基于NARX神经网络的湖南省空气质量预报效果评估研究

Evaluation of Air Quality Forecast Performance Based on NARX Neural Network Model in Hunan Province

针对湖南省的带外源输入的非线性自回归(Nonlinear autoregressive with external input,NARX)神经网络模型,评估了2019年10月1日-2020年9月30日的污染物浓度预报准确率,并进一步评估模式对O_(3)和PM_(2.5)等两个重点污染物的时空分布预报效果。结果表明:模式预报的未来第1天级别准确率为91%,首要污染物准确率为70%;在空气质量等级为优良时,其预报能力较好。从时间序列上看,模式预报的PM和臭氧日最大8小时滑动平均浓度(O_(3)_8hr)与实测浓度的变化趋势较为一致,但模式对PM浓度有所高估;从空间分布上看,模式对夏季和冬季O_(3)_8hr整体的空间分布预报能力较优,且能够捕捉冬季PM污染过程的发生。

The forecast performance of the NARX neural network model was evaluated,with results from October 1,2019 to September 30,2020 in Hunan Province.The evaluation included two parts,one was the forecast accuracy of air quality level and primary pollutants,the other was the temporal and spatial distribution of O_(3)and PM_(2.5).The future 1-day forecast accuracy of air quality level was 91%,while that of primary pollutants was 70%.When the air quality index was less than 100,the forecast accuracy of air quality level achieved above 90%.The simulated temporal variation of PMand the maximum daily 8-h average(O_(3)_8hr)corresponded to the observation,but the model slightly overestimated PM.The model was capable of predicting the spatial distribution of O_(3)_8hr in both summer and winter,and capturing the PMpollution events in winter.