基于注意力机制和CNN-GRU模型的脱硫系统pH值预测

Prediction of pH Value in Desulfurization System Based on CNN-GRU and Attention Mechanism Model

针对火电厂石灰石-石膏烟气湿法脱硫系统中所面临的浆液pH值测量不准确、时间长的问题,提出了一种基于注意力机制和CNN-GRU模型来预测吸收塔内浆液pH值。首先,对火电厂监测系统(SIS)数据库中的数据进行预处理,然后使用相关性分析来确定它们之间的关联性。接下来可使用注意力机制(ATT)来自适应分配与pH值相关联的输入数据的权重,并根据权重大小来区分强弱特征变量,以此来解决预测精度低和不准确的问题。其后利用卷积神经网络(CNN)来二次提取和降维这些特征数据,并对送入门控循环神经单元网络(GRU)中的数据进行优化,可大大加快神经网络训练进程,并且能够更准确地处理复杂的动态脱硫变化。对某电厂2×350 MW机组运行数据进行测试,并通过与其他主流算法对比得出所建pH值预测模型具备更高的精确度和稳定性。最后结合模型预测控制(MPC),验证了该模型的实用性。

CNN-GRU model and attention mechanism-based method is proposed to forecast the pH of the slurry in thermal power plant's limestone gypsum flue gas wet desulfurization system,in order to tackle the issue of inaccurate and protracted pH measurements within the absorption tower.Firstly,correlation analysis is conducted on the data stored in the SIS data base.Subsequently,Attention(Attention,ATT)is utilized to adaptively assign weight to the input data related to pH value,distinguishing between strong and weak feature variables based on the weight.Finally,Convolutional Neural Network(CNN)is employed to extract and reduce the dimension of the chosen feature data.The data sent to the Gate Recurrent Unit(Gate Recurrent Unit,GRU)network are optimized so as to improve the training speed and prediction accuracy of the neural network.Test of the operation data of 2×350 MW units in a domestic power plant verifies that the pH value prediction model has higher accuracy and stability by comparing it with other mainstream algorithms.Finally,the practicability of the model is verified by combined with the MPC.