Mean Radiant Temperature Calculation Using Neural Networks in Python

Reducing energy consumption is a very important topic for engineers and researchers. The need for this reduction, combined with reductions in energy generation using fossil sources, with the aim of mitigating climate change, has led to research and the proposal of new ways to make the equipment responsible for energy consumption more efficient. An academic or office building installed in a hot climate exhibits very inefficient use of electricity for air conditioning; generally, buildings are overcooled, and this causes high energy consumption due to the set points used. Conventional control methods for air conditioning equipment use hysteresis temperature control and only work with the temperature in the space to calculate the operating set point. The mean radiant temperature (MRT) of a specific area is the average temperature of the objects surrounding a person and influences the person’s heat loss or gain due to the exchange of thermal radiation. To directly measure the MRT, a meter called a radiometer is used, which is very expensive and sophisticated; its measurement is carried out using a black balloon thermometer, and empirical formulas are used for its calculation, such as the method of Fanger. In order to generate savings in the use of electrical energy in an academic or office building, the proposed MRT calculation algorithm can help to reduce electrical energy consumption through the use of the estimated predicted mean value (PMV) method proposed by the American Society of Heat, Refrigeration, and Air Conditioning Engineers (ASHRAE). This method uses the ambient temperature, black balloon temperature, and air speed to calculate the MRT. By calculating the set points of air conditioning equipment using the PMV index, it will be possible to reduce electrical energy consumption without sacrificing user comfort. The general objective is to develop an algorithm for calculating the value of the MRT using neural networks in Python. The neural network model developed in Python uses 20 data input points, such as the air temperature, to predict the MRT for single-hidden-layer and double-hidden-layer models, and it obtains R2 values of 0.58 and 0.65, respectively. The three-layer neural network model, using normalized data and 104 input data points, obtains an R2 value of 0.71. Using a second input variable (occupancy) in the developed neural network model increases the R2 coefficient to 0.89.