Scholarlink

<p>Crop yield prediction is an essential element for agricultural planning and policy development, although the results may be less accurate because of the high variability that could exist in soil types, weather patterns, and crop growing dynamics for each region. The existing approaches for crop yield prediction have been hampered by the fact that they have been using limited data sources and rigid hyperparameter optimization techniques, thus inhibiting the ability to cope with the diversified agricultural data. The problem encountered by the existing approaches is addressed by the proposed work. The proposed framework focuses on the comprehensive integration of crop yield statistics, climatic variables, soil property data, and satellite-based vegetation indices acquired from various open-access datasets. This multi-source data integration thus allows capturing complementary information in the models regarding crop growth, which is not possible in any of the single datasets alone. To determine the optimized parameter settings for the models, a novel adaptive cooperative PSO–DE–Jaya (AC-PDJ) metaheuristic is proposed by hybridizing the global searching capability of PSO, the variation mechanism of DE, and the exploitation strength of the Jaya algorithm. The cooperation among the participating strategies will be adaptively regulated to enhance the convergence with better preservation of solution diversity. The proposed methodology will be evaluated using various machine learning and deep learning models and compared against various state-of-the-art optimization techniques. Experimental results on different crops in various geographical locations have demonstrated that AC-PDJ-optimized models result in lower prediction errors and improved goodness-of-fit with faster and more stable convergence performance. The framework uses only opensource datasets and computation analysis, hence making it suitable for practical expert decision-support applications with no requirement of actual field experiments or specialized equipment.</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

A multi-dataset crop yield prediction framework using adaptive cooperative metaheuristic optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Prakash Chandra Sharma,
  • Narendra Singh Yadav,
  • Tarun Dhar Diwan,
  • Pankaj Chandra

摘要

Crop yield prediction is an essential element for agricultural planning and policy development, although the results may be less accurate because of the high variability that could exist in soil types, weather patterns, and crop growing dynamics for each region. The existing approaches for crop yield prediction have been hampered by the fact that they have been using limited data sources and rigid hyperparameter optimization techniques, thus inhibiting the ability to cope with the diversified agricultural data. The problem encountered by the existing approaches is addressed by the proposed work. The proposed framework focuses on the comprehensive integration of crop yield statistics, climatic variables, soil property data, and satellite-based vegetation indices acquired from various open-access datasets. This multi-source data integration thus allows capturing complementary information in the models regarding crop growth, which is not possible in any of the single datasets alone. To determine the optimized parameter settings for the models, a novel adaptive cooperative PSO–DE–Jaya (AC-PDJ) metaheuristic is proposed by hybridizing the global searching capability of PSO, the variation mechanism of DE, and the exploitation strength of the Jaya algorithm. The cooperation among the participating strategies will be adaptively regulated to enhance the convergence with better preservation of solution diversity. The proposed methodology will be evaluated using various machine learning and deep learning models and compared against various state-of-the-art optimization techniques. Experimental results on different crops in various geographical locations have demonstrated that AC-PDJ-optimized models result in lower prediction errors and improved goodness-of-fit with faster and more stable convergence performance. The framework uses only opensource datasets and computation analysis, hence making it suitable for practical expert decision-support applications with no requirement of actual field experiments or specialized equipment.