LONG SHORT-TERM MEMORY TRANSFER LEARNING FOR PREDICTING AIR POLLUTANTS IN THE CONSTRUCTION SITE

The issue of air pollution on construction sites has grown significantly in recent years, threatening the health of workers and people living nearby. Accordingly, several studies based on artificial intelligence have been put out to forecast the spread of pollutants on the construction site. Although different models and characteristics are used, these methods have not yet predicted multiple types of pollutants. In addition, they have a prediction window size of a few hours and need long-term maintenance of the sensor stations for data collection. To overcome these limitations, this paper introduces a new deep learning framework based on long short-term memory transfer learning to predict multiple air pollutants 12 hours in advance at the construction site. Transfer learning can reduce the sensors’ cost and save operational maintenance at various stations. An experiment conducted using data from Airqino stations on a construction site proves that the proposed framework is able to overcome different competing methods, showing very promising prediction results.