基于改进的PSO-BP神经网络的参考作物腾发量预测

针对传统PSO算法易陷入局部极值的缺点,提出了一种改进的粒子群算法(ADAPPSO)。该算法通过利用表现为非线性递减特性的自适应惯性权重来有效减少PSO算法在运算过程中出现局部极值的情况。利用ADAPPSO算法对BP神经网络所涉及的一系列参数进行优化,并在此基础上建立ADAPPSO-BP模型对参考作物腾发量(ET-0)进行预测。以商丘地区数据为例,通过平均影响值法(MIV)对变量进行筛选,并在此基础上建立了BP神经网络、PSO-BP和ADAPPSO-BP 3种预测模型。根据相关实验结果表明,BP模型、PSO-BP模型和ADAPPSO-BP模型的决定系数R2分别为0.8983、0.9527和0.9606,3种模型的平均绝对误差MAE分别为0.3558、0.2401和0.2056。3种模型中,ADAPPSO-BP模型的R2值最大,MAE最小,这表明提出的ADAPPSO-BP模型能够有效地提高ET-0的预测能力。

Reference Crop Evapotranspiration Prediction Based on Improved PSO-BP Neural Network

Aiming at the shortcoming that the traditional PSO algorithm is easy to fall into local extreme value,this article proposes an improved particle swarm algorithm(ADAPPSO).This algorithm effectively reduces the occurrence of local extreme values during the operation of the PSO algorithm by using adaptive inertial weights that exhibit a nonlinear decreasing characteristic.The ADAPPSO algorithm was used to optimize a series of parameters involved in the BP neural network,and based on this,an ADAPPSO-BP model was established to predict the reference crop evapotranspiration(ET0).Taking the data in Shangqiu area as an example,the variables were screened by the mean influence value method(MIV),and three prediction models were established based on BP neural network,PSO-BP and ADAPPSO-BP.According to relevant experimental results,the determination coefficients R2 of the BP model,the PSO-BP model,and the ADAPPSO-BP model were 89.83%,95.27%,and 96.06%,and the average absolute errors MAE of the three models were 0.3558,0.2401,and 0.2056.Among the three models,the R2 of the ADAPPSO-BP model was the largest and the MAE was the smallest,which indicated that the proposed ADAPPSO-BP model could effectively improve the prediction ability of ET0.