Apparatus and method for predicting control variables for smart farm environment control, and recording medium for performing same

Abstract

An apparatus for predicting control variables for smart farm environment control comprises: a data preprocessing module for preprocessing environmental data collected through a sensor in a smart farm environment to convert the environmental data into environmental variables; a modality-specific embedding layer which receives the preprocessed environmental variables, independently processes same according to the distribution of each of the environmental variables, and generates embedding data; a transformer-based multilayer structure which receives the generated embedding data and quantifies and encodes interactions between the environmental variables; a decoder which receives encoded data generated in the transformer-based multilayer structure, defines a relationship between the environmental variables and the control variables as a translation problem, and simultaneously predicts all control variables in a parallel non-autoregressive manner; and an output conversion module for converting the control variables predicted by the decoder into an operation command of a control device of a smart farm, the output conversion module including an independent output projection layer and an activation function for each control variable. Accordingly, the present invention can effectively learn complex interactions between the environmental variables and the control variables in a smart farm environment by utilizing a transformer-based artificial intelligence model, thereby improving the prediction accuracy of the control variables and the operation efficiency of the smart farm.

Images