Demand Calibration of Multimodal Microscopic Traffic Simulation using Weighted Discrete SPSA

Abstract

This paper presents a stochastic approximation framework to solve a generalized problem of off-line calibration of demand for a multimodal microscopic (or mesoscopic) network simulation using aggregated sensor data. A key feature of this problem is that demand, although typically treated as a continuous variable is in fact discrete, particularly in the context of agent-based simulation. To address this, we first use a discrete version of the weighted simultaneous perturbation stochastic approximation (W-DSPSA) algorithm for minimizing a generalized least squares (GLS) objective (that measures the distance between simulated and observed measurements), defined over discrete sets. The algorithm computes the gradient at each iteration using a symmetric discrete perturbation of the calibration parameters and a multimodal weight matrix to improve the accuracy of the gradient estimate. The W-DSPSA algorithm is then applied to the large-scale calibration of multimodal origin–destination (OD) flows (including private vehicle (PVT) and public transit (PT) trips) in a microscopic network simulation model of Singapore. The results indicate that an acceptable margin of error on the vehicle loop count (VLC) and bus passenger count (BPC) are achieved at convergence with an improvement of 60%~80% in root mean squared errors. Lastly, we validate the calibration results with observed travel times on the network. Statistical comparison shows good agreements on both point-to-point travel time (PTT) and public buses’ stop-to-stop ride-time (SRT) with the field observations.

Evaluating Automated Demand Responsive Transit Using Microsimulation

“…SimMobility is a high-fidelity agent-based microsimulation laboratory, which coherently integrates individuals' choices and travel behavior emphasizing the principle of activitybased accessibility across different timescales represented by three core models: the Long-term model that captures yearto-year evolution of land use and agents' socio-economic activities [35], the Mid-term model, which predicts day-today individual activity patterns and travel behavior, and network dynamics at a mesoscopic level [36], and the Short-term model that simulates agents' movements at the microscopic granularity [37], [38]. The overall structure and information flow across models are depicted in Figure 1.…”

Composite Surrogate for Likelihood-Free Bayesian Optimisation in High-Dimensional Settings of Activity-Based Transportation Models

“…Calibration of Preday ABM has been performed in several studies, whereby parameters are estimated by Simultaneous perturbation stochastic approximation (SPSA) method with its variants [28,29,31]. However, all studies considered reduction of the dimensionality of the parameter space either by sensitivity analysis (SA) [34] or principal component analysis (PCA) [19].…”

“…Application of the Preday ABM in simulating various environments requires systematic adjustments or calibration of a large set of parameters (further referred to as ABM parameters), in order to align the associated outputs more closely to the observed values or true output statistics. For that purpose, various optimisation methods are adopted, including primarily gradient-free metaheuristics [28,29,31]. However, Bayesian inference with the recent developments provides a valuable analytical approach for the calibration process [35,36], a great advantage of which is the elimination of necessity to simulate a large sample set in finding the global optima [14,18,20,36].…”

Integrated agent-based microsimulation framework for examining impacts of mobility-oriented policies

“…The objective function is the ordinary least square (OLS) type that minimizes the difference between the observed traffic volume and predicted traffic volumes (only cars) on links where data is available. The algorithm used is SPSA (simultaneous perturbation stochastic algorithm) to find out the new set of parameters to be used for the next calibration iteration as used in [Oh et al 2019]. It should be noted that within one calibration simulation there are 250 MATSIM simulation runs, and therefore, the process is quite time-consuming.…”