Time-Series RF-EMF Predictions for Exposure Assessments Using Machine Learning Techniques

This paper performs time-series forecasts for radio-frequency electromagnetic field (RF-EMF) exposure assessments, leveraging machine learning (ML) methods. More specifically, a deep bidirectional long short-term memory (BiLSTM) network is proposed, which is trained and tested on long-term electric field time-series data collected by a monitoring station at frequencies between 100 kHz and 7 GHz. The adopted ML network performs very well during the testing phase delivering quite accurate time-series forecasts for unknown inputs. Finally, future electric field sequences are forecasted where the encountered electric field periodic patterns, which are correlated with the daily life, remain consistent.