基于二次分解、LSTM-ELM和误差修正的空气质量指数预测模型  

Air quality index prediction model utilizing quadratic decomposition,LSTM-ELM,and error correction techniques

作　　者：周建国[1] 秦远 周路明 ZHOU Jianguo;QIN Yuan;ZHOU Luming(Department of Economics and Management,North China Electric Power University,Baoding 071003,Hebei,China)

机构地区：[1]华北电力大学经济管理系,河北保定071003

出　　处：《安全与环境学报》2025年第1期322-334,共13页Journal of Safety and Environment

摘　　要：精准预测空气质量指数(Air Quality Index,AQI)对于制定有效的空气污染治理策略至关重要。为了进一步提升AQI的预测精度,提出了一种新的预测模型,并结合了二次分解(Secondary Decomposition,SD)、优化算法、双尺度预测和误差修正的方法。首先,采用改良的自适应白噪声完全集合经验模态分解(Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,ICEEMDAN)和样本熵(Sample Entropy,SE)对原始AQI序列进行分解并重构,获得高频、中频和低频3个频率分量。其次,利用经过北方苍鹰算法(Northern Goshawk Optimization,NGO)优化的变分模态分解(Variational Mode Decomposition,VMD)对高频分量进行二次分解,进一步降低其复杂度。再次,引入向量加权平均算法(Weighed Mean of Vectors Algorithm,INFO)对长短期记忆网络(Long Short-Term Memory,LSTM)和极限学习机(Extreme Learning Machine,ELM)的关键参数进行优化,同时利用INFO-LSTM预测高频分量分解后的子序列,进而利用INFO-ELM分别预测中、低频分量,并将所得预测结果进行线性叠加。最后,利用NGO-VMD和INFO-ELM对误差序列进行分解和预测,并对初次预测结果进行修正,得到最终的AQI预测值。研究选取北京、上海和成都3个典型城市为例进行实证分析,并对比了7个对照试验,发现基于二次分解、LSTM-ELM和误差修正的模型具有最高的预测精度。该模型可为治理空气污染提供理论和技术上的帮助。To enhance the prediction accuracy of the Air Quality Index(AQI),this study proposes a novel prediction model that integrates Secondary Decomposition(SD),an optimization algorithm,dual-scale prediction,and error correction techniques.Initially,the study utilized the Enhanced Intrinsic Mode Function Ensemble Empirical Mode Decomposition with Adaptive Noise(ICEEMDAN)and Sample Entropy(SE)method to decompose and reconstruct the original AQI sequence.This approach yielded three distinct frequency components:high-frequency,mid-frequency,and low-frequency.Subsequently,the research employed the Northern Goshawk Optimization(NGO)algorithm to enhance Variational Mode Decomposition(NGO-VMD),further decomposing the high-frequency component and thereby reducing its complexity.This study optimized the key hyperparameters of Long Short-Term Memory(LSTM)and Extreme Learning Machine(ELM)using the Vector Weighted Mean(INFO)algorithm.Predictions were then performed on the high-frequency sub-sequence using INFO-LSTM,while INFO-ELM was employed to forecast the mid-frequency and low-frequency components.After completing these stages,we combined the predicted results through linear aggregation.Subsequently,we further decomposed and predicted the error sequence using NGO-VMD and INFO-ELM.By adjusting the initial predictions based on this analysis,we derived the final predicted AQI.As part of the empirical analysis,we selected Beijing,Shanghai,and Chengdu for comparison against seven control experiments.The results demonstrated that the proposed model achieved the highest prediction accuracy.Additionally,the secondary decomposition algorithm employed reduces the complexity of the original data,thereby shortening the duration and easing the prediction process.The application of Sample Entropy in reconstructing components helps minimize the accumulation of prediction errors.Furthermore,the dual-scale prediction approach effectively addresses different frequency components,thereby enhancing overall prediction accuracy.The error correction

关 键 词：环境工程学 空气质量指数预测 二次分解 长短期记忆网络 极限学习机 向量加权平均算法 误差修正模型 

分 类 号：X513[环境科学与工程—环境工程]