Paper: arXiv 2405.17070
Authors: Monika Zimmermann, Florian Ziel
Abstract
Accurate mid-term (weeks to one year) hourly electricity load forecasts are essential for strategic decision-making in power plant operation, ensuring supply security and grid stability, planning and building energy storage systems, and energy trading. While numerous models effectively predict short-term (hours to a few days) hourly load, mid-term forecasting solutions remain scarce. In mid-term load forecasting, capturing the multifaceted characteristics of load, including daily, weekly and annual seasonal patterns, as well as autoregressive effects, weather and holiday impacts, and socio-economic non-stationarities, presents significant modeling challenges. To address these challenges, we propose a novel forecasting method using Generalized Additive Models (GAMs) built from interpretable P-splines that is enhanced with autoregressive post-processing. This model incorporates smoothed temperatures, Error-Trend-Seasonal (ETS) modeled and persistently forecasted non-stationary socio-economic states, a nuanced representation of effects from vacation periods, fixed date and weekday holidays, and seasonal information as inputs. The proposed model is evaluated using load data from 24 European countries over more than 9 years (2015-2024). This analysis demonstrates that the model not only has significantly enhanced forecasting accuracy compared to state-of-the-art methods but also offers valuable insights into the influence of individual components on predicted load, given its full interpretability. Achieving performance akin to day-ahead Transmission System Operator (TSO) forecasts, with computation times of just a few seconds for several years of hourly data, underscores the potential of the model for practical application in the power system industry.
Complexity vs Empirical Score
- Math Complexity: 4.5/10
- Empirical Rigor: 8.5/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper employs interpretable Generalized Additive Models (GAMs) with P-splines, which is a standard but moderately complex statistical technique, and focuses heavily on practical application with extensive multi-country, multi-year backtesting, performance benchmarks against TSO forecasts, and computational efficiency.
Research Flowchart
flowchart TD
A["Research Goal<br>Develop Accurate &<br>Interpretable Mid-Term<br>Hourly Load Forecasting"] --> B["Data Collection & Input<br>9+ Years (2015-24)<br>24 European Countries"]
B --> C{"Core Methodology<br>Generalized Additive Models with P-Splines"}
C --> D["Input Features Processing"]
D --> E["Smoothed Temperatures"]
D --> F["ETS Modeled Socio-Economic<br>States (Forecasted persistently)"]
D --> G["Holiday Effects<br>Fixed date & Weekday"]
D --> H["Seasonal Information<br>Daily, Weekly, Annual"]
E & F & G & H --> C
C --> I["Autoregressive<br>Post-Processing"]
I --> J["Key Outcomes & Findings"]
J --> K["Significantly Enhanced<br>Forecasting Accuracy"]
J --> L["Full Interpretability<br>Insights into individual<br>component impacts"]
J --> M["Computational Efficiency<br>Seconds for years of<br>hourly data"]
J --> N["Performance comparable<br>to Day-Ahead TSO Forecasts"]
style A fill:#e1f5fe
style J fill:#f3e5f5
style C fill:#fff3e0
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a novel two-stage forecasting method using Generalized Additive Models with interpretable P-splines and autoregressive post‑processing.
Source sentence
we propose a novel forecasting method using Generalized Additive Models (GAMs) built from interpretable P-splines that is enhanced with autoregressive post‑processing.
Assumes 6
- The time series begins on January 1st at 00:00. data
Source sentence
Without loss of generality, we assume that the time series begins on January 1st at 0 : 00.
- Hourly observations are used (S = 24 per day) but the method works for any S ≥ 3. model
Source sentence
As outlined in Section 2, our dataset encompasses hourly observations, resulting in S = 24 observations per day. Nevertheless, the methodology introduced in the subsequent sections is formulated for arbitrary observations per day S ∈N≥3
- Forecasting horizon is 52 weeks (H = 168 × 52) and experiments are randomly sampled. execution
Source sentence
observations is rolled forward by 24 hours, whereby the first hour to forecast is always 9 a.m. The forecasting horizon is chosen as 52 weeks, i.e. H = 168 × 52.
