Multi-type traffic sensor location problem for origin–destination estimation considering spatiotemporal correlation and sensor failure

The challenge of locating traffic sensors for origin–destination (OD) demand estimation involves determining the optimal number, location, and type of sensors needed to accurately gauge the traffic flow between OD pairs within an urban road network. However, analysis of traffic big data reveals that OD demand exhibits spatiotemporal correlation, indicating interconnected traffic flows between various OD pairs during different time periods. Additionally, it is important to acknowledge that sensors are prone to failure, with failure patterns varying over time and among different sensor types. In this paper, we propose new robust models for multi-type sensor location that consider the spatiotemporal correlation of OD demands and sensor failures. We estimate the mean and covariance of OD demands for different OD pairs cross various time periods to address spatiotemporal correlation, while establishing upper bounds on estimation errors to mitigate reliance on true OD demands. Furthermore, based on its bathtub curve characteristic, we employ the Weibull distribution to characterize the time-varying sensor failure rate, and utilize a nonlinear least squares method to estimate the parameters of failure rate function. Subsequently, we establish combination location models for count and automatic vehicle identification (AVI) sensors in scenarios with either no existing sensors or pre-existing sensors respectively, with the goal of minimizing upper bounds on error in both OD mean and covariance estimation while concurrently considering the spatiotemporal correlation of OD demand and time-varying sensor failure rate. Given the bi-objective nature and non-linear characteristics of these proposed models, we develop a non-dominated sorting genetic algorithm as a solution approach. Finally, we present numerical examples demonstrate how factors such as the spatiotemporal correlation of OD demands, time-varying sensor failure rate, sensor type, and budget constraints influence the sensor location strategy. The results confirm that our proposed models and algorithms deliver enhanced precision and faster convergence speed.