| 摘 要: 目前对空气质量的预报常使用WRF-CMAQ模拟体系，但受限于模拟条件，预测结果并不理想，因此基于某监测点的污染物浓度实测数据，在预报过程中使用这些实测数据对一次预报数据进行修正以达到更好的预报效果。利用极限学习机模型训练对数据的预测，以AQI和首要污染物的误差这两个指标的加权组合作为适应度，通过遗传算法来优化模型，得到更准确地预测结果。并在对位置时间数据进行预测时采用滚动预测的方法以降低预测误差，相较于一次预测的预测误差降低了5%以上。结果表明：优化后的模型在空气质量预测的准确率方面有很大的提高。 |
| 关键词: 大气污染；插值；极限学习机；遗传算法 |
|
中图分类号: TP183
文献标识码: A
|
|
| Secondary Air Quality Forecast based on Extreme Learning Machine Model |
|
ZHU Shengkai, CHEN Jinjie
|
(University of Shanghai for Science and Technology, Shanghai 200093, China )
kaisss163@163.com; 2502526194@qq.com
|
| Abstract: At present, WRF-CMAQ simulation system is often used to forecast air quality, but the forecast results are not satisfactory due to its limited simulation condition. Therefore, based on the actual measured data of pollutant concentrations at a monitoring site, this paper proposes to use these actual data to correct the primary forecast data during the forecasting process to achieve better forecasting results. Using Extreme learning machine model to train the forecasting of data, taking the weighted combination of two indicators, AQI (Air Quality Index) and the error of primary pollutants, as the fitness, the model is optimized by genetic algorithm to obtain more accurate forecast results. The rolling forecast method is used to reduce the forecast error when forecasting the location and time data. The forecast error is reduced by more than 5% compared with that of primary data, and the results show that the optimized model has a great improvement in the accuracy of air quality forecast. |
| Keywords: air pollution; interpolation; extreme learning machine; genetic algorithm |
