耦合模拟退火优化最小二乘支持向量机的日参照蒸散量模拟计算

Simulation of Daily Reference Evapotranspiration Based on Least Squares Support Vector Machine Optimized by Coupled Simulated Annealing

针对传统最小二乘支持向量机模型的训练速度慢、不易在线训练、计算量大及参数选择困难等缺陷,提出采用耦合模拟退火优化最小二乘支持向量机算法,建立基于耦合模拟退火优化最小二乘支持向量机的参照作物蒸散量预测模型。选取陕西省的榆林、安康和西安气象站监测的1971-2014年气象资料进行模型训练、测试与验证,研究气象监测获取原始数据作为网络输入,参照蒸散量ET0为输出,构建CSA-LSSVM预测模型,并将CSA-LS-SVM预测结果与LSSVM模型及经典ET0模型模拟计算结果进行比较。结果表明,CSA-LS-SVM模型模拟计算精度和总ET0模拟模型都优于LSSVM模型及其他经典模型模拟结果。该研究CSA-LS-SVM模型为陕西地区气象资料缺乏情况下ET0精确计算提供科学依据,为作物需水量的智能决策提供参考。

According to the traditional Least squares support vector machine had defects, which mainly include training with a slow speed, the data sample, and selecting parameter difficulty, the prediction model of CSA-LS-SVM of daily reference crop evapotrans- piration was examined in the paper. In the current study, the applicability of CSA-LS-SVM in ETo modeling was assessed using the original meteorological data of 1971--2014 in Yulin, Ankang and Xi＇an of Shaanxi, China, which put that eight parameters as input, the reference evapotranspiration values as output. In addition, LSSVM, Hargreaves, Priestley-Taylor models and the CSA-LS-SVM model were tested against the FAO-56 PM model in terms of performance by using commonly used criteria. Experimental results demonstrate that the performed of CSA-LS-SVM model better than the model of ELM, also significantly better than other empirical model. Furthermore, the same result of the total ETo estimation. The research could provide a reference to accurate ETo estimation and give a reference of crop water requirement of the intelligent irrigation decision in Shaanxi.