- Mean absolute error (MAE) is used as the primary error metric. statistical
Source sentence
We evaluate forecasting accuracy using the mean absolute error MAE10 in three distinct ways
Show 2 more
- Temperature inputs are exponentially smoothed. model
Source sentence
This model uses exponentially smoothed temperatures
- Socio‑economic non‑stationary states are modeled via ETS and persistently forecasted. model
Source sentence
Error‑Trend‑Seasonal (ETS) modeled and persistently forecasted non‑stationary socio‑economic states
Data
- Hourly load data from 24 European countries other public
Source sentence
The proposed model is evaluated using load data from 24 European countries over more than 9 years (2015-2024).
- Temperature data from weather stations alternative unknown
Source sentence
Utilizing temperature data Tempt ∈R observed at weather stations
Costs
Not applicable
Finds 5
- The proposed GAM models outperform all benchmark models (except FNN and some GAM‑type models) across all 24 countries. performance
Source sentence
our proposed GAM models (GAM, \AR, \HLD, \Temp, \States) distinctly outperform all benchmark models except the FNN and GAM-type models (CLIM, CL-ETS) in every country included in the study.
- The best‑performing GAM model shows superior performance compared to all benchmarks across all countries. performance
Source sentence
Our best-performing GAM model has superior performance compared to all benchmarks across all countries.
- Including temperature effects improves forecasting accuracy, with a hierarchy \Temp < CLIM < CL‑ETS < GAM. predictive-accuracy
Source sentence
Specifically, we find the following performance hierarchy in most countries: \Temp < CLIM < CL-ETS < GAM which proves the skill of our temperature model over a climatological approach.
- Vanilla benchmark models exhibit low forecasting accuracy, but adding actual or forecasted temperatures improves them. performance
Source sentence
... However, including actual temperatures and recency temperature effects, even when only forecasted by seasonality, improves Vanilla forecasting accuracy.
- The model runs in a few seconds for year‑ahead hourly forecasts, enabling practical application. other
Source sentence
computation times of just a few seconds for several years of hourly data
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (71)
- Short Term Electric Load Forecasting (2012)
- GEFCom2012: Electric load forecasting and backcasting with semi-parametric models (2013) — link
- Trends and cycles in economic time series: A Bayesian approach (2006) — link
- Short-Term Load Forecasting With Seasonal Decomposition Using Evolution for Parameter Tuning (2015) — link
- Local Short and Middle Term Electricity Load Forecasting With Semi-Parametric Additive Models (2014) — link
- Generalized Additive Models for Large Data Sets (2014) — link
- The evaluation of extrapolative forecasting methods (1992) — link
- Are fluctuations in energy variables permanent or transitory? A survey of the literature on the integration properties of energy consumption and production (2012) — link
and 63 more
- Seasonal climate forecasts for medium-term electricity demand forecasting (2014) — link
- On the asymmetry of the symmetric MAPE (1999) — link
- Trends and random walks in macroeconmic time series (1982) — link
- Comparing Predictive Accuracy (1995) — link
- Long Term Probabilistic Load Forecasting and Normalization With Hourly Information (2013) — link
- A study of outliers in the exponential smoothing approach to forecasting (2011) — link
- A unit root model for trending time-series energy variables (2014) — link
- Medium term system load forecasting with a dynamic artificial neural network model (2005) — link
- Testing of unit root and other nonstationary hypotheses in macroeconomic time series (1997) — link
- Short-Term Load Forecasting Based on a Semi-Parametric Additive Model (2011) — link
- Further evidence on breaking trend functions in macroeconomic variables (1997) — link
- Global Energy Forecasting Competition 2012 (2013) — link
- Generalized Additive Models: An Introduction With R (2007) — link
- Explaining Cointegration Analysis: Part II (2001) — link
- Load Forecasting Using Support Vector Machines: A Study on EUNITE Competition 2001 (2004) — link
- A note on the Mean Absolute Scaled Error (2015) — link
- Probabilistic electric load forecasting: A tutorial review (2016) — link
