C207 ニューラルネットモデル内包学習による3次元空間の速度・濃度場の推定

抄録

This paper proposes a novel algorithm using an artificial neural network for modelling both 3-D flow velocity vector and concentration fields from measured values and boundary conditions. By using its trained neural network, we can estimate 3-D velocity vectors and concentrations over the entire field. The network is trained by using sparsely measured values and boundary conditions as teaching data so that the output of the network agrees well with those data. In addition, the continuity and the diffusion equations are systematically satisfied by the model inclusive learning. In order to evaluate the effectiveness, the proposed method is applied to air flow fields in the full size model of an infection-free hospital room. The concentration distributions in the room are estimated and then compared with the experimental smoke concentration data over the field for validating the proposed algorithm. The estimated results grasp the experimental diffusion behavior.