关于空气中PM2.5质量浓度预测研究

Research on Prediction Mass Concentration of PM2.5

针对PM2.5受到天气因素、大气污染物的的影响巨大,具有明显的非线性、不确定特征,传统预测方式很难得到有效的预测结果,对PM2.5质量浓度预测问题,提出了采用遗传算法优化的神经网络的预测方法。采用遗传算法对BP神经网络的权值及阈值进行优化,在处理该类问题上具有很好的学习、泛化、映射能力。以宝鸡市监测站每小时监测数据为研究对象,进行PM2.5小时浓度预测建模。仿真结果表明,采用遗传算法优化的BP神经网络PM2.5预测的拟合和预测平均绝对误差分别为8.6%和12.3%,空气质量等级预测正确率分别为86%和85%。将模型与BP神经网络预测结果进行对比分析,结果表明,遗传算法优化的BP神经网络的预测结果具有更高的准确度和精确度。

The mass concentration of PM2.5 is influenced greatly by the weather, air pollutions and other factors. A prediction method based on GA-BP was proposed in this paper. Genetic algorithm was used to optimize the thresh- olds and weights of BP neural network, which is of great capabilities of learning, generalization and mapping in deal- ing with such nonlinear issue. Using the monitoring data of Baoji monitoring station as study object, a PM2.5/hour concentration prediction model was established. The simulation results show that the fitting error and prediction mean absolute error of PM2.5 prediction based on BP-GA are 8.6% and 12.3% respectively, and the accuracy rates of Air quality levels prediction are 86% and 85% respectively. The Comparison and analysis of the predicted results of GA-BP and BP neural network show that the prediction result of GA-BP has higher accuracy and precision.