一种前向小波神经网络实现PM_(2.5)浓度的预测

Forward wavelet neural network for prediction of PM_(2.5) concentration

PM_(2.5)作为主要的空气污染物,科学有效地预测其浓度能够让人们采取预防措施以减少对人体的伤害。使用传统的方法检测空气中PM_(2.5)的浓度收效甚微且成本高,这是因为PM_(2.5)组成成分复杂,并且其浓度的变化是一种非线性动态过程。因此,进行快速有效的预测PM_(2.5)浓度意义重大。文中采用了具有时域和频域二维信号处理能力且收敛速度较快的前向小波神经网络预测空气中的PM_(2.5)浓度,预处理后的数据输入到网络中进行训练和测试。结果表明,相比于BP和PSO⁃BP神经网络,小波神经网络对PM_(2.5)浓度的预测精度较高,降低了错误率,并且有效地减小了预测偏差,说明该方法用于空气的质量预测是可行的。

Scientifically and effectively predicting the concentration of PM_(2.5)(a major air pollutant)can enable people to take precautionary measures to reduce the impact on humans.Traditional methods have little effect and high cost in detecting PM_(2.5) concentration in air,because the composition of PM_(2.5) is complex and the change of its concentration is a nonlinear dynamic process.Therefore,it is of great significance to predict the concentration of PM_(2.5) quickly and effectively.In this paper,the forward wavelet neural network with two⁃dimensional signal processing ability in time domain and frequency domain and fast convergence speed is used to predict PM_(2.5) concentration in the air,and the preprocessed data is input into the network for training and testing.The results show that,in comparison with BP and PSO⁃BP neural networks,the wavelet neural networks have higher prediction accuracy for the concentration of PM_(2.5),can reduce the error rate,and effectively minish the prediction deviation.Therefore,the proposed method is feasible for air quality prediction.