Simulation Framework Reconstructs Full Weekly Travel Schedules From Single-Day Survey Data

A computational modeling framework can accurately simulate complete weekly activity and travel schedules for urban populations by relying primarily on single-day travel diaries combined with weekly participation frequencies, according to an investigation published in Computational Urban Science. The simulation architecture, designated as the iTLE Week-Sim module, was engineered by researchers to overcome the longstanding limitation in transportation planning known as the typical-day assumption. Traditional travel demand models have predominantly captured mobility patterns for an isolated 24-hour weekday, failing to record how individual schedules fluctuate across an entire week. By coupling a first-order Markov Chain Monte Carlo activity sequence generator with a multi-objective optimization procedure, the investigators demonstrated that synthetic multi-day schedules can reproduce empirical weekly behavioral dynamics without requiring resource-intensive longitudinal surveys.

Transportation planning has progressively evolved from aggregate flow models that evaluate network capacity to disaggregate activity-based models that analyze travel as a derived demand arising from daily obligations and personal choices. While modern activity-based models capture interactions across individual activity tours, their real-world application has been constrained by the scarcity of multi-day household survey data. Longitudinal travel panels require significant financial investments and frequently suffer from participant fatigue, attrition, nonresponse bias, underreporting, and seasonal distortion between survey waves. Consequently, operational tools capable of simulating an entire week have remained extremely rare, with few platforms such as mobiTopp and MATSim attempting multi-day representations. Even within those models, data acquisition barriers have prevented broad regional adoption, prompting the need for surrogate computational frameworks that extract longitudinal properties from standard cross-sectional surveys.

To build and calibrate the simulation framework, the researchers utilized empirical records from the Halifax Travel Activity Survey, conducted across the Halifax Regional Municipality in Canada between July 2022 and March 2023. The sampling initiative engaged 3,731 households and 5,095 individuals, collecting comprehensive 24-hour travel diaries spanning from 3:00 AM to 3:00 AM across separate weekdays. These diaries encompassed 710 logs recorded on Mondays, 890 on Tuesdays, 754 on Wednesdays, 609 on Thursdays, and 450 on Fridays, alongside self-reported weekly frequencies for key daily activities measured on a five-point Likert scale. A comparative evaluation against the 2021 Canadian Census verified that the survey sample closely reflected the population across household attributes, individual demographic characteristics, and commuting metrics, establishing a representative empirical foundation without requiring post-stratification weighting.

The module expands the Integrated Transportation, Land Use, and Energy simulation system, an established microsimulation environment developed by the Dalhousie Transportation Collaboratory. Within this architecture, a long-term decision simulator models annual sociodemographic changes, while a short-term decision simulator schedules daily activities, mode choices, and vehicle allocations for synthetic agents. In the standard implementation, the short-term module generates a baseline activity chain for a single weekday by executing a first-order Markov chain, wherein the transition to each consecutive activity state depends exclusively on the immediate prior state. The iTLE Week-Sim module expands this foundation by running the stochastic sequencing process across five consecutive weekdays, explicitly updating decision variables to account for daily schedules crossing midnight, individualized tour start times, and daily shared vehicle allocations.

Because repeating an uncalibrated single-day Markov process across five days merely duplicates static proportions without reflecting day-of-week dynamics, the framework incorporates an optimization algorithm inspired by genetic algorithm principles. The system adjusts transition probability modification factors through an iterative calibration routine that minimizes the absolute deviation between observed and simulated activity proportions across all five weekdays. The model incorporates behavioral constraints derived from survey questions specifying how many times per week individuals engage in distinct activities, alongside the empirical likelihood of performing an activity on a specific weekday. If a generated schedule violates these constraint conditions, the algorithm resimulates the chain up to six iterations, preventing excessive computational loops while driving the system toward empirical alignment.

The calibration specifically targeted three primary activity categories: in-home work, in-person work, and shopping, adjusting their transition probability matrices across multiple simulation iterations. The multi-objective optimization was set to terminate when the total weekly objective value dropped below 0.20, representing an average absolute deviation of approximately 0.04 per weekday between simulated and empirical proportions. Calibration experiments showed rapid error convergence across consecutive parameter configurations, reducing deviations from initial values exceeding 0.34 for in-home work and 0.42 for in-person work down to 0.190 for in-home work, 0.164 for in-person work, and 0.152 for shopping. Model validation against independent activities that were not adjusted during optimization—specifically eating out and recreation—achieved close empirical agreement, yielding absolute validation deviations of 0.043 and 0.023, respectively.

