基于VMD-Transformer-ECM模型的空气中有害气体浓度预测  

Prediction of harmful gas concentrations in air based on VMD−Transformer−ECM model

作　　者：张子煜 刘浩哲 陈娟[1] ZHANG ZiYu;LIU HaoZhe;CHEN Juan(College of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029;Beijing Institute of Telemetry Technology,Beijing 100076,China)

机构地区：[1]北京化工大学信息科学与技术学院,北京100029 [2]北京遥测技术研究所,北京100076

出　　处：《北京化工大学学报（自然科学版）》2024年第6期102-111,共10页Journal of Beijing University of Chemical Technology(Natural Science Edition)

基　　金：国家自然科学基金(60971019)。

摘　　要：空气中有害气体的浓度序列具有较强的复杂性、非线性及波动性,为气体浓度的准确预测带来了很大挑战。针对该问题,提出了一种基于变分模态分解(VMD)和误差补偿(ECM)的Transformer预测模型(VMD-Transformer-ECM)。首先通过VMD将气体浓度时间序列分解成不同频率的本征模态函数(IMF),以降低预测模型输入的复杂性和非平稳性;然后通过Transformer模型对分解所得的各模态分量进行预测,并对预测结果进行重构以得到初步预测值;最后通过ECM模型对误差序列进行预测,并使用误差预测值来补偿初步预测值,以进一步提高模型的预测精度。在不同数据集中对所提模型进行了验证,结果显示,与其他模型相比,VMD-Transformer-ECM模型对CO_(2)及其他有害气体浓度预测的平均绝对百分比误差(MAPE)和均方根误差(RMSE)最小,决定系数(R^(2))最大,其中在预测步长为3 h时,本模型对CO_(2)浓度预测的MAPE为4.38%,RMSE为35.44×10^(-6),R^(2)为0.94,表明所提模型的预测精度较高,预测性能较好。The concentration sequence of harmful gases in the air has high complexity,nonlinearity and volatility,which results in the accurate prediction of gas concentration being very challenging.In view of this,a Transformer prediction model(VMD-Transformer-ECM)based on variational mode decomposition(VMD)and an error compensation model(ECM)is proposed.Firstly,the gas concentration time series is decomposed into intrinsic mode functions(IMF)with different frequencies by VMD to reduce the complexity and non‑stationarity of the prediction model input.The Transformer model is then used to predict the modal components obtained by decomposition,and the prediction results are reconstructed to obtain the preliminary prediction value.Finally,the error sequence is predicted by ECM,and the error prediction value is used to compensate the preliminary prediction value,so as to further improve the prediction accuracy of the model.The proposed model was validated using different datasets.The results show that compared with other models,the VMD-Transformer-ECM model has the smallest mean absolute percentage error(MAPE)and root mean square error(RMSE)and the largest determination coefficient R^(2) for the prediction of CO_(2) and other harmful gas concentrations.When the prediction step is 3 h,the MAPE of the model for CO_(2) concentration prediction is 4.38%,RMSE is 35.44×10^(-6),and R^(2) is 0.94.Our proposed model has higher prediction accuracy and better prediction performance than other models currently employed.

关 键 词：有害气体 Transformer 变分模态分解(VMD) 误差补偿(ECM) 气体浓度预测 

分 类 号：TN911.7[电子电信—通信与信息系统]