Spatial Electric Load Forecasting Method Based on High-Level Encoding of High-Resolution Remote Sensing Images

Abstract

Spatial load forecasting (SLF) is important for regional power infrastructure construction planning and power grid management. However, for rapidly developing urban regions, SLF is generally inaccurate due to insufficient historical data. Hence, it is important to introduce the spatial load density (SLD) from similar regions to improve the accuracy of SLF. To select similar regions appropriately and acquire SLDs with limited available auxiliary data, this study proposes a spatial electric load forecasting method based on the high-level encoding of high-resolution remote sensing images called SELF-HE. In particular, SELF-HE introduces high-level ground object features as a key index to describe the characteristics of electric loads in a region and can establish connections between the remote sensing image features and SLD similarity. Based on this functionality, SELF-HE achieves more accurate SLF in regions with insufficient historical data. In the experiments, SELF-HE was compared with four traditional methods, and the results revealed that SELF-HE achieved improved SLF accuracy. Given that the high-resolution remote sensing images fully covered urban areas and were readily obtained, the proposed method can improve the accuracy of SLF with extremely low data collection costs and is applicable to rapidly developing urban regions.