基于BP神经网络的佛山空气质量预报模型的研究  

Study on model for predicting air quality in Foshan based on BP neural network

作　　者：刘永红[1] 谢敏[1] 蔡铭[1] 李璐[1] 

机构地区：[1]中山大学工学院智能交通研究中心,广东省智能交通重点实验室,广州510275

出　　处：《安全与环境学报》2011年第2期125-130,共6页Journal of Safety and Environment

基　　金：广东省重大科技专项(2008A080800012)

摘　　要：提出非数值的佛山惠景城站点空气质量预报模型,将BP神经网络引入到预报模型中,以降雨量、风速、风向、温度、湿度和云量等气象参数和前1 d污染物浓度为模型输入参数,建立了结构为7-7-1的非季节预报模型和夏季预报模型。结果表明,夏季模型无论在模型检验还是在实际预报精度方面都略优于非季节模型。夏季模型的级别预报评分基本在90分以上,综合评分比非季节模型高10%。对夏季模型进行了参数敏感性分析,结果表明具有较好的稳定性。In this paper, a non-numeric air quality prediction model aiming at Huijingcheng, Foshan, Guangdong was proposed. The authors introduced BP neural network into the prediction model, and established the non-seasonal as well as summer model with the 7-7-1 structure. After the input factors analysis, which mainly discussing the impacts of different meteorological factors on the pollutants concentration, a few parameters were selected as the input neurons into BP neural network. The parameters included rainfall, wind speed, wind direction, temperature, humidity, cloud cover and the air quality of previous day. A series of pretreatment of these input data were carried out before entering into the training system. The result showed that the summer model was slightly better for both model test and actual prediction accuracy if compared to the non-season model. According to the grading criterion of air quality prediction issued by China meteorological administration, some evaluations were done to these models. It showed that the scores, level and API of summer model in the prediction of main pollutants were all higher than the non-season model. In addition, the summer model scored over 90 points in the level prediction which was 11.1 points higher than the non-season model, and its comprehensive score was 10% higher than the non-season model. For summer model, all three pollutants predictions have reached prediction result better than the non-season model. Among the results, NO2 made a score of 91.5 which was 16.0 higher than the latter model in the level prediction. Moreover, the summer model showed good stability in the sensitivity analysis. The output value presented certain regularity when a disturbance quantity was given to the input parameters and the variation was small. Therefore, the summer model could be used as reference in the non-numerical air quality prediction in Foshan.

关 键 词：环境工程学 空气质量预报 BP神经网络 非季节模型 夏季模型 

分 类 号：X51[环境科学与工程—环境工程]