Justifying Short-Term Load Forecasts Obtained with the Use of Neural Models

Tadeusz A Grzeszczyk,Michal K Grzeszczyk

doi:10.3390/en15051852

Tadeusz A Grzeszczyk, Michal K Grzeszczyk

Open Access

https://doi.org/10.3390/en15051852

Copy DOI

Journal: Energies	Publication Date: Mar 2, 2022
Citations: 7	License type: CC BY 4.0

Affiliation: Warsaw University of Technology

Abstract

There is a lot of research on the neural models used for short-term load forecasting (STLF), which is crucial for improving the sustainable operation of energy systems with increasing technical, economic, and environmental requirements. Neural networks are computationally powerful; however, the lack of clear, readable and trustworthy justification of STLF obtained using such models is a serious problem that needs to be tackled. The article proposes an approach based on the local interpretable model-agnostic explanations (LIME) method that supports reliable premises justifying and explaining the forecasts. The use of the proposed approach makes it possible to improve the reliability of heuristic and experimental neural modeling processes, the results of which are difficult to interpret. Explaining the forecasting may facilitate the justification of the selection and the improvement of neural models for STLF, while contributing to a better understanding of the obtained results and broadening the knowledge and experience supporting the enhancement of energy systems security based on reliable forecasts and simplifying dispatch decisions.

Highlights

Power load forecasting is of increasing importance due to the role of electricity and the daily functioning of the information society in running a business; there is a need to synchronize three processes: power generation, transmission and utilization, difficulties with storing large amounts of electric energy, and inevitable changes in power systems towards highly complex and intelligent solutions [1]
From the point of view of the forecasting horizon and the dissimilarities of the problems to be solved related to it, load forecasting in power systems can be classified as follows: from a few minutes to hour-ahead scheduling—very short-term load forecasting (VSTLF); from hourly, daily and weekly to yearly time series (TS)—short-term load forecasting (STLF); up to 3 years ahead—medium-term load forecasting (MTLF) and up to 10 years ahead—long-term load forecasting (LTLF) [3]
The type of neural networks used and the structural parameters of the models are, to a large extent, selected heuristically. These are black-box models that provide results that are difficult or impossible to interpret, on the one hand, and, on the other, regarding the selection of critical parameters and tasks related to control processes, that work for the safety and planning operation of power systems

Summary

Introduction

Power (electric) load forecasting is of increasing importance due to the role of electricity and the daily functioning of the information society in running a business; there is a need to synchronize three processes: power generation, transmission and utilization, difficulties with storing large amounts of electric energy, and inevitable changes in power systems towards highly complex and intelligent solutions [1]. The type of neural networks used and the structural parameters of the models are, to a large extent, selected heuristically These are black-box models that provide results that are difficult or impossible to interpret, on the one hand (most often there is no justification for the forecasts obtained for specific input data), and, on the other, regarding the selection of critical parameters and tasks related to control processes, that work for the safety and planning operation of power systems. Explaining the forecasting may facilitate the justification of the selection and the improvement of neural models for STLF, contribute to a better understanding of the obtained results, simplify dispatch decisions and broaden the knowledge and experience supporting the enhancement of energy systems security based on reliable forecasts. The last section briefly highlights the significant results achieved and outlines possible directions for further research

Possibility of Justifying and Explaining Neural Forecasts

Results and Discussion

Explanations