基于遗传优化图神经网络的企业贸易经济预测  

Enterprise Trade Economic Forecast Model Based onGenetic Optimization Graph Neural Networks

作　　者：余先玲 胡乔雅 YU Xianling;HU Qiaoya(School of Economics and Management,Guangzhou Huanan Business College,Guangzhou 510650,China;School of Economics and Management,Foshan University,Foshan 528225,China)

机构地区：[1]广州华南商贸职业学院经济管理学院,广东广州510650 [2]佛山大学经济管理学院,广东佛山528225

出　　处：《常州工学院学报》2024年第4期44-51,共8页Journal of Changzhou Institute of Technology

基　　金：广东省特色创新项目(2021WTSCX308)。

摘　　要：在智能化算法研究的基础上,设计基于遗传优化的图神经网络(GOGNN)模型,用于实现企业贸易经济预测功能。根据经济贸易特征因子,通过引入种群适应度的平均值和适应度值的离散程度来动态调整交叉概率和变异概率。根据种群的进化情况动态调整这些概率,避免模型陷入局部最优解。选择影响贸易经济的评价指标构建经济贸易预测图数据,使用图神经网络对指标因素进行聚合,以预测准确率作为适应度函数的输出值。结合来自国家统计局、国家外汇管理局、行业调查研究等网站的历史外贸经济数据,将提出的GOGNN模型与其他模型做对比。实验结果表明:GOGNN模型对训练数据的均方误差大约为0.74335,小于其他模型;GOGNN模型对测试数据集的预测误差值在0.03以下,且相对于其他模型有着较好的收敛精度。The economic benefits of foreign trade enterprises are influenced by multiple related factors such as product type,policy and tax rate,so it is necessary to make a scientific analysis of the future foreign trade economic forecast.On the basis of intelligent algorithm research,a genetic optimization graph neural network(GOGNN)model is designed to realize the function of enterprise trade economic prediction.According to the characteristic factors of economy and trade,the crossover probability and mutation probability are dynamically adjusted by introducing the average value of population fitness and the dispersion degree of fitness value.According to the evolution of the population,these probabilities are dynamically adjusted to avoid the model falling into the local optimal solution.The evaluation indexes that affect the trade economy are selected to construct the economic and trade forecast graph data,the graph neural network is used to aggregate the index factors,and the forecast accuracy is taken as the output value of the fitness function.Combined with the historical foreign trade economic data from the National Bureau of Statistics,the State Administration of Fo-reign Exchange,industry research and other websites,the proposed GOGNN is compared with the other models.The experimental results show that the mean square error of GOGNN to the training data is about 0.74335,which is smaller than the other models.The prediction error of GOGNN model for test data set is below 0.03,and it has better convergence accuracy than other models.

关 键 词：企业贸易经济 经济预测 图神经网络 遗传优化 遗传算子 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F742[自动化与计算机技术—控制科学与工程]