摘要
针对前馈神经网络预测粮食产量的方法易陷入局部最优的问题,提出一种基于粒子群算法和人工蜂群算法的改进BP神经网络模型.利用粒子群优化算法和人工蜂群算法在全局搜索能力上的不同优势,结合两者对BP神经网络的权值和阈值进一步优化,以提升粮食产量预测模型的准确性与鲁棒性.给出基于粒子群和人工蜂群混合的ABPSO算法的具体实现,并选择1979年至2012年我国粮食的产量及影响其产量的8项因素作为数据集进行试验.结果表明:改进的BP神经网络能够较好地预测国内近几年的粮食产量变化趋势;相比未优化的BP模型,新算法预测误差平均值由847 780 t降低至240 320 t,误差范围由1 894 200 t降低至586 800 t.
To solve the problem of feedforward neural network for predicting grain yield with easy falling into local optimum,an improved BP neural network model was proposed based on particle swarm optimization and artificial bee colony algorithm. According to the different advantages of particle swarm optimization algorithm and artificial bee colony algorithm in global search ability,the weight and the threshold of BP neural network were further optimized to improve the accuracy and robustness of grain yield prediction model. The specific implementation of artificial bee particle swarm optimization (ABPSO) algorithm was given based on particle swarm and artificial bee colony. The eight factors on Chinese grain yield and the yields from 1979 to 2012 were selected as data sets. The results show that the trend of grain yield in China in recent years can be better predicted by the improved BP neural network. Compared with unimproved BP model,the average prediction error of the new algorithm is decreased from 847 780 t to 240 320 t,and the error range is decreased from 1 894 200 t to 586 800 t.
作者
庄星
韩飞
ZHUANG Xing;HAN Fei(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang,Jiangsu 212013,China)
出处
《江苏大学学报（自然科学版）》
EI
CAS
北大核心
2019年第2期209-215,共7页
Journal of Jiangsu University：Natural Science Edition
基金
国家自然科学基金资助项目(61572241
61271385)
国家重点研发计划项目(2017YFC0806600)
江苏省"六大人才高峰计划"高层次人才计划项目(2015-DZXX-024)
江苏省"333工程"第三层次人才计划项目((2016)III-0845)
关键词
粮食产量
预测
BP神经网络
粒子群
人工蜂群
混合群智能
grain yield
prediction
BP neural network
particle swarm
artificial bee colony
hybrid swarm intelligence
