摘要
针对燃煤电厂锅炉燃烧工况复杂多变和脱硫系统惯性大,影响因素多,导致的出口SO_(2)浓度频繁大范围波动且难以预测的问题,提出一种基于浣熊优化算法(coati optimization algorithm,COA)优化变分模态分解(variational mode decomposition,VMD)算法与融合卷积神经网络(convolutional neural network,CNN),双向长短期记忆网络(bidirectional long short-term memory networks,BiLSTM)和注意力机制的出口SO_(2)浓度预测模型。首先使用k-近邻互信息法筛选出与出口SO_(2)浓度相关性高的辅助变量,求取出各个辅助变量对应的时延补偿,然后对补偿后的变量用COA-VMD算法进行分解,保留分解结果中与输出变量相关性最大的变量子集进行重构,并将其作为模型的输入,最后使用CNN-BiLSTM-Attention建立出口SO_(2)浓度预测模型。仿真结果表明,相比其他模型该模型的均方根误差、平均绝对百分比误差最小,预测精度最高,分别为0.5777 mg/m^(3),0.2705%,0.9732。
In response to the problem of frequent and wide range fluctuations in outlet SO_(2)concentration caused by the complex and variable combustion conditions of coal-fired power plant boilers,as well as the large inertia and influencing factors of the desulfurization system,an outlet SO_(2)concentration prediction model based on the coati optimization algorithm(COA)optimized variational mode decomposition(VMD)algorithm,convolutional neural network(CNN),bidirectional long short term memory networks(BiLSTM),and attention mechanism was proposed.Firstly,the k-nearest neighbor mutual information method was used to screen out auxiliary variables with high correlation with the outlet SO_(2)concentration,and the corresponding time delay compensation for each auxiliary variable was obtained.Then,the compensated variables were decomposed using the COA-VMD algorithm,retaining the subset of variables with the highest correlation with the output variables in the decomposition results for reconstruction and using them as inputs to the model.Finally,a prediction model for outlet SO_(2)concentration was established using CNN-BiLSTM-Attention.The simulation results show that compared with other models,this model has the smallest root mean square error and average absolute percentage error,and the highest prediction accuracy,which are 0.5777mg/m^(3),0.2705%,0.9732,respectively.
作者
畅晗
金秀章
赵术善
赵大勇
CHANG Han;JIN Xiuzhang;ZHAO Shushan;ZHAO Dayong(School of Control and Computer Engineering,North China Electric Power University,Baoding,Hebei 071003,China)
出处
《计量学报》
北大核心
2025年第7期1041-1050,共10页
Acta Metrologica Sinica
基金
国家重点研发计划项目(2018YFB0604300)。
关键词
SO_(2)浓度预测
浣熊优化算法
VMD分解
卷积神经网络
双向长短期记忆网络
注意力机制
SO_(2)concentration prediction
coati optimization algorithm
VMD decomposition
convolutional neural networks
bidirectional long short term memory network
attention mechanism
作者简介
第一作者:畅晗(2000-),男,山西永济人,华北电力大学硕士研究生,研究方向燃煤电厂脱硫系统优化控制。Email:2962218791@qq.com。
