Adaptive Training of Heat Demand Prediction using Continual Learning

Type: MA thesis

Status: finished

Supervisors: Siming Bayer, Adithya Ramachandran, Andreas Maier

One of the primal challenges faced by utility companies is ensuring efficient supply with minimal greenhouse gas emissions. On the track of facilitating the energy transition and mitigating the anthropomorphic climate change, framework for heat demand forecast poses the basis for different applications, impacting the operational as well as the economical efficiency of a utility company. More concretely, VK Energy [1] optimize combined heat and power (CHP) generation systems to increase their overall efficiency and flexibility. In order to operate the CHP system in heating networks with heat storage adaptively to the demand of the electricity system, a real-time, accurate, robust and user friendly (i.e., ideally without extensive hyperparameter tuning) forecast of the heat demand in the respective heating network is indispensable.

As heat demand forecast is an on-going research topic, numerous methods [2-5] have been proposed in the recent years. Although advanced machine learning (ML) and deep learning (DL) methods proposed in the state-of-the-arts demonstrate tremendous capability to predict the heat demand accurately, their performance is limited by the training data. As the heat consumption strongly depends on the weather condition, which is a dynamic environment, ML/DL models trained initially may not be valid anymore with changing consumption behavior and climate. Therefore, continual learning paradigm [6] should be considered to improve the applicability of ML/DL algorithms for heat demand forecast.

In this thesis, following aspects need to be considered:

  • Literature review of heat consumption prediction and continual learning.
  • Development and implementation of a continual learning framework for heat demand prediction comprising different forecasting methos and retraining mechanism.
  • Comprehensive evaluation of the performance of the implemented framework w.r.t. accuracy, robustness, run-time, and potentially usability.

[1] https://www.vk-energie.de/

[2] Y. Zhao, Y. Shen, Y. Zhu and J. Yao, “Forecasting Wavelet Transformed Time Series with Attentive Neural Networks,” 2018 IEEE International Conference on Data Mining (ICDM), 2018, pp. 1452-1457, doi: 10.1109/ICDM.2018.00201.

[3] Chatterjee, Satyaki and Bayer, Siming and Maier, Andreas K “Prediction of Household-level Heat-Consumption using PSO enhanced SVR Model”, NeurIPS 2021 Workshop on Tackling Climate Change with Machine Learning, 2021, https://www.climatechange.ai/papers/neurips2021/42

[4] Kováč, Szabolcs and Micha’čonok, German and Halenár, Igor and Važan, Pavel,” Comparison of Heat Demand Prediction Using Wavelet Analysis and Neural Network for a District Heating Network”, Special Issue “Artificial Intelligence in the Energy Industry”, https://www.mdpi.com/1996-1073/14/6/1545

[5] Chatterjee S., Ramachandran A., Neergaard TF., Maier A., Bayer S., Heat Demand Forecasting with Multi-Resolutional Representation of Heterogeneous Temporal Ensemble. In: NeurIPS 2022 Workshop Tackling Climate Change with Machine Learning, 2022

[6] https://towardsdatascience.com/how-to-apply-continual-learning-to-your-machine-learning-models-4754adcd7f7f