基于生成对抗网络模型的小样本PM_(2.5)预测  

A SMALL SAMPLE OF PM_(2.5) PREDICTION BASED ON THE GENERATIVE ADVERSARIAL MODEL

作　　者：汪祖民[1] 张嘉峰 胡玲艳[1] 邹启杰[1] 盖荣丽[1] 刘艳[1] Wang Zumin;Zhang Jiafeng;Hu Lingyan;Zou Qijie;Gai Rongli;Liu Yan(College of Information Engineering,Dalian University,Dalian 116622,Liaoning,China)

机构地区：[1]大连大学信息工程学院,辽宁大连116622

出　　处：《计算机应用与软件》2023年第10期114-119,共6页Computer Applications and Software

基　　金：大连市科技创新基金项目(2020JJ26SN058,2020JJ27SN106)。

摘　　要：针对目前数据驱动的方法在小样本下PM_(2.5)预测准确率较低的问题,提出一种基于生成对抗性网络(GAN)模型PME-GAN,用于在线预测PM_(2.5)浓度值。在生成器中加入长短期记忆网络(LSTM)并用于提取输入数据的时序特征,在判别器中加入多层感知机网络(MLP),通过生成器对PM_(2.5)浓度值进行预测。与LSTM、GRU、CNN-LSTM和CNN-GRU 4种模型进行对比实验,结果表明,该方法在小样本数据集上具有更高的预测准确率,对保定测试集的后25%数据开始预测,预测效果很好。To solve the problem of low PM_(2.5) prediction accuracy of current data-driven methods under small samples,PME-GAN model based on generative adversarial network(GAN) is proposed for online prediction of PM_(2.5) concentration.A long short-term memory network(LSTM) was added to the generator and used to extract the timing characteristics of the input data,and a multi-layer perceptron network(MLP) was added to the discriminator,and the PM_(2.5) concentration value was predicted by the generator.Comparative experiments were carried out with 4 models of LSTM,GRU,CNN-LSTM and CNN-GRU.The experimental results show that this method has higher prediction accuracy on small sample data sets.Starting to predict the last 25% of the Baoding test set,the prediction effect is very well.

关 键 词：小样本 PM_(2.5)预测 生成对抗性网络 博弈 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]