Large scale spatial electric load forecasting framework based on spatial convolution

doi:10.1016/j.ijepes.2019.105582

Large scale spatial electric load forecasting framework based on spatial convolution

2020-05-01

12

PRE

AI

DOI：10.1016/j.ijepes.2019.105582

Original

Original request for help

Save

Abstract

En 中文

This paper presents a novel framework for spatial electric load forecasting based on spatial convolution, which is applied to simulate the neighbors influence in a given region. In space domain, spatial convolution defines the connectivity between nodes and the weights of the convolution matrix define the relative influence of the neighbors. The convolution matrix is defined such a way to guarantee that the summation of the spatial prediction is equal to the global prediction. In the frequency domain, the convolution can be interpreted as a low-pass filter, i.e. it smooths the transitions between neighbors. By convolution, predicted load spreads to areas without any load by neighbor load induction, i.e. existing loads induce load growth also in areas initially without load. Moreover, within this framework concepts such as attraction poles, constraint areas and minimal load are easily applied. This methodology is very efficient in computational aspects, being very adequate to large datasets. Results from a real world dataset are presented and discussed.

Keywords:

Spatial load forecast

Spatial convolution

Spatial autocorrelation

Large scale spatial electric load forecasting framework based on spatial convolution

Abstract

AI Summary

Journal

Organization