Multi-Band Forecasting of Ambient RF Power to Guide Optimal Placement of Energy-Harvesting Systems

RF energy harvesting (RF-EH) is emerging as a practical power source for low-power IoT. We develop multi-frequency forecasting models that predict ambient available RF power across multiple bands to guide best placement, time window, and frequency band for EH deployment. Measurements were done using a broadband omnidirectional antenna and a Fieldfox handheld RF analyzer at three different locations across 16 frequencies, which are then used to train on two locations and evaluate on a held-out third. We compare LSTM, GRU, and 1D-CNN against a hybrid fusion (LSTM+GRU+1D-CNN). The hybrid architecture achieves MAE = 0.77 dBm, RMSE = 1.43 dBm, while the 1D-CNN reaches MAE = 0.78 dBm, RMSE = 1.53 dBm. 1-D CNN has a substantially lower computational power and training time while the hybrid model could be preferred when peak accuracy is required.