基于卡尔曼滤波融合算法的空气质量指数预测  

Prediction of air quality index based on Kalman filtering fusion algorithm

作　　者：郭利进[1,2] 井海明[1,2] 南亚翔 修春波[1,2] GUO Lijin JING Haiming NAN Yaxiang X IU Chunbo(College of Electrical Engineering and Automation, Tianj in Polytechnic University, Tianjin 300387 Laboratory of New Technology about Energy and Electrical Engineering, Tianjin Polytechnic University, Tianjin 300387)

机构地区：[1]天津工业大学电气工程与自动化学院,天津300387 [2]天津工业大学电工电能新技术天津市重点实验室,天津300387

出　　处：《环境污染与防治》2017年第4期388-391,共4页Environmental Pollution & Control

基　　金：国家自然科学基金资助项目(No.61203302);天津市应用基础与前沿技术研究计划项目(No.14JCYBJC18900)

摘　　要：分析了卡尔曼滤波算法的基本原理及其对空气质量指数(AQI)的预测机制。利用自回归滑动平均模型(ARMA)为卡尔曼滤波建立数学模型,提出了将径向基函数(RBF)神经网络融合于卡尔曼滤波,实现了新的融合算法对AQI进行预测。根据AQI时间序列的特点,建立了自回归预测模型,进而建立卡尔曼滤波的状态方程和测量方程。采用随机梯度逼近训练算法训练RBF神经网络,用RBF神经网络的输出作为卡尔曼滤波测量方程的观测值。仿真结果表明,融合了RBF神经网络后的卡尔曼滤波预测算法改善了单一方法预测滞后的现象,减小了误差,提高了预测精度。The basic principle of the Kalman filtering algorithm and its prediction mechanism of air quality index（AQI）were analyzed.This paper using auto-regressive and moving average model（ARMA）to establish a mathematical model for Kalman filtering and put forward radial basis function（RBF）neural network merging with Kalman filtering to achieve a new fusion algorithm for AQI forecast.According to the characteristics of AQI time series,auto-regressive prediction model was established,then Kalman filtering state equation and measurement equation were established.The stochastic gradient approximation algorithm was used to train the RBF neural network,and the output of RBF neural network was used as the observation value of the Kalman filtering measurement equation.Simulation results showed that compared with a single method,the Kalman filtering prediction algorithm combined with RBF neural network had improved the lag phenomenon,reduced errors and raised the prediction accuracy.

关 键 词：卡尔曼滤波 空气质量指数 自回归滑动平均模型 径向基函数 

分 类 号：X831[环境科学与工程—环境工程]