Prediction of Chongqing's grain output based on support vector machine

Jia Wang, Guixian Tian, Yongchao Tao, Chengwu LuFrontiers in Sustainable Food Systems 2023 · DOI

Abstract

Scientific prediction of agricultural food production plays an essential role in stabilizing food supply. In order to improve the accuracy of grain yield prediction and reduce the error of grain yield prediction in Chongqing, this paper proposes a new method for the grain yield prediction in Chongqing by using support vector machine (SVM). In this paper, based on the support vector regression structure, the support vector regression algorithm is designed, and then the support vector machine is adopted in the replacement of the error back propagation process in BP neural network. The results of case analysis show that the method based on support vector machine can effectively reduce the error of grain yield prediction.

Introduction

Predicting agricultural output is a cornerstone of national food security, particularly for nations like China with vast populations. Chongqing, a critical agricultural hub in the upper reaches of the Yangtze River, faces complex challenges including rapid urbanization, shrinking arable land, and unpredictable climate change impacts.

Traditional forecasting methods, such as econometric models, often struggle with non-linear variables. Standard Back-Propagation (BP) neural networks, while powerful, frequently fall into "local optima," settling on suboptimal solutions during training. This research addresses these limitations by integrating Support Vector Machines (SVM) with neural network structures to create a high-precision forecasting model for grain production in Chongqing.

Methodology

The core innovation is an adaptive learning support degree regression algorithm. Instead of traditional gradient-descent-based back-propagation, the model utilizes Support Vector Regression (SVR) to determine optimal weights and thresholds.

Technical Workflow

The prediction process follows a structured sequence as shown in the flowchart below:

  1. Normalized Processing: Grain yield data is normalized to ensure consistency.
  2. Dimension Reduction: Partial correlation coefficients filter out non-essential attributes, focusing the model on high-impact variables.
  3. Structural Risk Minimization: Unlike models that minimize only training error (empirical risk), SVM minimizes "structural risk" based on VC dimension theory, enhancing generalization to unseen data.
  4. Optimization: Network parameters are treated as "particles" optimized via a fitness function based on mean square error to reach a global optimum.
Schematic diagram of the grain yield prediction process using SVR.

Data Sources

The study utilized historical agricultural data from Chongqing (2015–2019), focusing on total grain output (measured in 10,000 tons) and crop-specific yields for rice, wheat, and corn. Data was sourced from regional statistical yearbooks and established academic benchmarks.

Results and Discussion

The SVM-based model demonstrated exceptional fit and stability. By replacing the standard BP process, the model avoids the "local optimum" trap, leading to a more accurate mapping of inputs—such as fertilizer usage and climate factors—to actual output.

Performance Metrics

The model achieved significant precision across all categories:

  • Total Grain: The average residual error was a mere 0.46%, with relative errors consistently below 1% across the five-year test period.
  • Crop-Specific Accuracy: Rice predictions achieved the highest precision (errors < 0.7%), while wheat and corn remained highly accurate with errors under 3.5% and 3.1%, respectively.
Comparison between actual and predicted grain output values from 2015 to 2019.

The model's performance significantly outperformed standard approaches like the Adaptive Neuro-Fuzzy Inference System (ANFIS) and the Weighted K-Nearest Neighbor (WKNN) algorithm, showing a much tighter fit to historical yield curves.

Conclusion

The integration of SVM into grain yield prediction offers a robust solution for "small sample" problems typical in agriculture, where deep learning might otherwise overfit. While effective, future work should address the computational complexity of SVMs for larger datasets and provide theoretical guidance for automated kernel selection. Expanding the data scope to include real-time remote sensing and granular meteorological variables will further enhance early warning capabilities for food supply chains.

Summary updated

Abstract

Scientific prediction of agricultural food production plays an essential role in stabilizing food supply. In order to improve the accuracy of grain yield prediction and reduce the error of grain yield prediction in Chongqing, this paper proposes a new method for the grain yield prediction in Chongqing by using support vector machine (SVM). In this paper, based on the support vector regression structure, the support vector regression algorithm is designed, and then the support vector machine is adopted in the replacement of the error back propagation process in BP neural network. The results of case analysis show that the method based on support vector machine can effectively reduce the error of grain yield prediction.
BibTeX
@inproceedings{wang2023prediction,
  author = {Jia Wang and Guixian Tian and Yongchao Tao and Chengwu Lu},
  title = {{Prediction of Chongqing's grain output based on support vector machine}},
  booktitle = {Frontiers in Sustainable Food Systems},
  year = {2023},
  doi = {10.3389/fsufs.2023.1015016},
}
Figure