摘要
提出一种基于最大相关最小冗余(mRMR)算法和蜉蝣算法优化正则化极限学习机(MA-RELM)的出口SO_(2)质量浓度预测模型。通过机理分析确定初始输入变量,利用改进的时延分析方法对初始输入变量进行时延补偿,采用mRMR算法对各个初始输入变量进行重要性排序,搭建正则化极限学习机(RELM)预测模型,并利用蜉蝣算法确定模型参数。结果表明:与最小二乘支持向量机(LSSVM)、长短期记忆网络(LSTM)和极限学习机(ELM)相比,RELM预测模型的均方根误差分别降低了36%、38%和26%;与粒子群算法(PSO)和灰狼算法(GWO)寻优后的模型相比,MA-RELM预测模型误差最低,该模型能够对出口SO_(2)质量浓度进行准确预测。
A prediction model of outlet SO_(2) mass concentration based on max-relevance and min-redundancy(mRMR) algorithm and regularized extreme learning machine optimized by mayfly algorithm(MA-RELM) was proposed. The initial input variables were determined through mechanism analysis method, and the time delay of the initial input variables was compensated by improved time delay analysis method, the importance of each initial input variable was sorted by mRMR algorithm. Finally, the RELM prediction model was established, and the model parameters were determined by mayfly algorithm. Results show that compared with least square support vector machine(LSSVM), long short-term memory(LSTM) and extreme learning machine(ELM), root mean square error of the RELM prediction model is reduced by 36%, 38% and 26%, respectively. Compared with the model optimized by particle swarm optimization(PSO) and gray wolf algorithm(GWO), the error of the MA-RELM prediction model is the lowest, and the model can accurately predict outlet SO_(2) mass concentration.
作者
金秀章
刘岳
赵文杰
于静
JIN Xiuzhang;LIU Yue;ZHAO Wenjie;YU Jing(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,Hebei Province,China)
出处
《动力工程学报》
CAS
CSCD
北大核心
2022年第7期664-670,676,共8页
Journal of Chinese Society of Power Engineering