I joined the Smart Charge Project in 2020 during the system development and validation phase. This Interreg-funded collaboration between UiT, Narvik, and The Lapland University of Applied Sciences in Rovaniemi, Finland, provided an opportunity to apply advanced predictive modeling techniques to real-world energy management challenges in Arctic environments. The project’s focus on integrating electric mobility with smart home technology presented unique, real-world datasets and modeling requirements that aligned with my research interests (at the time) in time series prediction and energy system optimization.
The Smart Charge initiative, which ran from October 2019 to June 2023, brought together diverse stakeholders, including Aurora Powertrains, Saga Energy, Narvik Municipality, Vattenfall, and the Snow Hotel. My involvement centered on the data analysis and predictive modeling components that would facilitate intelligent energy management decisions within the system’s multi-agent framework.
Technical Context and Energy System Challenges
The project addressed the energy management complexities inherent to Arctic rural areas. Conventional energy infrastructure in such regions faces limitations due to geographic isolation and extreme environmental conditions. The system under development required predictive capabilities to manage bidirectional energy flows between electric snowmobiles and residential properties. This necessitated reliable forecasting models that could account for the patterns of Arctic energy consumption dictated by environmental characteristics.
When I joined the project team, the foundational architecture was established—electric snowmobiles functioning as mobile energy storage units within a vehicle-to-grid (V2G) framework. My task involved developing the analytical basis that would enable these vehicles to make intelligent charging and discharging decisions based on predicted energy demands, pricing fluctuations, and system constraints.
Methodology and Model Development
My primary contribution focused on developing machine learning and neural network models capable of handling single and multi-variate predictions across varying temporal horizons. The complexity of Arctic energy systems requires approaches to capture seasonal variations, tourism-related demand fluctuations, and weather-dependent consumption patterns.
For short-term predictions, I implemented machine learning (e.g., random forest, XGBoost, etc.) and neural network architectures optimized for hourly and daily forecasting. They incorporated real-time data streams from smart home sensors, weather stations, and grid monitoring systems. These models needed to process multiple input variables simultaneously—temperature fluctuations, occupancy patterns, historical consumption data, and equipment operational states—to generate reliable energy demand forecasts.
The long-term prediction models presented different challenges. They required approaches to identify and extrapolate seasonal trends while accounting for the stochastic nature of Arctic weather patterns. I implemented deep learning approaches, specifically focusing on CNN and LSTM-based architectures that could maintain prediction accuracy across extended forecasting horizons spanning weeks to months.
The multi-variate prediction models proved valuable for optimizing the V2G system’s operational efficiency. By incorporating variables such as electricity pricing from Norwegian markets and weather forecasts, the models could anticipate energy demand patterns with sufficient accuracy to enable proactive grid management decisions.
Model validation revealed interesting characteristics specific to Arctic energy consumption. The seasonal variations proved more pronounced than initially anticipated, with winter tourism periods creating distinct demand signatures that required specialized handling within our prediction algorithms. Additionally, the interaction between heating demands and vehicle charging requirements demonstrated complex interdependencies that benefited from our multi-variate modeling approach.
Outcomes and Dissemination of Knowledge
The project culminated in presentations at CIRED 2023 in Rome, where I presented research findings titled “Using Light Electric Vehicles for V2G Services in the Arctic.” This paper detailed the machine learning methodologies developed for energy prediction in Arctic environments, particularly emphasizing the comparative performance of single-step versus multi-step prediction approaches.
The research demonstrated measurable improvements in prediction accuracy when incorporating Arctic-specific variables into model architectures, contributing to the broader literature on energy forecasting in extreme environments. The project leader and my adviser, Bernt Arild Bremdal, presented complementary work titled “Predicting Peak Prices in the Current Day-Ahead Market” and provided additional context for understanding the economic optimization aspects of our predictive models.
Beyond the conference presentations, this work contributed to ongoing research initiatives within our AI Group, including investigations into federated learning approaches for distributed energy systems and edge computing applications for real-time demand response optimization. Furthermore, the Smart Charge Project provided valuable insights into applying predictive modeling techniques within constrained energy environments. The success of our multi-variate approaches suggests promising directions for further research in Arctic energy management, particularly regarding the scalability of these methods to larger distributed energy networks.