基于CNN的空气中苯浓度预测模型

A Prediction Model of Benzene Concentration in Air Based on CNN

针对空气中苯浓度预测困难的问题.我们利用空气中多种成份之间关系为依据,以卷积神经网络(Convolutional Neural Network,CNN)为基础,设计了一种苯浓度预测模型.模型由多个卷积神经网络层、批归一化层和多个全连接神经网络层组成,模型输入为连续多个时刻空气中多种成份特征值,模型输出为预测的苯浓度值.通过对CNN模型的超参进行调整提高模型效果,实验结果显示在CNN核宽度为4,输入序列长度为8时模型取得最好效果.调整后的CNN模型的与基于支持向量机(Support Vector Machine,SVM)和基于长短期记忆神经网络(Long Short-Term Memory,LSTM)的模型进行比较,在测试集上的RMSE值与SVM模型和LSTM模型相比,分别降低1.51和1.54.在预测时模型计算需花费的时间上,CNN模型需要0.57ms,高于SVM模型需要的0.055ms,低于LSTM模型需要的1.15ms.

It is difficult to predict the benzene concentration in the air.We design a benzene concentration prediction model with the Convolutional Neural Network(CNN)based on the relationships between various components in the air.The model consists of multiple convolutional neural network layers,batch normalization layers,and multiple fully connected neural network layers.The model input is the feature values of multiple components in the air at multiple consecutive times,and the model output is the predicted benzene concentration value.The effect of the model is improved by adjusting the hyper-parameters of the CNN model.The experimental results show that the model achieves the best results when the CNN kernel width is 4 and the input sequence length is 8.The adjusted CNN model is then compared with the model based on Support Vector Machine(SVM)and the model based on Long Short-Term Memory(LSTM).Compared with values derived from the SVM model and the LSTM model,the RMSE value of the CNN model on the test set is reduced by 1.51 and 1.54,respectively.In terms of the time cost of prediction,the CNN model will take 0.57ms,which is longer than that with the SVM model(0.055ms)and shorter than that with the LSTM model(1.15ms).