Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
April 18, 2026IET conference proceedings.

Weather on edge: deploying an MLP-based forecasting model on edge devices for hyper-local monitoring

View Full Paper
Ask AI
Bookmark
Share

Authors

JRJonas Gabriel L. RusianaGLGwyn Ann S. LobatonADAzriel Peter S. Deduyo

Discussion

Loading...

Member takes

Overview

Deploys an edge-capable forecasting model to monitor localized weather conditions, suggesting its practical application for real-time forecasts.

Key Points

  • The aim is to develop an offline weather forecasting model using edge devices for accurate local monitoring.
  • Deployed a Multilayer Perceptron (MLP) model on a Raspberry Pi 4.
  • Collected temperature, humidity, pressure, and rainfall data from onboard sensors.
  • Evaluated forecasts using RMSE, MAE, sMAPE, MASE, and R2 metrics compared to an Exponential Smoothing baseline.
  • Conducted residual quantile analysis for detailed performance insights.
  • MLP model demonstrated higher accuracy and stability than the ETS baseline, with R2 exceeding 0.97.
  • Achieved low-latency performance of less than 0.05 ms.
  • Realized a 65% reduction in model size through INT8 quantization.
  • Forecast errors were within sensor tolerance ranges, confirming system feasibility.

Cite This Study

Rusiana et al. (2026) studied this question.

synapsesocial.com/papers/69e3205140886becb653f779https://doi.org/10.1049/icp.2026.0976
View Full Paper
Ask AI
Bookmark
Share