Source-linked AI summary
A High-Resolution Synthetic EV Charging Dataset for Cold-Climate Distribution Grid Impact Analysis: Trondheim, Norway (2020-2030)
Hanieh Taraghi Nazloo, Petr Musilek
TL;DR
Distribution-grid studies need realistic, long-horizon EV charging data that captures cold-climate, behavioral, and contextual variation. This paper constructs a Trondheim synthetic dataset by combining empirical records with CTGAN–KDE generation, scenario scaling, and physical feasibility correction, yielding a validated medium-adoption profile for 2020–2030. Its principal limitation is that long-term changes in vehicle technology, V2G, tariffs, and routing behavior are not explicitly modeled, while secondary network constraints remain downstream requirements.
Problem
Distribution-grid analysis requires high-fidelity EV charging data that represents temporal behavior, user heterogeneity, seasonal effects, and cold-climate demand dependencies.
Method
The dataset combines empirical charging sessions with calendar, weather, and adoption features, hybrid CTGAN–KDE synthesis, and post-generation 7.2 kW feasibility correction.
Results
The validated dataset preserves observed charging-duration, arrival/departure, correlation, and power-demand relationships across a medium-adoption Trondheim scenario.
Takeaways & Limitations
The dataset supports cold-climate EV forecasting, distribution-grid impact and transformer-loading studies, and data-driven energy-management or reinforcement-learning models.
Takeaways & Limitations
Long-term changes in battery technology, V2G adoption, tariffs, and routing behavior are not modeled, while transformer, feeder, and phase-unbalance constraints must be applied downstream.
Abstract
from arXiv · showhide
This data article presents a high-resolution, long-term synthetic electric-vehicle (EV) charging dataset for Trondheim, Norway, spanning February 2020 to December 2030. Empirically grounded in 14 months of historical charging logs from December 2018 to January 2020, the dataset captures session-level behavioral patterns, including delivered energy, plug-in duration, connection schedules, user categorization (private vs. shared), seasonal variations, public-holiday effects, and daily ambient temperature dependencies. To model future electrification dynamics, the synthetic generation pipeline integrates historical session records, calendar and weather features from MET Norway, annual EV-adoption growth multipliers derived from Statistics Norway (SSB) registration trajectories, a daily session-count model, a Conditional Tabular Generative Adversarial Network (CTGAN), seasonal Kernel Density Estimation (KDE), and post-generation physical charger-power feasibility correction. Under a standardized 7.2 kW AC charging constraint, the resulting medium EV-adoption scenario dataset contains 76,993 hourly charging-activity records. The hourly profile is activity-based rather than a complete continuous hourly time series; hours with no allocated EV charging energy are not included. The records provide total hourly charging energy, equivalent average charging power, active session counts, private/shared user load breakdowns, ambient temperature features, and calendar indicators. The dataset provides a validated cold-climate benchmark for distribution-grid impact assessment, transformer-loading analysis, EV charging-demand forecasting, charger-capacity planning, energy-management optimization, and the development of data-driven smart-charging control strategies.
Value of the Data
The dataset supports long-term, cold-climate distribution-grid studies by combining behavioral fidelity, temperature dependencies, EV-growth scenarios, and physically bounded charging profiles.
- Long-Term Demand Horizons: The dataset provides an 11-year synthetic EV charging profile for Trondheim, Norway, supporting multi-year grid planning and infrastructure studies.Its horizon spans 2020–2030 and is calibrated to Trondheim.
- Behavioral & Contextual Fidelity: The hybrid CTGAN–KDE architecture preserves temporal connection patterns, user categories, seasonal dynamics, and holiday effects from empirical charging logs.These features support modeling of heterogeneous charging behavior.
- Cold-Climate Parameterization: Daily temperature dependencies and cold-weather scaling factors provide a specialized benchmark for cold-climate load forecasting and thermal-impact analyses.
- Multi-Scenario Penetration Modeling: Parametrized EV-growth trajectories model escalating fleet penetration and its effects on hourly demand, peak-to-average ratios, and station utilization.
- Physical Feasibility Bounding: Post-generation correction enforces a 7.2 kW per-charger limit while retaining nonlinear charging dynamics for downstream power-flow modeling.
- Multi-Scale Application Readily: Session-level and hourly aggregate formats interface with power-flow models, transformer assessments, smart-charging algorithms, and reinforcement-learning environments.
Background
EV charging datasets enable analysis of demand, session behavior, station utilization, and distribution-network interactions, but cold-climate applications require locally grounded synthetic modeling of behavioral, calendar, and weather variation.
- Background: High-resolution charging records capture plug-in times, charged energy, session duration, and identifiers for analyzing temporal demand and user heterogeneity.
- Background: Synthetic and AI-augmented datasets address the need to preserve observed charging distributions while supporting conditional generation for future scenarios.Prior work includes large-scale event synthesis and CTGAN–KDE daily charging generation.
- Background: Accurate hourly load profiles support transformer-loading, feeder-congestion, flexible-charging, and probabilistic planning studies because unmanaged charging can intensify grid stresses.
- Background: Ambient temperature affects EV energy consumption and charging demand, with reported seasonal variation up to 16% and winter monthly-consumption increases up to 30%.
- Background: A Trondheim-oriented dataset should combine local session statistics with behavioral, calendar, weather, adoption, and charger-power constraints rather than mechanically extrapolating historical sessions.Norwegian evidence identifies three distinct driver classes differing in charging frequency, location preference, and target state-of-charge.
