基于BP网络的空气污染指数预报研究

Forecast of air pollution index based on BP neural network

为了研究对空气污染进行预报的更有效方法,从气象学角度对各种污染物和气象因素进行了分析,在确定了空气污染指数影响因子基础上,综合考虑BP网络的逼近能力和泛化能力,提出了空气污染指数BP网络预报模型,并进行了对比预报检验.结果表明,BP方法预报模型的预报准确度明显高于通常使用的逐步回归方法,特别是对骤升骤降趋势也能得到准确度较高的预报结果.

According to meteorology, the affecting factors which influence the air pollution index are chosen by analyzing several kinds of contaminations and climatic factors. The optimal network structure is determined by making an integrated survey of the approximation capability and the generalization of the network. The fitting and forecasting results indicate that compared with the normal step wise regress method, the forecast precision with the BP method is improved. Moreover, we can gain a highly precise forecast result in the trend of increasing dramatically and declining sharply. The experiment results show that this method can be put into practical use.