论文部分内容阅读
针对常用的回归和神经网络作物建模方法存在的输出单一、参数优化困难和预测精度不足等问题,利用基因表达式编程优异的全局搜索能力和RBF神经网络多输出任意非线性函数逼近特点,设计了1种GEP优化的RBF多输出模型算法GEP-RBF。以水稻和番茄的5个关键环境因子为输入、以叶片CO_2交换率和蒸腾速率为输出,进行建模验证。结果显示,在预测的均方根误差指标上,GEP-RBF模型与GA-RBF和RBF相比,水稻的CO_2交换率和蒸腾速率分别降低了约28.4%、38.0%和89.9%、62.8%,番茄的CO_2交换率和蒸腾速率则分别降低了约56.9%、48.4%和75.3%、67.1%;在多输出结果的平衡性指标上,相比GA-RBF和RBF,GEP-RBF模型提高了约16.4%~77.4%。结果表明,GEP-RBF模型具有良好的预测精度和多输出平衡性,是一种有效的作物生长建模方法。
Aiming at the problems of single output, parameter optimization difficulty and poor prediction accuracy of crop modeling methods commonly used in regression and neural networks, the global search ability of gene expression programming and the approximation of RBF neural network with multi-output arbitrary nonlinear function approximation are designed. A GEP-optimized RBF multi-output model algorithm GEP-RBF is proposed. Five key environmental factors of rice and tomato were taken as input, the leaf CO 2 exchange rate and transpiration rate were output, and the model was validated. The results showed that compared with GA-RBF and RBF, the CO 2 exchange rate and transpiration rate of rice decreased by 28.4%, 38.0%, 89.9% and 62.8%, respectively, in the prediction of root mean square error (RMSE) The CO 2 exchange rate and transpiration rate of tomato decreased by 56.9%, 48.4% and 75.3%, 67.1% respectively. Compared with GA-RBF and RBF, the GEP-RBF model improved about about the balance of multiple output 16.4% ~ 77.4%. The results show that GEP-RBF model is an effective crop growth modeling method with good prediction accuracy and multiple output balance.