Dataset Content
The study delivers a validated Trondheim dataset containing activity-based hourly charging records derived from corrected synthetic sessions, with temporal, load, environmental, and scenario fields.
- Dataset Content: The primary output is a validated synthetic hourly EV charging-load dataset for Trondheim under a medium-adoption scenario and 7.2 kW AC charger limit.The hourly profile is derived from a corrected session-level synthetic dataset.
- Dataset Content: 76,993 hourly records are included, with one record per hourly interval; hours without allocated EV charging energy are omitted.The profile is therefore activity-based rather than a complete continuous hourly time series.
- Dataset Content: Each record includes timestamps and calendar fields, hourly energy and power measures, active-session counts, private/shared load breakdowns, temperature, adoption scenario, charger case, and power limit.
- Dataset Content: The temporal scope covers February 2020 through December 2030, with a minor early-January 2031 extension for sessions crossing the horizon boundary.The extension preserves energy conservation between session-level and hourly datasets.
- Dataset Content: 256,826 corrected sessions produce approximately 3.26 GWh, with mean hourly energy of 42.33 kWh and peak hourly demand of 1559.87 kWh.The peak reflects regional future EV demand rather than transformer- or site-constrained capacity.
- Dataset Content: Validation confirms the 7.2 kW constraint and identical energy totals between session-level logs and hourly aggregates, supporting forecasting, grid studies, and energy-management models.
Experimental Design, Materials, and Methods
The study combines cleaned historical charging logs with calendar, weather, EV-adoption, stochastic session-count, CTGAN–KDE, and charger-feasibility procedures to generate synthetic Trondheim charging sessions. Validation compares synthetic and observed behavioral distributions, dependencies, timing patterns, and physical charging bounds.
- Data preparation: Historical session logs were cleaned before synthesis by removing incomplete, invalid, inconsistent, and duplicate records.The workflow excluded records with missing timestamps or durations, negative energy, non-positive plug-in durations, inconsistent timestamps, and duplicate identifiers.
- Feature construction: Calendar features encoded year, month, day, weekday, weekend status, season, Norwegian holidays, and special periods including Christmas/New Year, Easter, summer holidays, and bridge days.These features were constructed for both baseline and future synthetic periods to represent calendar-dependent demand shifts.
- Feature construction: Temperature observations from the MET Norway Frost archive supplied daily mean, minimum, maximum, and heating-degree-day variables for Trondheim.Baseline weather variables were obtained from Trondheim-representative stations using daily aggregated air-temperature observations.
- Scenario construction: EV-adoption scaling used Trondheim private electric-passenger-car registrations from SSB, with Trøndelag fallback data and national figures retained for comparison.The stock was treated as end-of-year registered electric-car stock, and 2019 was the baseline for subsequent growth multipliers.
- Scenario construction: Future daily session counts were sampled from a Negative Binomial model after scaling 2019 month-and-weekday charging intensity by EV growth, temperature, and holiday factors.The distribution addressed overdispersion in historical daily session counts.
- Synthetic generation and correction: The hybrid CTGAN–KDE pipeline generated synthetic sessions, then capped delivered energy at E ≤ PchargerD under the primary 7.2 kW AC case.Sessions exceeding the physical limit were adjusted rather than removed, preserving session-count structures and adoption assumptions.
- Validation: Validation found close agreement for plug-in duration, arrival and departure timing, major correlations, behavioral envelopes, and marginal densities between observed and synthetic sessions.Reported comparisons include mean plug-in durations of 11.50 h observed versus 11.69 h synthetic, mean arrivals of 16.82 h versus 16.25 h, and mean departures of 13.05 h versus 13.14 h.
- Validation: The synthetic data preserved key physical and behavioral relationships while attenuating the departure-hour versus average-power correlation.The preserved relationships include inverse duration–power dependence and bounded energy–duration distributions; the correlation changed from r = 0.42 observed to r = 0.22 synthetic.
Limitations
The dataset’s projections are constrained by a short historical baseline, simplified adoption and weather representations, omitted distribution-network limits, behavioral abstraction, and Trondheim-specific calibration.
- Baseline Historical Span: The generative models rely on charging records from December 2018 to January 2020 and do not explicitly model several long-term technology, tariff, or behavior shifts through 2030.Unmodeled factors include battery chemistry and capacity, V2G adoption, real-time tariffs, and driver routing behavior.
- Adoption Dynamics: EV adoption scaling adjusts overall daily session frequencies but does not dynamically represent individual decisions, charger-infrastructure expansion, or spatial redistribution.The formulation uses annual municipal growth multipliers rather than microlevel agent decisions.
- Grid Bounds: The 7.2 kW per-charger correction omits transformer, feeder, and phase-unbalance constraints that must be imposed in downstream studies.The unconstrained hourly loads represent unmitigated aggregated charging demand.
- Exogenous Weather Features: Weather effects use daily mean temperature and cold-weather scaling, excluding snowfall, precipitation, road-surface friction, and extreme wind conditions.
- Anonymization and Behavioral Abstraction: Synthetic trajectories preserve population-level joint distributions rather than exact driver schedules, so they represent behavioral samples rather than traceable individual profiles.
- Geographic Generalizability: Direct application beyond Trondheim requires recalibration against local baseline data because the parameterization is climate-, adoption-, and grid-context specific.
Ethics Statement
The study used secondary, fully anonymized observational and synthetic data without active human-subject experimentation, animal testing, or clinical or personally identifiable information collection.
- The study used secondary, fully anonymized observational data and synthetic generation techniques without active human-subject experimentation, animal testing, or clinical or PII collection.