基于BP神经网络的空气污染指数预测模型研究

BP神经网络已成为研究空气污染预测的有效工具之一。文章利用近十年北京市地面气象观测资料和空气污染指数数据,通过BP神经网络技术构建了不同季节的空气污染指数预测模型,对北京市空气污染指数进行了预测。通过相关系数分析法,对比分析了预测结果与实际监测结果,研究结果表明:春、夏、秋、冬季的预测值与监测值线性相关系数分别为0.81、0.84、0.89、0.85。北京春季常伴随有沙尘天气,而文章并没有考虑沙尘天气对预测模型的影响,因此春季BP神经网络预测精度在四季中最低,其预测值与监测值的线性相关系数为0.81。由于秋季不同空气质量级别的数据都有较多分布,因此该季节构建的网络更具有代表性,其预测精度在四季中最高,预测值与监测值的线性相关系数高达0.89。总之,BP神经网络模型对于北京空气污染指数预测是行之有效的。

Forecasting Model of Air Pollution Index Based on BP Neural Network

As a widely used neural network,BP network has a widespread application in air pollution forecast.Based on the analysis of nearly 10 years Beijing meteorological data in the ground and air pollution index data,different season's air pollution index forecasting model was built through BP neural network technology and the air pollution index was predicted.Through the correlation coefficient analysis,the prediction results were compared with practical monitoring results.Results of the analysis showed that linear correlation coefficients of predicted value and monitoring value were 0.81,0.84,0.89 and 0.85 respectively corresponding to spring,summer,autumn and winter.Beijing often accompanies with the dust weather in spring,and influence of dust weather on the model was not considered,so the spring BP neural network prediction accuracy was the lowest in the four seasons which the linear correlation coefficients was 0.81.Due to distributions of data of different air quality levels in autumn,the forecast model for autumn was more representative,with autumn BP neural network prediction accuracy as the highest in the four seasons with linear correlation coefficients of 0.89.BP neural network model for Beijing's air pollution index forecast was effective.