基于静电法联合长短时记忆神经网络的入炉煤质辨识方法

Coal quality identification method based on electrostatic method combined with long short-term memory neural network

燃煤机组每台磨煤机的实时入炉煤质对锅炉的燃烧优化调整具有重要意义,但传统煤质检测程序繁琐,且存在检测数据滞后的情况。因此,提出一种基于静电法联合长短时记忆(LSTM)神经网络的入炉煤质辨识方法。根据不同煤粉传输过程中存在静电变化的特性,安装静电传感器检测不同煤粉通过管道时的静电信号强度,结合风煤比、煤粉流速、磨煤机出口温度等影响参数,并结合现场数据分析参数相关性与迟延性,构建煤质辨识LSTM神经网络模型。以某600 MW机组锅炉实测数据为例,经参数寻优后采用LSTM神经网络模型辨识煤质准确率达到86.84%,对4种煤质分类结果的评估指标AUC值均在0.9以上,并与其他机器学习模型进行对比实验,结果表明LSTM神经网络模型具有更高的辨识精度,验证了该方法的可行性和准确性。

The ability to obtain the coal quality of each coal mill in real time is of great significance to optimization and adjustment of boiler combustion.The conventional coal quality detection procedure is cumbersome and the detection data lag behind.Therefore,a coal quality identification method based on electrostatic method combined with long short term memory(LSTM)neural network is proposed.In this method,according to the characteristics of electrostatic changes in the transmission process of different pulverized coal,electro-static sensors are installed to detect the electrostatic signal strength of different pulverized coal passing through the pipeline,the correlation and delay of several parameters such as air-coal ratio,pulverized coal flow rate and coall mill outlet temperature is analyzed by combining with the site data,and an LSTM neural network model for coal quality identification is established.Taking the measured data of a 600 MW unit boiler as an example,after parameter optimization,the coal quality identification accuracy of the LSTM neural network model reaches 86.84%,and all the AUC values of the four coal quality classification results are above 0.9.Compared with other machine learning models,the LSTM neural network model has higher identification accuracy,which is feasible and accurate.