Method and device for model training and smell source positioning

The invention discloses a method and a device for model training and smell source positioning. The method comprises the following steps: acquiring environment data corresponding to an environment in which a carrier for searching an odor source is located at a first set moment, and inputting the environment data into a preset feature fusion model, so that the feature fusion model outputs environment features corresponding to the environment data. And then, inputting the environment characteristics into a preset time sequence network, so that the time sequence network outputs environment state information of the environment characteristics at the first set moment based on the environment characteristics at each set moment before the first set moment. And further, inputting the environment state information into a trained prediction model, so that the prediction model predicts a smell distribution map of the environment state information at a second set moment according to the environment state information, and