Scholarlink

<p>{A dynamically adaptive hybrid fuzzy–ANN–SVM architecture with dynamic fuzzy partitioning is developed to address uncertainty and achieve significant improvements in agro-climatic yield forecasting.} Forecasting wheat yields under increasing climatic volatility presents persistent methodological challenges, particularly where abrupt environmental deviations interact nonlinearly with agronomic variables. Models that have been traditionally adopted, such as ARIMA and regression-based models, lack the flexibility in their structure to respond to changing climatic inputs and the complex responses of yields within a short period. This study proposes a hybrid forecasting architecture that combines the fusion of fuzzy time series (FTS) logic with machine learning, specifically Support Vector Machines (SVM) and Artificial Neural Networks (ANN), based on a new dynamic partitioning algorithm. This algorithm re-tunes fuzzy sets using real-time climatic input. This particular innovation involves coupling temporal fuzzification with responsive adjustments of the intervals, which means that the system can redefine climate descriptors, such as heatwaves, rainfall distribution, and soil moisture anomalies, as dynamically evolving constructs rather than fixed classes. These fuzzy-based intervals enter the SVM and ANN layers, and then the model can learn the temporal causality patterns and continue contextualising climatic variables. The ANN component captures multi-dimensional, non-linear patterns, whereas SVM facilitates regularization and generalization in high-dimensional input spaces. The architecture was also experimented with on longitudinal datasets (2010–2024) from Punjab, India, and the Central Valley of California, whose climatic regimes are divergent and predominantly wheat-based. The hybrid model was also superior to the classical models in all climate conditions, particularly in anomalies (e.g., the 2016 North Indian heatwave), exhibiting significant decreases in Mean Squared Error (MSE) and Average Forecast Error (AFER). It also had a strong ability to withstand missing and noisy data, which confirms its field-readiness. The framework has great potential for climate-resilient wheat yield prediction in the agro-climatic regions researched and has relevance in the methodological application to decision support systems in a similar data and climatic setup.</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 dynamically adaptive hybrid fuzzy–ANN–SVM architecture for agro-climatic yield forecasting under uncertainty

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • R. Premalathasri,
  • E. Priyadarshini

摘要

{A dynamically adaptive hybrid fuzzy–ANN–SVM architecture with dynamic fuzzy partitioning is developed to address uncertainty and achieve significant improvements in agro-climatic yield forecasting.} Forecasting wheat yields under increasing climatic volatility presents persistent methodological challenges, particularly where abrupt environmental deviations interact nonlinearly with agronomic variables. Models that have been traditionally adopted, such as ARIMA and regression-based models, lack the flexibility in their structure to respond to changing climatic inputs and the complex responses of yields within a short period. This study proposes a hybrid forecasting architecture that combines the fusion of fuzzy time series (FTS) logic with machine learning, specifically Support Vector Machines (SVM) and Artificial Neural Networks (ANN), based on a new dynamic partitioning algorithm. This algorithm re-tunes fuzzy sets using real-time climatic input. This particular innovation involves coupling temporal fuzzification with responsive adjustments of the intervals, which means that the system can redefine climate descriptors, such as heatwaves, rainfall distribution, and soil moisture anomalies, as dynamically evolving constructs rather than fixed classes. These fuzzy-based intervals enter the SVM and ANN layers, and then the model can learn the temporal causality patterns and continue contextualising climatic variables. The ANN component captures multi-dimensional, non-linear patterns, whereas SVM facilitates regularization and generalization in high-dimensional input spaces. The architecture was also experimented with on longitudinal datasets (2010–2024) from Punjab, India, and the Central Valley of California, whose climatic regimes are divergent and predominantly wheat-based. The hybrid model was also superior to the classical models in all climate conditions, particularly in anomalies (e.g., the 2016 North Indian heatwave), exhibiting significant decreases in Mean Squared Error (MSE) and Average Forecast Error (AFER). It also had a strong ability to withstand missing and noisy data, which confirms its field-readiness. The framework has great potential for climate-resilient wheat yield prediction in the agro-climatic regions researched and has relevance in the methodological application to decision support systems in a similar data and climatic setup.