Tailoring the Tails: Enhancing the Reliability of Probabilistic Load Forecasts

Roberto Baviera, Pietro Manzoni
Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications, PMLR 230:508-521, 2024.

Abstract

Quantifying predictive uncertainty regarding future electricity demand is the main goal of probabilistic load forecasting. A good probabilistic model is often identified with forecasted densities that are as concentrated (“sharp”) as possible. However, this goal is frequently achieved by sacrificing forecast reliability, i.e. the statistical compatibility between forecasted densities and observed frequencies. In real-world applications, reliability is the crucial measure of model quality, especially when predicting distribution tails. We propose a new methodology for probabilistic load forecasting, introducing a novel loss function which allows an excellent balance between forecast sharpness and reliability. We apply the proposed modelling approach for predicting the electricity load on a benchmark dataset. Experimental results show that the obtained density forecasts are extremely reliable and also close to optimal in terms of sharpness and point accuracy.

Cite this Paper

BibTeX
@InProceedings{pmlr-v230-baviera24a, title = {Tailoring the Tails: Enhancing the Reliability of Probabilistic Load Forecasts}, author = {Baviera, Roberto and Manzoni, Pietro}, booktitle = {Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications}, pages = {508--521}, year = {2024}, editor = {Vantini, Simone and Fontana, Matteo and Solari, Aldo and Boström, Henrik and Carlsson, Lars}, volume = {230}, series = {Proceedings of Machine Learning Research}, month = {09--11 Sep}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v230/main/assets/baviera24a/baviera24a.pdf}, url = {https://proceedings.mlr.press/v230/baviera24a.html}, abstract = {Quantifying predictive uncertainty regarding future electricity demand is the main goal of probabilistic load forecasting. A good probabilistic model is often identified with forecasted densities that are as concentrated (“sharp”) as possible. However, this goal is frequently achieved by sacrificing forecast reliability, i.e. the statistical compatibility between forecasted densities and observed frequencies. In real-world applications, reliability is the crucial measure of model quality, especially when predicting distribution tails. We propose a new methodology for probabilistic load forecasting, introducing a novel loss function which allows an excellent balance between forecast sharpness and reliability. We apply the proposed modelling approach for predicting the electricity load on a benchmark dataset. Experimental results show that the obtained density forecasts are extremely reliable and also close to optimal in terms of sharpness and point accuracy.} }
Endnote
%0 Conference Paper %T Tailoring the Tails: Enhancing the Reliability of Probabilistic Load Forecasts %A Roberto Baviera %A Pietro Manzoni %B Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications %C Proceedings of Machine Learning Research %D 2024 %E Simone Vantini %E Matteo Fontana %E Aldo Solari %E Henrik Boström %E Lars Carlsson %F pmlr-v230-baviera24a %I PMLR %P 508--521 %U https://proceedings.mlr.press/v230/baviera24a.html %V 230 %X Quantifying predictive uncertainty regarding future electricity demand is the main goal of probabilistic load forecasting. A good probabilistic model is often identified with forecasted densities that are as concentrated (“sharp”) as possible. However, this goal is frequently achieved by sacrificing forecast reliability, i.e. the statistical compatibility between forecasted densities and observed frequencies. In real-world applications, reliability is the crucial measure of model quality, especially when predicting distribution tails. We propose a new methodology for probabilistic load forecasting, introducing a novel loss function which allows an excellent balance between forecast sharpness and reliability. We apply the proposed modelling approach for predicting the electricity load on a benchmark dataset. Experimental results show that the obtained density forecasts are extremely reliable and also close to optimal in terms of sharpness and point accuracy.
APA
Baviera, R. & Manzoni, P.. (2024). Tailoring the Tails: Enhancing the Reliability of Probabilistic Load Forecasts. Proceedings of the Thirteenth Symposium on Conformal and Probabilistic Prediction with Applications, in Proceedings of Machine Learning Research 230:508-521 Available from https://proceedings.mlr.press/v230/baviera24a.html.

Related Material