基于RF-LSTM混合神经网络的固废焚烧烟气排放浓度预测研究

Study on Prediction of Flue Gas Emission Concentration of Solid Waste Incineration Based on RF-LSTM Hybrid Neural Network

固废焚烧会产生有毒有害烟气,烟气排放浓度预测可以辅助固废焚烧工艺参数的高效调整。自编码器(Autoencoder,AE)、卷积神经网络(Convolutional Neural Networks,CNN)和长短期记忆(Long Short-Term Memory,LSTM)网络是3种常见的人工神经网络,而随机森林(Random Forest,RF)是一种高度灵活的机器学习算法。基于RF和LSTM网络,构建混合神经网络模型,结合成都市某固废焚烧发电厂运行工况数据,开展氮氧化物(NO_(x))浓度预测与分析。结果表明,RF-LSTM模型的均方根误差、平均绝对误差较AE-LSTM模型分别减少38.58%、46.56%,较CNN-LSTM模型分别减少23.77%、31.96%;RF-LSTM模型的决定系数较AE-LSTM模型增加22.54%,较CNN-LSTM模型增加16.00%。原始样本进行插值补缺时,步长为3 h的RF-LSTM模型预测精度最高,能够有效预测NO_(x)排放浓度。

The incineration of solid waste produces toxic and harmful flue gas,and predicting the concentration of flue gas emissions can assist in the efficient adjustment of process parameters for solid waste incineration.Autoencoder(AE),Convolutional Neural Networks(CNN),and Long Short Term Memory(LSTM)networks are three common artificial neural networks,while Random Forest(RF)is a highly flexible machine learning algorithm.Based on RF and LSTM networks,a hybrid neural network model is constructed,combined with operational data from a solid waste incineration power plant in Chengdu city,to predict and analyze nitrogen oxide(NO_(x))concentration.The results show that the root mean squared error and mean absolute error of the RF-LSTM model are reduced by 38.58%and 46.56%respectively compared to the AELSTM model,and by 23.77%and 31.96%respectively compared to the CNN-LSTM model;the coefficient of determination of the RF-LSTM model increases by 22.54%compared to the AE-LSTM model and 16.00%compared to the CNN-LSTM model.When interpolating and flling in gaps in the original samples,the RF-LSTM model with a step size of 3 h has the highest prediction accuracy and can effectively predict NO_(x) emission concentrations.