神经网络在空气污染预报中的应用研究

空气污染预报是一项复杂的系统工程，是当今环境科学研究的热点，国内外已有将神经网络法应用于大气污染预报的研究。本论文以PM2．5为例，采用伦敦市PM2．5的小时平均浓度数据，使用传统的BP神经网络建立预报模型，定量预测伦敦市PM2.5的小时平均浓度，探讨了大气污染预报网络的建模过程中，扩大样本集、去除样本集数据噪声和在输入向量中加入气象变量等因素对建模所产生的影响。最后得出结论，适当的选择样本集、气象变量，有利于提高所建立网络模型的预测精度。

Study on Application of Artificial Neural Network in Air Pollution Forecast

As one of the research hotspots in environmental sciences, air pollution forecast is a complex systems engineering. There are researches on applying the artificial neural network （ANN） in the air pollution forecast at home and abroad. As an example, this paper uses the data of hourly concentration of PMz.5 in London and adopts the traditional BP neural networks to build up forecast model to quantitively forecast the hourly concentration of PM2. 5 in London. The impacts of enlarging sample set, removing noise of sample set and adding weather factors into inputting vector on establishment of the model for the air pollution forecast network is discussed. Finally, it comes to the conclusion that properly selecting sample set and weather factors is beneficial to improve forecasting precision of the network model.