基于BP神经网络模型的贵阳市臭氧浓度预报研究
Research on Forecast of Ozone Concentration in Guiyang City Based on BP Neural Network Model

本文基于BP神经网络模型，利用贵阳市环监站2016年6月~2017年9月O₃浓度逐时数据以及对应时刻气象观测数据，在系统分析贵阳市臭氧浓度变化特征的基础上，利用BP神经网络模型，对贵阳市大气O₃浓度进行预测，结果表明：1) 贵阳市O₃浓度总体呈现出春夏高、秋冬低的波动变化特征，年均值为58 μg/m³；5月(86 μg/m³)、11月(40 μg/m³)分别为浓度最高、最低月。在季节时间尺度上表现为春季 > 夏季 > 冬季 > 秋季的特点，夏季均值明显高于其他三季；日变化特征为单峰型，18:00出现峰值，峰值为82 μg/m³；2) O₃浓度与气压、相对湿度呈负相关，与温度、风速呈现正相关，相关系数分别为？0.61、？0.75、0.81和0.76；3) 基于只考虑气象因素和考虑气象因素加污染因素两种输入方案的BP神经网络均能较好地预测O₃质量浓度，前者实测O₃质量浓度与预测O₃质量浓度相关系数为0.69，后者为0.81，且两者预测结果与实测数据的平均绝对误差分别为40.2%和29.6%。说明基于考虑了气象和污染双重因子的输入方案BP神经网络预测结果相较于基于只考虑气象因子的输入方案的BP神经网络预测结果更优。
By using the hourly data of O₃ concentration from June 2016 to September 2017 at Guiyang envi-ronmental monitoring station and the meteorological observation data at the corresponding time, based on the systematic analysis of the characteristics of the ozone concentration change in Guiyang, the BP neural network model was used to predict O₃ concentration. The results showed that: 1) The O₃ concentration in Guiyang showed a fluctuating characteristic, the value was high in spring and summer and low in autumn and winter, with an annual average value of 58 μg/m3; May (86 μg/m³) and November (40 μg/m³) were the highest and the lowest months respectively. On the seasonal time scale, the characteristics are spring > summer > winter > autumn. The average value of summer is significantly higher than which of the other three seasons. The daily variation was characterized by a single peak, with a peak at 18:00 (82 μg/m³); 2) O₃ concentration was negatively correlated with air pressure and relative humidity, and positively correlated with temperature and wind speed, the correlation coefficients were ？0.61, ？0.75, 0.81 and 0.76 respectively; The BP neural network of the two input schemes can predict the O₃ mass concentration well, the correlation coefficient of the measured O₃ mass concentration and predicted O₃ mass concentration was 0.69 of the former, and was 0.81 of the latter. The average absolute error between the predicted results of both was 40.2% and 29.6% respectively. The result of the input scheme that takes into account the dual factors of weather and pollution was better than the result of the input scheme which only considers the meteorological factors.