- Additive models and robust aggregation for GEFCom2014 probabilistic electric load and electricity price forecasting (2016) — link
- Lasso estimation for GEFCom2014 probabilistic electric load forecasting (2016) — link
- Probabilistic energy forecasting: Global Energy Forecasting Competition 2014 and beyond (2016) — link
- GEFCom2014 probabilistic electric load forecasting using time series and semi-parametric regression models (2016) — link
- Relative Humidity for Load Forecasting Models (2016) — link
- Benchmarking robustness of load forecasting models under data integrity attacks (2017) — link
- Long term load forecasting with hourly predictions based on long-short-term-memory networks (2018) — link
- Modeling public holidays in load forecasting: a German case study (2018) — link
- Short term load forecasting and the effect of temperature at the low voltage level (2018) — link
- Optimal Deep Learning LSTM Model for Electric Load Forecasting using Feature Selection and Genetic Algorithm: Comparison with Machine Learning Approaches † (2018) — link
- Forecasting principles and practice
- Creating a proof-of-concept climate service to assess future renewable energy mixes in Europe: An overview of the C3S ECEM project (2018) — link
- Quantile regression for the qualifying match of GEFCom2017 probabilistic load forecasting (2018) — link
- Enhanced Deep Networks for Short-Term and Medium-Term Load Forecasting (2018) — link
- Global energy forecasting competition 2017: Hierarchical probabilistic load forecasting (2019) — link
- Combining Weather Stations for Electric Load Forecasting (2019) — link
- Making and Evaluating Point Forecasts (2009) — link
- Changes in electricity demand pattern in Europe due to COVID-19 shutdowns (2020) — link
- The impact of different COVID-19 containment measures on electricity consumption in Europe (2020) — link
- How to model European electricity load profiles using artificial neural networks (2020) — link
- Adaptive Methods for Short-Term Electricity Load Forecasting During COVID-19 Lockdown in France (2021) — link
- N-BEATS neural network for mid-term electricity load forecasting (2021) — link
- Texts in Statistical Science (2018) — link
- Forecasting: theory and practice (2022) — link
- The non-linear link between electricity consumption and temperature in Europe: A threshold panel approach (2008) — link
- Comparing Predictive Accuracy (1994) — link
- Practical Smoothing (2021) — link
- Probabilistic Forecasting of Regional Net-load with Conditional Extremes and Gridded NWP (2022) — link
- Probabilistic Forecasting of Regional Net-Load With Conditional Extremes and Gridded NWP (2021) — link
- Modeling Energy Demand—A Systematic Literature Review (2021) — link
- Global changes in electricity consumption during COVID-19 (2021) — link
- State-Space Models for Online Post-Covid Electricity Load Forecasting Competition (2022) — link
- A data-driven knowledge-based system with reasoning under uncertain evidence for regional long-term hourly load forecasting (2022) — link
- Smoothed Bernstein Online Aggregation for Short-Term Load Forecasting in IEEE DataPort Competition on Day-Ahead Electricity Demand Forecasting: Post-COVID Paradigm (2022) — link
- Intelligence based Accurate Medium and Long Term Load Forecasting System (2022) — link
- Introduction to Time Series and Forecasting (2016) — link
- Generalized Additive Models (2006) — link
- Probabilistic forecasting method for mid-term hourly load time series based on an improved temporal fusion transformer model (2022) — link
- Adaptive Probabilistic Forecasting of Electricity (Net-)Load (2023) — link
- High-Resolution Load Forecasting on Multiple Time Scales Using Long Short-Term Memory and Support Vector Machine (2023) — link
- Modelling the unit root properties of electricity data—A general note on time-domain applications (2023) — link
- Incorporating air temperature into mid-term electricity load forecasting models using time-series regressions and neural networks (2023) — link
- Forecasting and Analytics with the Augmented Dynamic Adaptive Model (ADAM) (2023) — link
- Using weather data in energy time series forecasting: the benefit of input data transformations (2023) — link
- Carbon Monitor Power-Simulators (CMP-SIM v1.0) across countries: a data-driven approach to simulate daily power generation (2024) — link
- Influence of population mobility on electricity consumption in seven U.S. cities during the COVID-19 pandemic (2024) — link
0 of 71 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.