摘要
气候问题是全人类始终关注的问题,因此有必要对其进行预测研究。本文采用广州市黄埔区的气象数据作为研究对象,分析的数据包括当地的风速、平均气温、降水量等多个不同指标。以气温(即气候温度,单位℃)作为因变量,其余变量视作影响气温的自变量。在数据预处理方面,选择主成分分析方法(PCA)对样本进行降维处理,从中选出影响权重最大的自变量,以筛选过后的数据作为基础,采用回声状态网络预测气温的变化。实验结果表明,预测值和实际值的总体偏差较小,均保持在较低水平,绝对预测误差普遍处于15%以下,预测效果良好。采用对比模型(LSTM模型和ARIMA模型)对同样的数据进行预测。在实验中,我们发现对比模型的预测效果并不理想,在均方误差(MSE)的计算中,预测值和实际值的拟合精度较低。验证了ESN预测效果的优越性。
Climate issues are a constant concern for all mankind and therefore prediction studies are necessary.This paper uses meteorological data from Huangpu District,Guangzhou City,as the subject of the study.The data analyzed include a number of different indicators such as local wind speed,average temperature and precipitation.Air temperature(i.e.climatic temperature in °C) is used as the dependent variable and the remaining variables are considered as independent variables affecting air temperature.In terms of data pre-processing,Principal Component Analysis(PCA) was chosen to reduce the dimensionality of the samples,from which the independent variables with the greatest influence weights were selected,and the filtered data were used as the basis for predicting the changes in temperature using an echo state network.The experimental results show that the overall deviation between the predicted and actual values is small and remains low,with the absolute prediction error generally below 15% and the prediction effect is good.A comparison model(LSTM model and ARIMA model) was used to predict the same data.In our experiments,we found that the prediction results of the comparison models were not satisfactory,and the accuracy of fitting the predicted and actual values was low,as can be seen from the calculation of the mean squared error(MSE).The superiority of the prediction effect of ESN is verified.
作者
王人杰
刘海忠
朱洋
Wang Renjie;Liu Haizhong;Zhu Yang(School of Mathematics and Physics,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《科学技术创新》
2022年第19期76-79,共4页
Scientific and Technological Innovation
关键词
气温预测
主成分分析
回声状态网络
绝对预测误差
均方误差
Temperature prediction
Principal component analysis
Echo state network
Absolute prediction error
Mean square error
