposed framework permits parallel, scalable computation and offers considerable potential in the strategic planning and deployment of various sensors for expansive, real-world highway networks.

Keywords

• Traffic Sensor Location Problem
• topological embedding
• information theory
• Kullback–Leibler divergence
• genetic algorithm
• physics-guided random walk