Time-series simulation and forecasting for the energy transition.
Electricity markets, batteries, V2G, flex industry — based in the
Netherlands, working at
Birdview Energy and
independently via
Treehouse Energy.
What I work on
I build models and tooling for the Dutch electricity grid: forward
price curves, BESS dispatch and revenue stacking, scenario analysis,
and grid power-flow studies. Most of it lives in Python on top of
PyPSA,
linopy, and
HiGHS — with a custom HiPO build for
the heavier scenarios.
Day to day that means a lot of TenneT, JAO, ENTSO-E, and Ned.nl data;
a lot of optimisation; and a steady stream of small upstream PRs back
into the open-source stack I depend on.
Electricity markets
I trade the Dutch market across its full stack — from cleared
day-ahead curves down to imbalance settlement at quarter-hour
resolution. The work touches every product where a battery, V2G
fleet, or flexible industrial load can earn revenue, plus the
cross-border flows that determine when those products are
binding.
Wholesale
Day-ahead · EPEX SPOT, Nord Pool
Intraday · continuous & auctions
ID3 · ID1 · index pricing
Forwards · cleared & OTC curves
NL · DE · BE · FR · GB · NO · DK
Balancing & ancillary
FCR · symmetric, 4-hour blocks
aFRR capacity · up & down
aFRR energy · merit-order activation
mFRR · imbalance settlement
TenneT MARI/PICASSO integration
Cross-border
Flow-based · JAO domain, PTDFs, RAMs
NTC · legacy bilateral capacity
Single Day-Ahead Coupling (SDAC)
XBID · cross-border intraday
Import/export profile reconstruction
Truth sources
ENTSO-E Transparency Platform
TenneT imbalance, FCR/aFRR
EPEX SPOT · Nord Pool
Ned.nl · NL renewables truth
JAO · flow-based
Generation fleet & commodity fundamentals
Forward power-curve modelling means knowing the fleet — every
CCGT and OCGT marginal cost, every renewables capacity factor,
every retirement and new-build. Below: the technologies I track
and the commodity benchmarks that drive marginal price across
the merit order.
Thermal & firm
CCGT · OCGT · gas turbines
CHP · industrial & district heat
Coal · hard coal & lignite
Biomass · co-fired & dedicated
Nuclear · Oil · Diesel peakers
Renewables fleet
Solar PV · utility & rooftop
Onshore wind
Offshore wind · NL North Sea build-out
Hydro · run-of-river & pumped
Curtailment & redispatch tracking
Storage & flex
Lithium BESS · 1h, 2h, 4h, 8h
V2G · fleet aggregation
Electrolysers · alkaline & PEM
Heat pumps · industrial & district
Demand response · industrial flex
Commodity benchmarks
Gas TTF · NL hub
Coal API2 · ARA delivery
CO₂ EUA · ETS futures
Brent · Diesel · oil-linked peakers
Fuel-to-power conversion w/ heat rates
Renewables modelling
Hourly capacity factors built bottom-up from reanalysis weather
and panel/turbine specs — never trust a published CF, build it
yourself. Calibrated against
Ned.nl for NL solar and wind
(the truth source) and ENTSO-E for the rest of Europe.
Solar PV
PVGIS · panel-level CF
pvlib · physical model chain
atlite · gridded CF profiles
Tilt, azimuth, soiling, snow
Inverter clipping & DC/AC ratio
Wind
atlite · ERA5 → CF
windpowerlib · power curves
Hub-height extrapolation (log law)
Wake losses & availability
Onshore vs offshore turbine classes
Weather datasets
ERA5 · ECMWF reanalysis
ERA5-Land · 9 km surface
SARAH-3 · solar irradiance
MERRA-2 · NASA reanalysis
KNMI · NL weather stations
Cross-checks
Renewables.ninja · Pfenninger & Staffell
Ned.nl · NL realised generation
ENTSO-E · per-zone CFs
IEA PVPS performance ratios
Project-level monitoring data
Time-series & forecasting
Imbalance prices, intraday spreads, and renewables shortfalls —
short-horizon forecasting that has to run live, recalibrate
daily, and survive distribution shifts. Tree-based regressors do
most of the work; deep models earn their keep on longer horizons
and when the feature space is genuinely high-dimensional.
Tree boosting
XGBoost · default for tabular
LightGBM · fast on wide features
CatBoost · categorical-heavy markets
Quantile regression for prediction intervals
SHAP values for diagnostics
Classical & statistical
statsmodels · SARIMAX, VAR
Prophet · seasonal baselines
Exponential smoothing · Holt-Winters
Kalman filters · state-space
Ridge / Lasso for stable benchmarks
Feature engineering
Lag · rolling · expanding windows
Calendar (DST · holidays · DoW · Settlement)
Weather (T · GHI · wind speed at 10/100m)
Market structure (residual load · merit order)
Cross-zone spreads & flow-based shadow prices
Deep & validation
PyTorch · Lightning
Temporal Fusion Transformer · N-BEATS
Walk-forward / time-series CV
Conformal prediction intervals
Backtests against realised settlement
Optimisation & energy systems
Capacity expansion, dispatch, and BESS revenue stacking — all
linear or mixed-integer, all on the open-source
PyPSA stack. HiGHS with HiPO for
the heavier scenarios; small upstream PRs back into linopy and
HiGHS when something blocks a real run.
