基于递归神经网络的火电机组污染物排放研究

Research on Quantity of Power Unit Pollutant Emission Based on Recurrent Neural Networks

该文通过研究海量的发电机组历史污染物排放数据,提出一种基于LSTM-RNN深度学习的改进型发电机组排放预测算法ALSTM-RNN(A-R)。A-R算法可以有效地提取出模型特征量,结合数据的归一化对模型的结果进行优化调整,以降低模型训练时间,提高预测精度。通过在不同的发电机组测试试验,A-R算法较最小二乘法(LSM),支持向量机回归(SVR)具有较小的均方误差值,较LSTM-RNN模型预测方差更小,更加稳定,具有较好的鲁棒性。

Through studying massive historical data of pollutant emissions in generation sets,this paper puts forward an improved algorithm ALSTM-RNN（A-R）for predicting emissions of generation sets based on deep learning algorithm LSTM-RNN. A-R algorithm can extract features of the model effectively. Through the normalization of data to optimize and adjust results of the model,which is able to reduce the training time of our model and improve the predictive accuracy. By testing on the different generator sets,A-R algorithm has a relatively small mean square error compared with the least squares method（LSM）and support vector machine regression（SVR）. Meanwhile,compared with LSTM-RNN,A-R method owns small predictive variance,a stable model and good robustness.