MPC Predictor SESS Benchmark

This repository contains the code, and evaluation scripts accompanying the paper “A Comprehensive Evaluation of Prediction Techniques and Their Influence on Model Predictive Control in Smart Energy Storage Systems”.

It provides a reproducible benchmark of forecasting models (Linear, XGBoost, RNN, TimeMixer, and TimesNet) for load, PV generation, and electricity price prediction. These models are evaluated both on standard error metrics (MSE, RMSE, MAE, MAPE, R²) and in combination with a model predictive controller (MPC) to assess their real-world impact on smart energy storage system (SESS) performance.

Results

Predictor Tuning

• linear
  • 20250204_141500: Linear, load
  • 20250204_184612: Linear, PV
  • 20250204_222016: Linear, price
• xgboost
  • 20250128_182124: XGBoost, load
  • 20250129_211746: XGBoost, PV
  • 20250131_074032: XGBoost, price
• recurrent-net
  • 20250201_203130: RNN, load
  • 20250202_122543: RNN, PV
  • 20250203_061524: RNN, price
• time-mixer
  • 20250128_151138: TimeMixer, load
  • 20250129_071826: TimeMixer, PV
  • 20250131_143014: TimeMixer, price
• times-net
  • 20250128_135659: TimesNet, load
  • 20250131_104458: TimesNet, PV
  • 20250203_150716: TimesNet, price

Predictor Test

• linear
  • 20250207_072124: Linear, load
  • 20250207_072235: Linear, PV
  • 20250207_072340: Linear, price
• xgboost
  • 20250207_071942: XGBoost, load
  • 20250207_073746: XGBoost, PV
  • 20250207_072613: XGBoost, price
• recurrent-net
  • 20250207_070515: RNN, load
  • 20250207_070826: RNN, PV
  • 20250207_071556: RNN, price
• time-mixer
  • 20250206_205215: TimeMixer, load
  • 20250206_205225: TimeMixer, PV
  • 20250207_084358: TimeMixer, price
• times-net
  • 20250207_074319: TimesNet, load
  • 20250207_075541: TimesNet, PV
  • 20250211_072513: TimesNet, price

Predictor MPC Evaluation

• base: 20250208_092114
• perfect: 20250207_143021
• standard
  • linear: 20250722_203255
  • xgboost: 20250722_205950
  • recurrent-net: 20250722_204540
  • times-net: 20250723_032007
  • time-mixer: 20250723_030559
• retrain
  • linear: 20250723_033326
  • recurrent-net: 20250723_034609
  • time-mixer: 20250723_035851
  • times-net: 20250723_043739
  • xgboost: 20250725_124430

Acknowledgments

Some of the prediction-model implementations are adapted from the Time-Series-Library. Many thanks to that community for making their work available.

Citation

Paper: https://doi.org/10.1016/j.segy.2025.100202

If you use this repository or find it helpful in your own research, please cite:

@misc{Ludolfinger2025,
    title = {A comprehensive evaluation of prediction techniques and their influence on model predictive control in smart energy storage systems},
    journal = {Smart Energy},
    pages = {100202},
    year = {2025},
    issn = {2666-9552},
    doi = {https://doi.org/10.1016/j.segy.2025.100202},
    author = {Ulrich Ludolfinger and Thomas Hamacher and Maren Martens},
}

Licensing

• The code and documentation are released under the MIT License.
• The electricity price data in res/data/ee_prices.csv are © Bundesnetzagentur | SMARD.de and redistributed under the CC BY 4.0 license.
• The files opsd_building_*.csv in res/data originate from the Open Power System Data project and are also redistributed under the CC BY 4.0 license.