Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interpretable Load Forecasting via Representation Learning of Geo-distributed Meteorological Factors

Yangze Zhou, Guoxin Lin, Gonghao Zhang, Yi Wang

TL;DR

This work addresses the challenge of selecting geo-distributed meteorological factors for day-ahead load forecasting by introducing a graph-convolutional representation-learning framework that jointly learns MF representations $R_G$ from the graph $G=(X,A)$ and the forecast model $f$. An accelerated Shapley-value interpretation module is integrated to identify subgraphs $G_s$ whose MF most influence forecasts, enabling practical insights for MF deployment. Empirical results on two real datasets show improved forecasting accuracy, especially during extreme conditions like the accumulation temperature effect and sudden temperature changes, and reveal a strong link between MF importance and regional GDP/mainstay industries. The findings provide a principled approach for geo-distributed MF selection and station placement, with potential for robust, interpretable, and cost-efficient weather-informed load forecasting.

Abstract

Meteorological factors (MF) are crucial in day-ahead load forecasting as they significantly influence the electricity consumption behaviors of consumers. Numerous studies have incorporated MF into the load forecasting model to achieve higher accuracy. Selecting MF from one representative location or the averaged MF as the inputs of the forecasting model is a common practice. However, the difference in MF collected in various locations within a region may be significant, which poses a challenge in selecting the appropriate MF from numerous locations. A representation learning framework is proposed to extract geo-distributed MF while considering their spatial relationships. In addition, this paper employs the Shapley value in the graph-based model to reveal connections between MF collected in different locations and loads. To reduce the computational complexity of calculating the Shapley value, an acceleration method is adopted based on Monte Carlo sampling and weighted linear regression. Experiments on two real-world datasets demonstrate that the proposed method improves the day-ahead forecasting accuracy, especially in extreme scenarios such as the "accumulation temperature effect" in summer and "sudden temperature change" in winter. We also find a significant correlation between the importance of MF in different locations and the corresponding area's GDP and mainstay industry.

Interpretable Load Forecasting via Representation Learning of Geo-distributed Meteorological Factors

TL;DR

This work addresses the challenge of selecting geo-distributed meteorological factors for day-ahead load forecasting by introducing a graph-convolutional representation-learning framework that jointly learns MF representations from the graph and the forecast model . An accelerated Shapley-value interpretation module is integrated to identify subgraphs whose MF most influence forecasts, enabling practical insights for MF deployment. Empirical results on two real datasets show improved forecasting accuracy, especially during extreme conditions like the accumulation temperature effect and sudden temperature changes, and reveal a strong link between MF importance and regional GDP/mainstay industries. The findings provide a principled approach for geo-distributed MF selection and station placement, with potential for robust, interpretable, and cost-efficient weather-informed load forecasting.

Abstract

Meteorological factors (MF) are crucial in day-ahead load forecasting as they significantly influence the electricity consumption behaviors of consumers. Numerous studies have incorporated MF into the load forecasting model to achieve higher accuracy. Selecting MF from one representative location or the averaged MF as the inputs of the forecasting model is a common practice. However, the difference in MF collected in various locations within a region may be significant, which poses a challenge in selecting the appropriate MF from numerous locations. A representation learning framework is proposed to extract geo-distributed MF while considering their spatial relationships. In addition, this paper employs the Shapley value in the graph-based model to reveal connections between MF collected in different locations and loads. To reduce the computational complexity of calculating the Shapley value, an acceleration method is adopted based on Monte Carlo sampling and weighted linear regression. Experiments on two real-world datasets demonstrate that the proposed method improves the day-ahead forecasting accuracy, especially in extreme scenarios such as the "accumulation temperature effect" in summer and "sudden temperature change" in winter. We also find a significant correlation between the importance of MF in different locations and the corresponding area's GDP and mainstay industry.
Paper Structure (21 sections, 14 equations, 10 figures, 3 tables, 1 algorithm)

This paper contains 21 sections, 14 equations, 10 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Proposed Framework
  4. GCN-based Load Forecasting
  5. Accelerated Shapley Value-Based Model Interpretation
  6. Shapley Value
  7. Acceleration Strategy
  8. Case Studies
  9. Experimental Setups
  10. Data source
  11. Metric
  12. Benchmark
  13. Forecasting Performance
  14. City A
  15. City B
  16. ...and 6 more sections

Figures (10)

  • Figure 1: The proposed load forecasting framework utilizing geo-distributed MF.
  • Figure 2: An illustration of Shapley value for interpreting GNN
  • Figure 3: The MAE performance of Hongtao method in City A
  • Figure 4: Normalized forecasts of City A from Jul. 1 to Aug. 10
  • Figure 5: The range of temperature and humidity of 18 locations in City A from Jul. 8 to Jul. 12
  • ...and 5 more figures