中文摘要：

准确预测PM2．5浓度对人类生产生活具有重要意义，由于影响PM2．5浓度因素的复杂性和相关数据获取的不易性，使预测较困难。针对这些问题，提出了基于改进局域平均预测法的PM2．5浓度预测方法。该方法可避免上述问题，达到预测的目的。研究中首先采用C—C法得到PM2．5浓度时间序列的嵌入维和延迟时间并进行相空间重构，然后针对邻近相点选取问题这一难点，提出了基于邻近轨道的邻近相点选取方法。该方法能在一定程度上避免引入伪邻近相点。在回溯搜索优化算法的基础上，提出整数回溯搜索优化算法，并将其应用于邻近相点个数的确定。选取北京市2013--2015年PN2．5浓度数据进行仿真预测，结果表明改进后的局域平均预测法性能优于原算法，可短期实时预报PM2．5的浓度。

英文摘要：

Accurate predicting of the concentration of PM2.5 was significant to production and life. However, due to the complexity ot PM2.5 concentration factors and the difficulties of data acquisition, which made the forecasting more difficult. To deal with these problems, prediction model of PM2.5 concentration based on improved average local forecasting model was proposed. This method could avoid the above problems and achieve the purpose of prediction. Firstly, this paper used the C - C method to obtain embedding dimension and delay time, and reconstruct the phase space of time series of PM2.5 concentration. Secondly, a method for select the nearest neighboring points based on nearest neighboring orbits was presented, this method could avoid the introduction of nearest neighboring points. Thirdly, the integer backtracking search optimization algorithm was put forward based on backtracking search optimization algorithm, which applied to determine the number of nearest neighboring points. At last, selected the PM2.5 concentration data from 2013 to 2015 of Beijing for simulation and prediction. The results showed that the improved average local forecasting model was better than the original algorithm. The new model could real- time forecast short- term PM2.5 concentration.