基于神经网络预测的主蒸汽温度控制方法研究  

Research on Main Steam Temperature Control Method based on Neural Network Prediction

作　　者：周飞燃 曾德良[1] 胡勇 ZHOU Feiran;ZENG Deliang;HU Yong(School of Control and Computer Engineering,North China Electric Power University,Beijing,China,102206)

机构地区：[1]华北电力大学控制科学与工程学院,北京102206

出　　处：《热能动力工程》2024年第10期148-158,共11页Journal of Engineering for Thermal Energy and Power

基　　金：国家自然科学基金重点项目(61833011)。

摘　　要：为了解决火电机组参与灵活调峰运行时,负荷的大范围变动引起煤量、风量等参数频繁变化,进而导致主蒸汽温度剧烈波动的问题,以某600MW汽包锅炉为研究对象,提出一种基于神经网络预测的主蒸汽温度控制优化方法,采用随机森林算法挖掘对主蒸汽温度起主要作用的影响因素;根据各影响因素存在的迟延问题,采用互信息法对数据进行迟延对齐处理,以减小数据迟延时间对预测精度的影响;采用CNN-LSTM-Attention神经网络构建主蒸汽温度预测模型,对未来时刻主蒸汽温度的变化进行预测,并构建主蒸汽温度的预估补偿系统。仿真结果表明:经过时序对齐处理后的CNN-LSTM-Attention神经网络预测模型对主蒸汽温度的预测决定系数R^(2)值增长了8.45%,平均绝对误差S_(MAE)值降低了35.56%,均方根误差S_(RMSE)值降低了33.10%,有效提高了对主蒸汽温度的预测精度,通过设定值补偿方案,相较于传统PID控制,在不同负荷条件下,主蒸汽温度的波动幅度分别降低了57.7%,62.0%和56.8%,能提前抑制各扰动量对主蒸汽温度的影响。In order to solve the problem of frequent changes in parameters such as coal quantity and air volume caused by large-scale load fluctuations during flexible peak shaving operation of thermal power u-nits,which led to severe fluctuations in main steam temperature,a neural network-based prediction opti-mization method for main steam temperature control was proposed for a 600 MW steam drum boiler.The random forest algorithm was used to mine the influencing factors that play a major role in main steam tem-perature;based on the delay issues caused by various influencing factors,the mutual information method was used to perform delay alignment on the data,in order to reduce the impact of data delay time on pre-diction accuracy.A CNN-LSTM-Attention neural network was used to construct a main steam temperature prediction model,which predicted the changes in main steam temperature in the future and constructed an estimated compensation system for main steam temperature.The simulation results show that the CNN-LSTM-Attention neural network prediction model after time alignment processing has increased the R^(2) value of the main steam temperature prediction by 8.45%,reduced the S_(MAE) value by 35.56%,and reduced the S_(RMSE) value by 33.10%,effectively improving the prediction accuracy of the main steam temperature.Through the set value compensation scheme,compared with traditional PID control,the fluctuation amplitudes of the main steam temperature under different load conditions have been reduced by 57.7%,62.0%,and 56.8%,respectively,which can suppress the influence of various disturbances on the main steam temperature in advance.

关 键 词：主蒸汽温度 神经网络 预测控制 

分 类 号：TM621.4[电气工程—电力系统及自动化]