基于多光谱影像与模糊控制的精量灌溉决策模型研究

Decision-making Model of Precision Irrigation Based on Multispectral Image and Fuzzy Control

摘要:

为解决干旱地区农业灌溉精准度不高、灌溉决策智能化不足的问题,本研究构建基于多光谱影像与宽度学习的土壤含水率反演模型,并引入数据增强技术提高模型可靠性,在此基础上,利用多尺度土壤含水率反演结果计算水分亏缺率,生成空间差异化灌溉处方图,进而建立以水分亏缺率与有效积温为双输入参量的模糊控制器,构建基于多光谱和模糊控制深度融合的精量灌溉决策模型。结果表明:宽度学习算法(Broad learning system,BLS)结合数据增强技术显著提升反演模型性能,原始建模集决定系数为0.8596,均方根误差RMSE为0.0378g/g,平均绝对误差MAE为0.0175g/g;测试集决定系数为0.8258,均方根误差RMSE为0.0458g/g,平均绝对误差MAE为0.0181g/g;经数据增强处理后,测试集R2p提升8.1% ,同时测试集RMSE下降47.6% ,MAE下降7.7%。精量灌溉决策模型通过融合水分亏缺率与积温参数构建的模糊控制器,实现灌溉量动态优化;Pearson相关性分析表明,水分亏缺率与灌溉量在全生育期呈现显著正相关,Pearson相关系数均高于0.82,验证了规则库的合理性及空间时间适配性。同时,误差传播敏感度分析揭示精量灌溉模型具备智能分级响应特性,在高水分亏缺区域,敏感度显著降低,驱动灌溉决策形成抗扰动区,通过持续性高量级灌溉抵抗水分胁迫风险;在低水分亏缺区,敏感度增强触发灌溉量动态微调,实现资源精准优化;灌溉决策结果显示,模型输出的灌溉量随积温升高呈现规律性变化,平均灌溉量从生长初期的1.8mm增加至旺盛生长期的3.8mm,中等灌溉策略占比从1%提升至77.9% ,且各时期均保持显著的空间变异性,实现了时空协同的精量施灌。本研究结果可为干旱区精准灌溉与智能灌溉系统的优化设计提供理论支撑。

Abstract:

In order to solve the problems of low precision of agricultural irrigation and insufficient intelligence of irrigation decision-making in arid areas, an inversion model of soil moisture content was constructed based on multi-spectral image and broad learning, and data enhancement technology was introduced to improve the reliability of the model. On this basis, the water deficit rate was calculated by using the inversion results of multi-scale soil moisture content, and the spatial differential irrigation prescription map was generated. Then, a fuzzy controller with water deficit rate and effective accumulated temperature as double input parameters was established, and a precision irrigation decision-making model based on deep fusion of multi-spectral and fuzzy control was constructed. The results showed that the broad learning system combined with data enhancement technology significantly improved the performance of the inversion model. The determination coefficient of the original modeling set was 0.8596, the root mean square error (RMSE) was 0.0378g/g, and the average absolute error (MAE) was 0.0175g/g. The determination coefficient of the test set was 0.8258, the RMSE was 0.0458g/g, and the MAE was 0.0181g/g. After data augmentation, the test set R2 p was increased by 8.1% , and the test set RMSE was decreased by 47.6% , MAE was decreased by 7.7% . The precision irrigation decision model realized the dynamic optimization of irrigation amount by integrating the fuzzy controller constructed by water deficit rate and accumulated temperature parameters. Pearson correlation analysis showed that there was a significant positive correlation between water deficit rate and irrigation amount during the whole growth period, and the Pearson correlation coefficient was higher than 0.82, which verified the rationality and space-time adaptability of the rule base. At the same time, the error propagation sensitivity analysis revealed that the precision irrigation model had intelligent hierarchical response characteristics. In the high water deficit area, the sensitivity was significantly reduced, which drived the irrigation decision to form an anti-disturbance area, and resisted the risk of water stress through continuous high-order irrigation. In the low water deficit area, the sensitivity enhancement triggered the dynamic fine-tuning of irrigation amount to achieve accurate optimization of resources. The results of irrigation decision-making showed that the irrigation amount output by the model changed regularly with the increase of accumulated temperature. The average irrigation amount was increased from 1.8mm in the early growth period to 3.8mm in the vigorous growth period, and the proportion of medium irrigation strategies was increased from 1% to 77.9% , and significant spatial variability was maintained in each period. The results can provide theoretical support for the optimal design of precision irrigation and intelligent irrigation systems in arid areas.