To evaluate performance at an urban scale, the Week-Sim module executed full weekly microsimulations for a representative 10 percent synthetic sample of the Halifax metropolitan area, tracking 42,230 agents across five weekdays. Over this five-day simulation period, the model generated and scheduled a total of 744,290 distinct activities encompassing seven out-of-home types and four in-home categories, excluding passive baseline at-home periods. Computational profiling indicated that executing an optimized weekly simulation configuration required approximately 200 minutes per run. This sampling approach allowed the platform to maintain computational feasibility while retaining disaggregate demographic heterogeneity, tracking how individual age, living arrangements, and employment statuses govern weekly schedule assembly.

The simulated outputs revealed prominent intra-week rhythms and structural contrasts between physical and remote work arrangements. In-person work participation consistently concentrated into sharp morning peaks between 6:00 AM and 9:00 AM across all weekdays. In contrast, in-home work was dispersed broadly throughout the day, featuring shorter overall episode durations, greater variability in start times, and increased schedule adaptability that permitted individuals to avoid peak-period traffic congestion. Furthermore, the simulation captured a pronounced day-of-the-week effect toward the conclusion of the workweek, marked by a systematic reduction in work duration on Fridays as agents shifted time budgets toward discretionary and leisure pursuits, consistent with observations reported in recent independent empirical studies.

Analysis of demographic cohorts within the simulated population demonstrated distinct life-cycle and employment patterns in time allocation. In-home durations exceeded out-of-home durations across all population segments, but agents aged 25 to 50 exhibited the highest out-of-home engagement and lowest in-home duration, reflecting labor market commitments. Individuals under 25 and over 50 displayed greater residential orientation, corresponding to educational calendars and reduced workforce participation. Examining activity frequency across daily windows showed that midday periods between 10:00 AM and 4:00 PM concentrated the highest activity density across full-time workers, part-time workers, and students, whereas evening windows from 4:00 PM to 3:00 AM exhibited elevated discretionary activity, with part-time workers and students showing substantially higher day-to-day variability than full-time employees.

The authors noted specific operational limitations within the current surrogate modeling structure, pointing out that activity sequences are generated dynamically rather than through comprehensive multi-day utility maximization. Because the formulation relies on empirical frequency distributions and heuristic probability modification factors, it lacks an explicit long-range weekly planning mechanism that links activity choices directly to end-of-day feedback loops. The researchers emphasized that integrating profile-specific variability indices, developing day-to-day feedback structures, and conducting subsequent validation against multi-day panel surveys represent critical priorities for advancing the behavioral fidelity of computational urban simulation systems.

Transition matrices in the underlying simulation are constructed across 14 distinct population segments categorized by employment status and living arrangements, with sparsely populated groups merged to ensure statistical reliability. Daily activity sequencing begins at an initial entry state representing the home in the morning, progressing through state-specific transition probabilities until an absorbing exit state terminates the daily sequence when an individual remains home. Rather than incorporating travel time into activity spans, durations are sampled from continuous probability distributions conditioned on an agent’s total daily activity count and rounded to the nearest minute. Activity start times are subsequently assigned using joint probability distributions of activity duration and start timing.

Across 46 experimental simulation runs, the calibration procedure revealed the computational necessity of establishing a 0.20 convergence threshold. While early parameter updates achieved rapid error reduction, progress leveled off into marginal oscillations near the target threshold. Selecting stricter thresholds below 0.15 exacerbated the computational burden and triggered model instability from competing multi-objective trade-offs, whereas looser thresholds above 0.25 produced insufficient calibration. Non-calibrated validation activities, such as eating out and recreation, remained steady throughout testing, displaying error bounds between 0.02 and 0.06 deviation from observed proportions.

Disaggregate inspection of synthetic agents demonstrated how the micro-level scheduling captures realistic behavioral variability across diverse demographic profiles. In one modeled profile, a 52-year-old non-working female living with a partner concentrated activity hours into shopping, personal business, and recreation. In another case, an employed agent residing with a roommate performed workplace tasks on four of five weekdays, exhibiting noticeable fluctuations in daily start and end times along with a truncated shift on Friday. These individual itineraries confirm that the simulation reproduces personal schedule adjustments alongside aggregate travel trends, illustrating the framework’s disaggregate resolution.

Subject of Research: Computational microsimulation and optimization of weekly urban travel activities from single-day survey data

Article Title: A hybrid optimization and microsimulation framework for simulating weekly travel activities

Article References: Alam, M. J., Bhuiyan, M. R. H., Mandapati, V. V. R. R., & Habib, M. A. (2026). A hybrid optimization and microsimulation framework for simulating weekly travel activities. Computational Urban Science, 6(1), Article 24. https://doi.org/10.1007/s43762-026-00259-4

Image Credits: AI Generated

DOI: 10.1007/s43762-026-00259-4

Keywords: travel demand modeling, activity-based modeling, microsimulation, urban mobility, computational urban science, Markov Chain Monte Carlo, synthetic population, travel behavior, telecommuting, urban planning, Halifax Travel Activity Survey, optimization framework, schedule generation, daily rhythms, transportation engineering, travel diaries, tour-based models