The invention discloses a multi-
modal fusion
storage tank fire near-field
radiation prediction method, which comprises the following steps of: 1, building reduced scale experiment devices for different numbers of
storage tank fire under the action of a
wind field, acquiring a multi-oil-
pool flame image and a near-field
radiation signal, and constructing a multi-fire-source experiment
data set; 2, establishing a deep convolutional
feature extraction network based on VGG16, obtaining multi-channel deep features representing a
flame structure, brightness distribution and a turbulence form through a
flame image, mapping the number and the position of a fire source and the quality of an oil product into same-dimension vectors, and constructing a multi-source fusion
feature set; 3, constructing a
Transformer-CNN-BiLSTM near-field
radiation prediction model, fusing space and time characteristics, and realizing radiation
dynamic prediction under different
field conditions,
wind field disturbance and oil product
combustion states; and 4, constructing a
loss function for training an optimal prediction model. According to the invention, the real-time prediction of the
combustion rate and the near-field radiation distribution can be realized based on the flame image and the multi-source
physical information in a complex environment, and the accuracy and the dynamic response capability of fire monitoring are improved.