HEAT STRESS MODELING USING NEURAL NETWORKS TECHNIQUE

Abstract

Rising temperature especially in summer is currently a hot debate. Scientists around the world have raised concerns about Heat Stress Assessment (HSA). It depends on the urban geometry, building materials, greenery, environmental factor of the region, psychological and behavioral factors of the inhabitants. Effective and accurate heat stress forecasts are useful for managing thermal comfort in the area. A widely used technique is artificial intelligence (AI), especially neural networks, which can be trained on weather variables. In this study, the five most important meteorological parameters such as air temperature, global radiation, relative humidity, surface temperature and wind speed are considered for HSA. System dynamic approach and a new version of the Gated Recurrent Unit (GRU) method is used for the prediction of the mean radiant temperature, the mean predicted vote and the physiological equivalent temperature. GRU is a promising technology, the results with higher accuracy are obtained from this algorithm. The results obtained from the model are validated with the output of reference software named Rayman. Django's graphical user interface was created which allows users to select the range of thermal comfort scales based on their perception which depends on the age factor, local weather adaptability, and habit of tolerating the heat events. It also gives a warning to the user by color code about the level of discomfort which helps them to schedule and manage their outdoor activities. Future work consists of coupling this model with urban greenery factors to analyze the impact on the estimation of heat stress.