Energy-system modelling
PyPSA · network & dispatch
PyPSA-Eur · pan-European reference
linopy · LP/MIP on xarray
powerplantmatching · fleet datasets
Custom NL high-voltage power-flow models
Solvers
HiGHS · default, open-source
HiPO · custom build for hard scenarios
Gurobi · CPLEX · when licensed
SCIP · MIP fallback
YALMIP · MATLAB modelling layer
BESS dispatch
Multi-market revenue stacking
SoC-aware perfect-foresight LP
Rolling-horizon receding-horizon dispatch
Stochastic / scenario-based bidding
Degradation-aware throughput penalties
Grid & network
AC & DC OPF
N-1 contingency analysis
Congestion & redispatch modelling
Connection-point feasibility studies
JAO flow-based capacity reconstruction
Data engineering & ML stack
Datasets that don't fit in memory and queries that have to run
on a laptop and a cluster. Columnar formats are non-negotiable;
pandas where the API matters, polars where the throughput does,
DuckDB for everything in between.
DataFrames & arrays
numpy · pandas · the daily driver
polars · larger-than-memory pipelines
xarray · gridded / multi-dim
pyarrow · Arrow-native I/O
dask · for the rare cluster case
Storage & databases
DuckDB · OLAP on parquet, in-process
PostgreSQL · transactional, prod
SQLite · local truth files
TimescaleDB · long time-series
parquet · feather · zarr
ML & numerics
scikit-learn · pipelines & baselines
xgboost · lightgbm · catboost
scipy · statsmodels
pytorch · lightning
Optuna · Hyperopt for HPO
Apps & ops
FastAPI · pydantic
Streamlit · Dash · quick dashboards
Typer · Click · CLIs
pixi · uv · conda · envs
pytest · ruff · mypy
Geo / GIS & grid topology
Grid models start from geometry — substation coordinates,
transmission corridors, land cover, and the gemeente boundaries
that determine connection rights. OpenStreetMap is the only
open dataset that gets the Dutch HV network roughly right; the
rest is calibration.
Vector & raster
geopandas · shapely · pyproj
rasterio · rioxarray
fiona · gdal
EPSG:28992 (Amersfoort/RD) for NL
Sources
OSM · transmission & substations
AHN4 · NL elevation
BAG · BRT · NL administrative
Copernicus · land cover
Topology & viz
networkx · graph algorithms
folium · pydeck · kepler.gl
plotly · matplotlib · cartopy
osmnx · OSM pulls
MATLAB & Simulink
The numerical legacy from TU/e and URE — power-electronics
simulation, EV drivetrain modelling, and the original codebase
of my MSc thesis. I still reach for it when Simscape Electrical
is the right tool, or when a vendor reference model only ships
as a Simulink block diagram.
2014 – 2015University Racing Eindhoven (URE) — EV drivetrain design,
part-time alongside the bachelor.
2017 – 2019MSc Electrical Engineering, Technische Universiteit Eindhoven.
Master's thesis on day-ahead price formation in 2030 with high BESS & V2G penetration —
code ·
scholar.
2017 – 2020Taylor Solar — CTO, leading PV DC/DC optimiser design.
2021 – 2023Lightyear — Research Scientist, day-ahead market pricing models for the solar EV.
2023 – presentTreehouse Energy — independent consulting on energy markets,
BESS, grid congestion, and hourly time-series simulation in Python.
presentBirdview Energy — investment-grade BESS revenue & grid-risk
analytics across NL / DE / BE.
How a BESS valuation comes together
Most of my work flows through a pipeline like this — fundamentals
feed a forward curve, the curve plus asset specs feed a dispatch
optimiser, and the optimiser produces a revenue stack across every
market a battery can sit in.
flowchart LR
A[Fundamentals fuels · weather · capacity] --> D[Calibrate]
B[Asset specs MW · MWh · efficiency · degradation] --> E[Dispatch optimiser linopy + HiGHS]
C[Grid data TenneT · JAO · zonal] --> F[Congestion & curtailment]
D --> G[Hourly forward curve]
G --> E
E --> H[Revenue stack: DA · Intraday · FCR · aFRR · Imbalance]
F --> I[Investment-grade report]
H --> I
Selected projects
Birdcurve
Hourly forward power price curve for NL and neighbouring zones —
built from fundamentals (fuel forwards, capacity, renewables
generation profiles) and calibrated against the cleared
forward market. The curve I use to value BESS revenue across
day-ahead, imbalance, and ancillary stacks.
BESSview
Analytics and visualisation for battery energy storage — operational
telemetry, dispatch traces, revenue attribution by market, and
back-tests against alternative bidding strategies.
NL grid power flow
PyPSA-Eur–based model of the Dutch high-voltage network,
calibrated against TenneT data and used to study congestion,
curtailment risk, and connection-point capacity for new BESS and
industrial load.
Upstream contributions
Small but load-bearing PRs to the optimisation stack —
linopy (warmstart,
benchmarks),
HiGHS (HiPO build
path), and
PyPSA-Eur. The
goal is the same: make the tools faster and less surprising for
everyone solving the same problems.
From my GitHub
A handful of public repos — mostly Python, mostly energy data.
