循环神经网络模型下道路碳排放浓度预测

Prediction of road carbon emission concentration based on recurrent neural network model

以湖南省永州市永州大道基本路段CO_(2)浓度时序数据为研究对象,旨在实现道路CO_(2)浓度的实时预测。测得用于模型训练和预测精度计算的路段CO_(2)浓度数据,利用Savitzky-Golay滤波器对数据进行平滑去噪,在调试并建立循环神经网络最优模型结构的基础上,引入多元预测模型(MLR、SVR、BP)和时序预测模型(BP、RF、RNN、LSTM、GRU)进行预测性能对比,为路段CO_(2)浓度的实时预测提供参照。结果表明:时序预测模型相比于多元预测模型具有更好的预测效果,特别是循环神经网络模型中的GRU表现出较高的预测精度,其次是LSTM,最后是RNN;循环神经网络模型在处理路段CO_(2)浓度时序数据的训练和预测任务中具备突出性能,能够实时且精准预测道路路段CO_(2)浓度。

The article took the chronological data of CO_(2)concentration in the basic road section of Yongzhou Avenue in Yongzhou City,Hunan Province as the research object,aiming to realize the real-time prediction of CO_(2)concentration in road sections.The CO_(2)concentration data of road sections for model training and prediction accuracy calculation were measured,and the data were smoothed and denoised by Savitzky-Golay filter.On the basis of debugging and establishing the optimal model structure of recurrent neural network,multivariate prediction models(MLR,SVR,BP)and time-sequence prediction models(BP,RF,RNN,LSTM,GRU)were introduced for comparison of the prediction performance in order to provide references to real-time prediction of road section CO_(2)concentration,the results showed that time-sequence prediction model has better prediction effect compared with multivariate prediction model.The results show that the time-series prediction model has better prediction results than the multivariate prediction model,especially,the GRU in the recurrent neural network model shows higher prediction accuracy,followed by LSTM,and finally RNN.The article concludes that the recurrent neural network model has an outstanding performance in the task of training and prediction of chronological data of roadway section CO_(2)concentration,and it is able to predict the CO_(2)concentration of roadway sections in real time and with high accuracy.