基于Savitzky-Golay滤波的双向门控循环单元神经网络汽轮机热耗率预测  

Steam Turbine Heat Rate Prediction Based on Savitzky-Golay Filter and Bidirectional Gated Recurrent Unit Neural Network

作　　者：马良玉[1] 王永军 MA Liang-yu;WANG Yong-jun(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,保定071003

出　　处：《科学技术与工程》2020年第9期3623-3628,共6页Science Technology and Engineering

摘　　要：汽轮机热耗率是火电机组运行过程中的一项重要监测指标。为建立更加准确的汽轮机热耗率预测模型,借助某1 000 MW火电机组的真实历史数据,提出一种基于双向门控循环单元(gated recurrent unit,GRU)神经网络的汽轮机热耗率预测模型。针对火电机组现场运行数据噪声大的问题,采用SG(Savitzky-Golay)滤波器对所选变量数据进行降噪处理,将处理后的数据作为建模样本构建双向GRU神经网络汽轮机热耗率预测模型。并将其与BP(back propagation)神经网络、传统循环神经网络等2种算法的模型预测结果进行对比,结果表明:双向GRU神经网络热耗率预测模型的预测精度更高,泛化能力和鲁棒性更强,能够满足现场的实际需求。The steam turbine heat rate is an important monitoring index in the operation of thermal power units. In order to establish a more accurate prediction model of steam turbine heat rate, a prediction model based on bidirectional gated recurrent unit(GRU) neural network was proposed with the help of the real historical data of a 1 000 MW thermal power unit. In order to solve the problem of large noise in the field operation data of thermal power units, the Savitzky-Golay(SG) filter was used to reduce the noise of the selected variables. The heat rate prediction model based on bidirectional GRU neural network was constructed by using the processed data as modeling samples. Compared with the prediction results of back propagation(BP) network and traditional recurrent neural network, the results show that the BGRU neural network steam turbine heat rate prediction model has higher prediction accuracy, and its generalization ability and robustness are stronger, which can meet the actual needs of the field.

关 键 词：汽轮机热耗率 Savitzky-Golay 循环神经网络 门控循环单元 时间序列 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]