Forecasting Peak Load Profiles for New High-Power Charging Stations: A K-Means Clustering and Random Forest Approach
(2026) In Master’s Theses in Mathematical Sciences MASM02 20261Mathematical Statistics
- Abstract
- The increasing demand for electric vehicle charging stations creates new challenges for electric grid planning. This thesis proposes a framework for forecasting realistic 24-hour peak load profiles for new direct current fast-charging stations without historical observations. The method combines $K$-means clustering of existing stations’ hourly load patterns and random forest models that predict both the daily load shape and peak load magnitude of a new station. The resulting profile is scaled to create a low-risk predicted load curve and evaluated using leave-one-out cross-validation. The results show that the proposed profiles cover the majority of observed peak loads while remaining below theoretical maximum capacity for much of the... (More)
- The increasing demand for electric vehicle charging stations creates new challenges for electric grid planning. This thesis proposes a framework for forecasting realistic 24-hour peak load profiles for new direct current fast-charging stations without historical observations. The method combines $K$-means clustering of existing stations’ hourly load patterns and random forest models that predict both the daily load shape and peak load magnitude of a new station. The resulting profile is scaled to create a low-risk predicted load curve and evaluated using leave-one-out cross-validation. The results show that the proposed profiles cover the majority of observed peak loads while remaining below theoretical maximum capacity for much of the day. Small charging stations are more difficult to predict, as they can more easily approach full capacity. The framework is designed to support grid planners in making more efficient and risk-aware decisions when assessing new charging station connections. (Less)
- Popular Abstract
- Electric cars have gone from a futuristic idea to an everyday occurrence. Nowadays, it is hard to go about your day without seeing them charging their batteries. Shopping centers, gas stations and grocery stores, are but a few examples of where charging stations can be found. What happens when more and more people buy electric cars instead of fossil fuel cars? The need for charging stations increases. This increase in demand then brings the risk of straining the electric grid that provides the power to the charging stations. This thesis provides tools for grid planners, so that they can investigate whether a new charging station should be built with the current existing infrastructure or not.
A safe approach to this problem, would be... (More) - Electric cars have gone from a futuristic idea to an everyday occurrence. Nowadays, it is hard to go about your day without seeing them charging their batteries. Shopping centers, gas stations and grocery stores, are but a few examples of where charging stations can be found. What happens when more and more people buy electric cars instead of fossil fuel cars? The need for charging stations increases. This increase in demand then brings the risk of straining the electric grid that provides the power to the charging stations. This thesis provides tools for grid planners, so that they can investigate whether a new charging station should be built with the current existing infrastructure or not.
A safe approach to this problem, would be to assume that all charging stations are fully occupied and charge at maximum power, at all hours during the day, this is called the theoretical maximum capacity. When examining historical charging data, we see that this assumption is far from reality. The data consists of hourly measurements of power consumption from Direct Current fast-charging stations within Öresundskraft AB's electric grid. Öresundskraft AB is an energy company based in Helsingborg, southwest Sweden. The aim of this thesis, is to create more realistic peak charging station profiles, so that the electric grid can be used in a more efficient way.
The proposed method separates the problem into two parts: predicting the daily shape of charging demand, and predicting how high the daily peak load may become. These two are then combined, in order to form a more realistic charging station profile. In order to predict a new charging station's daily shape, existing stations are grouped according to their charging patterns. Examples of charging patterns could be if a station is more popular during weekends or weekdays. A machine learning model is then used to predict the new charging station's most likely group charging pattern. A similar machine learning model is also used to predict the new charging station's daily peak load.
The results show that low-risk predicted load profiles usually cover most observed peak loads while still staying below the station's theoretical maximum capacity for much of the day. This means that the method can help grid planners use the electricity grid more efficiently than if they always assume full power at every hour. There is a clear trade-off: more cautious predictions reduce the risk of underestimating demand, while less cautious predictions allow more grid capacity to be used. Small charging stations, with only a few charging points, were harder to predict because they can easily come close to full capacity. The framework can be used as decision support when assessing whether a new fast-charging station can be connected to the grid or whether reinforcements may be needed. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9231434
- author
- Gerholm, Markus LU
- supervisor
- organization
- course
- MASM02 20261
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Machine learning Forecasting peak load curves Electric grid planning Electric vehicle charging
- publication/series
- Master’s Theses in Mathematical Sciences
- report number
- LUNFMS-3144-2026
- ISSN
- 1404-6342
- other publication id
- 2026:E58
- language
- English
- id
- 9231434
- date added to LUP
- 2026-06-05 17:39:24
- date last changed
- 2026-06-05 17:39:24
@misc{9231434,
abstract = {{The increasing demand for electric vehicle charging stations creates new challenges for electric grid planning. This thesis proposes a framework for forecasting realistic 24-hour peak load profiles for new direct current fast-charging stations without historical observations. The method combines $K$-means clustering of existing stations’ hourly load patterns and random forest models that predict both the daily load shape and peak load magnitude of a new station. The resulting profile is scaled to create a low-risk predicted load curve and evaluated using leave-one-out cross-validation. The results show that the proposed profiles cover the majority of observed peak loads while remaining below theoretical maximum capacity for much of the day. Small charging stations are more difficult to predict, as they can more easily approach full capacity. The framework is designed to support grid planners in making more efficient and risk-aware decisions when assessing new charging station connections.}},
author = {{Gerholm, Markus}},
issn = {{1404-6342}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master’s Theses in Mathematical Sciences}},
title = {{Forecasting Peak Load Profiles for New High-Power Charging Stations: A K-Means Clustering and Random Forest Approach}},
year = {{2026}},
}