基于PSO Hammerstein模型的PM2.5预报

发布时间：2018-03-24 22:38

本文选题：PM2.5　切入点：Hammerstein模型　出处：《宁波大学》2017年硕士论文

【摘要】：随着人类生活水平的不断提高和科学技术的不断进步,工业生产和生活造成的大气污染愈发严重。近年来,雾霾天气对人体健康产生了巨大威胁,而导致雾霾天气的首要污染物就是PM2.5。因此,对PM2.5的监测和预报研究显得尤为重要。本文通过观察宁波市2013至2014年全年PM2.5的日平均浓度实际观测值变化曲线发现,其变化趋势大致呈周期性,且以一年为周期。因此,本文将以2013年除PM2.5实际观测值以外的全年空气质量指标作为样本数据,建立数学模型,对某天的PM2.5平均浓度值进行预测。文中首先对样本数据进行归一化处理和主成分分析(Principal Component Analysis,PCA),将实际观测的13维数据降至6维,大大简化了系统复杂度;其次将预处理后的6维数据作为输入,PM2.5预测值作为输出,建立多输入单输出ARMA模型,对空气中PM2.5的日平均浓度进行初步预测,并通过将预测结果的残差平方和、绝对误差、相对误差等作为目标函数,来检验预测精度;由于实际问题通常具有非线性特性,尝试将ARMA模型的输入中加入非线性环节来构建Hammerstein模型发现,预测精度由ARMA模型的0.3028左右提高到0.1910左右,预测效果有显著提高。群智能优化算法是近些年处理极值问题或优化问题时较为先进且高效的方法,其中粒子群优化(Particle Swarm Optimization,PSO)算法思想简单易实现,是进行模型辨识的首选算法。本课题中模型阶数的确定和参数的估计都是较为复杂的辨识过程,利用PSO算法可以较为高效地解决这些问题。但在实际应用时发现,传统的PSO算法收敛速度较慢,且预测精度不是很高,因此本文对传统PSO算法进行改进,有效地提高了收敛速度和预测精度。针对PM2.5建模的具体实现架构和思想是本文的新意所在。
[Abstract]:With the continuous improvement of human living standard and the progress of science and technology, the air pollution caused by industrial production and living is becoming more and more serious. In recent years, haze weather has posed a great threat to human health. PM2.5is the main pollutant causing haze weather. Therefore, it is very important to study the monitoring and forecasting of PM2.5. By observing the daily average concentration of PM2.5 in Ningbo from 2013 to 2014, it is found that, The variation trend is generally periodic and takes one year as the cycle. Therefore, the annual air quality index, other than the actual PM2.5 observations, will be taken as the sample data in this paper, and the mathematical model will be established. The average concentration of PM2.5 is predicted. Firstly, the sample data are normalized and principal component analysis (PCA) is used to reduce the observed 13 dimensional data to 6 dimension, which greatly simplifies the system complexity. Secondly, the pretreated 6-D data is taken as the input PM2.5 prediction value as the output, and the multi-input single-output ARMA model is established. The daily average concentration of PM2.5 in air is preliminarily predicted, and the absolute error is calculated by the sum of squared residuals of the predicted results. The relative error is used as the objective function to test the prediction accuracy. Because the practical problems are usually nonlinear, the nonlinear link is added to the input of the ARMA model to construct the Hammerstein model. The prediction accuracy is increased from about 0.3028 to 0.1910 of ARMA model, and the prediction effect is improved significantly. Swarm intelligence optimization algorithm is an advanced and efficient method in dealing with extreme value problem or optimization problem in recent years. Particle Swarm Optimization (PSO) algorithm is simple and easy to realize, and is the first choice for model identification. In this paper, the determination of model order and the estimation of parameters are more complex identification processes. These problems can be solved efficiently by using PSO algorithm, but in practical application, it is found that the convergence speed of traditional PSO algorithm is slow and the prediction accuracy is not very high. Therefore, the traditional PSO algorithm is improved in this paper. The convergence speed and prediction accuracy are improved effectively. The concrete implementation framework and idea of PM2.5 modeling is the new idea of this paper.
【学位授予单位】：宁波大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X513;TP18

【参考文献】