Accurate crop yield prediction via temporal convolutional network and variational autoencoder

Abstract

With the advancement of technology, smart agriculture assumes a pivotal role in enhancing the efficiency and sustainability of agricultural production. Nevertheless, accurately forecasting crop yields and understanding their determining factors often faces limitations inherent in conventional data processing techniques. To address this challenge, this study proposes a TCN-VAE framework, amalgamating temporal convolutional networks (TCNs) and variational autoencoders (VAEs), to tackle the crop yield prediction task in smart agricultural systems. The framework intricately models four principal categories of features—meteorological conditions, soil conditions, agricultural management, and environmental conditions—via TCN and compensates for missing data using VAE, thereby achieving a comprehensive analysis and prediction of multidimensional data and ultimately enhancing the precision of yield prediction. In the experiments, the TCN-VAE framework adeptly captures complex time series data and environmental features, demonstrating yield prediction performance that is markedly superior to that of traditional methods. The experimental results substantiate that the framework can enhance the accuracy and reliability of crop yield predictions, offering a novel technical approach and strategy for smart agriculture management. This research not only fosters the advancement of smart agriculture technology but also provides robust methodological support for farmers and agribusinesses, facilitating more scientific farm management and decision-making.