Crop Yield Prediction based on Attention-LSTM Model
Zirui Xiong, Xin Guo, Danjue Zhang, Shih-Pang Tseng, Yan Jiao Ji, Xuan Sun · 2024
With the widespread application of big data and artificial intelligence technologies in the agricultural sector, accurate prediction of crop yields has become crucial, not only for food security but also for significantly enhancing agricultural management efficiency. This paper delves into various prediction models, including traditional machine learning algorithms and deep learning-based time series prediction models, with a particular focus on the integration of the attention mechanism with Long Short-Term Memory (LSTM) networks. By constructing an LSTM model fused with the attention mechanism (Attention Mechanism LSTM Model, Attention-LSTM), this study significantly improves the accuracy of crop yield predictions, which is of great significance for ensuring the stability of food supply and optimizing the allocation of agricultural resources.In the experimental section, a large dataset encompassing factors such as fertilizer usage and soil properties was used to rigorously test the Attention-LSTM model. The results showed that, compared to models such as BP neural networks, GRU, BILSTM, and LSTM, the Attention-LSTM exhibited excellent performance in key evaluation metrics including RMSE, MSE, and MAPE, achieving significant improvement in prediction accuracy and enhancement of generalization ability.The contribution of this paper lies in the development of the Attention-LSTM model, which, with its efficient and precise crop yield prediction capabilities, provides a powerful tool for enhancing the level of agricultural intelligence and optimizing agricultural production decisions. It also showcases the vast potential of deep learning technologies in smart agriculture.