Skip to content

Efficiency on Edge Devices

Home / Open Questions / Efficiency on Edge Devices

Background: Transformer-based architectures are increasingly applied to solar power forecasting, yet these models face significant computational and memory hurdles when deployed on resource-constrained edge devices.

Question / Future Work: There is a need to develop more efficient versions of transformer-based solar forecasting models that are suitable for deployment on edge devices and in rural locations. Future research is required to explore techniques such as lightweight attention approximations, model pruning, quantization, or alternative methodologies to reduce the memory and computational footprint of these architectures.

Why It Matters: This is a critical bottleneck for real-world adoption of high-accuracy deep learning models in remote PV stations where hardware resources are limited.

Metadata & Links