Scholarlink

<p>Accurate soil moisture prediction and crop yield estimation play a vital role in planning and efficient management of crop water use in a water-stressed agricultural production. The existing methods, such as GRU, LSTM, and CNN-LSTM, may not capture the complex temporal dependence of time-series data in agriculture, which hinders the accuracy of prediction and its application to agricultural production. The present study suggests an AgriFusion-TFT framework to optimize by using the Monarch Butterfly Optimization (MBO) algorithm for integrated soil moisture and crop yield forecasting in paddy cultivation. The static and dynamic covariates, multi-head self-attention, and gated residual networks in the TFT architecture are able to learn long-range temporal dependencies, and the efficient hyperparameter tuning using MBO can increase convergence rate and model accuracy. The framework is applied to the field data from Thanjavur district in Tamil Nadu, India, and the results show a high R² value of 0.974, a low RMSE value of 0.023, and a low MAE value of 0.011, outperforming LSTM, GRU, and CNN-LSTM in all the metrics. The important variables that explain the most variation in crop yield include temperature, soil moisture, and rain. The results show great potential of the framework as a decision support tool for precision irrigation management, water management, and sustainable farm management in paddy growing regions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AgriFusion-TFT: a temporal fusion transformer-based framework for integrated soil moisture forecasting and crop yield prediction with adaptive optimization

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • R. Veerandra Kumar,
  • M. Anbarasan

摘要

Accurate soil moisture prediction and crop yield estimation play a vital role in planning and efficient management of crop water use in a water-stressed agricultural production. The existing methods, such as GRU, LSTM, and CNN-LSTM, may not capture the complex temporal dependence of time-series data in agriculture, which hinders the accuracy of prediction and its application to agricultural production. The present study suggests an AgriFusion-TFT framework to optimize by using the Monarch Butterfly Optimization (MBO) algorithm for integrated soil moisture and crop yield forecasting in paddy cultivation. The static and dynamic covariates, multi-head self-attention, and gated residual networks in the TFT architecture are able to learn long-range temporal dependencies, and the efficient hyperparameter tuning using MBO can increase convergence rate and model accuracy. The framework is applied to the field data from Thanjavur district in Tamil Nadu, India, and the results show a high R² value of 0.974, a low RMSE value of 0.023, and a low MAE value of 0.011, outperforming LSTM, GRU, and CNN-LSTM in all the metrics. The important variables that explain the most variation in crop yield include temperature, soil moisture, and rain. The results show great potential of the framework as a decision support tool for precision irrigation management, water management, and sustainable farm management in paddy growing regions.