基于CEEMDAN-IGWO-CNN-LSTM空气质量预测建模  

Modeling of Air Quality Prediction Based on CEEMDAN-IGWO-CNN-LSTM

作　　者：朱菊香 任明煜 谷卫 张雯柏 ZHU Ju-xiang;REN Ming-yu;GU Wei;ZHANG Wen-bai(Wuxi College of Rail Transportation,Wuxi Jiangsu 214105,China;Nanjing Institute of Automation,Nanjing University of Information Engineering,Nanjing Jiangsu 210000,China;National Maglev Transportation Engineering Technology Research Center,Tongji University,Shanghai 200092,China)

机构地区：[1]无锡学院(轨道交通),江苏无锡214105 [2]南京信息工程大学自动化学院,江苏南京210000 [3]同济大学国家磁悬浮交通工程技术研究中心,上海200092

出　　处：《计算机仿真》2025年第1期529-537,共9页Computer Simulation

基　　金：国家自然科学基金资助项目(52202473)。

摘　　要：为了提高PM_(2.5)浓度的预测精度,提出了一种基于CEEMDAN-IGWO-CNN-LSTM的混合预测模型。首先,针对原始PM_(2.5)序列不稳定、波动性大的问题,引入自适应噪声的完全集成经验模态分解(CEEMDAN)将原始不平稳的PM_(2.5)序列进行分解,去除噪声数据后得到多个平稳的固有模态分量和残差分量,利用卷积神经网络(CNN)高效的特征提取函数提取数据特征并输入到LSTM网络,其次,引入柯西变异策略、引入非线性函数调整控制参数和增加可变惯性权重改进灰狼算法,减小GWO陷入局部极值的风险,提高灰狼算法的搜索精度和收敛速度。最后,采用改进的灰狼算法(IGWO)优化CNN-LSTM模型的特征参数,最后将各个子序列进行叠加得到最终的预测结果。仿真结果表明,CEEMDAN-IGWO-CNN-LSTM模型具有较高的预测精度和较低的预测误差,CEEMDAN-IGWO-CNN-LSTM模型能够更精确地预测PM_(2.5)的浓度,达到了预期效果。In order to improve the prediction accuracy of PM_(2.5) concentration,a hybrid prediction model based on CEEMDAN-IGWO-CNN-LSTM is proposed in this paper.First,for the problem of unstable and volatile original PM2.5 sequence,the fully integrated empirical modal decomposition with adaptive noise(CEEMDAN)is introduced to decompose the original unstable PM2.5 sequence,and multiple smooth intrinsic modal components and residual components are obtained after removing the noisy data,and the efficient feature extraction function of convolutional neural network(CNN)is used to extract the data features and input to the Secondly,the gray wolf algorithm is improved by introducing the Corsi variation strategy,introducing the nonlinear function to adjust the control parameters and adding variable inertia weights to reduce the risk of GWO falling into local extremes and improve the search accuracy and convergence speed of the gray wolf algorithm.Finally,the improved gray wolf algorithm(IGWO)is used to optimize the feature parameters of the CNN-LSTM model,and finally the individual subsequences are superimposed to obtain the final prediction results.The simulation results show that the CEEMDAN-ICWO-CNN-LSTM model has high prediction accuracy and low prediction error,and the CEEMDAN-ICWO-CNN-LSTM model can predict the PM_(2.5) concentration more accurately and achieve the expected results.

关 键 词：自适应噪声完全经验模态分解 改进灰狼算法 卷积神经网络 长短期记忆网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术] TN98[自动化与计算机技术—计算机科学与技术]