Basis study about prediction to air flow environment in cross ventilated room by neural network

In many parts of Asia as typified by Japan, conditioning of the indoor thermal and air environments using natural ventilation since ancient times. When indoor thermal and air environments are predicted, the use of simulation technologies such as CFD and Heating and Ventilation Network Model has increased. Those have advantages and disadvantages. In addition, AI programs like Neural Network (NN) and Genetic Algorithm (GA) are increasingly utilized in other research areas. In architectural equipment field, there are examples of airconditioning system models with NN. These programs are relatively easy to use, finish the calculation quickly, and even conduct assessment and prediction. However, there are few application examples of NN in simulations of thermal and air environment of cross-ventilated room. This study examines fundamental investigations in the application of NN to cross-ventilation environmental simulation. As a result, it revealed that the results of the simulations with NN tended similarly to the results of CFD under the condition that 2 openings were open. Although, by the combination of the cases with 2 openings open, the simulation of NN with 3 openings in case,
which had small calculation load and high simulation accuracy, gained low accuracy and basically resulted in similar wind directions to the results of CFD.