摘要
针对汽轮机阀门流量特性的辨识,提出一种基于贝叶斯优化(Bayesian Optimization,BO)的线性自分段神经网络模型,确保辨识的结果存在反函数,从而可以更加精准地实现阀门调节参数优化。首先,在前馈神经网络(Forward Neural Network,FNN)中引入ReLU(Rectified Linear Unit,ReLU)激活函数,从而完成对阀门流量数据的分段线性辨识;其次,在FNN训练的过程中使用BO对FNN的超参数进行寻优,给出最优的线性分段数;最后,对山东省某火电机组阀门流量数据进行分段函数线性辨识,对文中提出的模型和优化方法进行了验证。
A linear self⁃segmented neural network model based on Bayesian optimization(BO)was proposed to identify the steam turbine valve flow.The identification result had inverse function,which could realize the valve regulation parameter optimization more accurately.Firstly,the rectified linear unit(ReLU)activation function was introduced into forward neural network(FNN)to complete the piecewise linear identification of valve flow.Secondly,BO was used to optimize the super parameters of FNN in the process of FNN training,and the optimal number of linear segments was given.Finally,the valve flow of a thermal power plant in Shandong province were identified by piecewise function linear identification,which verifies the correctness of the proposed model and optimization method.
作者
路宽
王文宽
孟祥荣
李军
杨子江
张森烨
LU Kuan;WANG Wenkuan;MENG Xiangrong;LI Jun;YANG Zijiang;ZHANG Senye(State Grid Shandong Electric Power Research Institute,Jinan 250003,China;College of Electrical Engineering and Automation,Shandong University of Science and Technology,Qingdao 266590,China;State Grid Yantai Power Supply Company,Yantai 264000,China)
出处
《山东电力技术》
2021年第9期1-5,16,共6页
Shandong Electric Power
基金
国家自然基金青年科学基金项目(61803233)。
关键词
FNN
高斯回归
贝叶斯优化
线性分段
汽轮机阀门流量
forward neural network
Guassian regression
Bayesian optimization
piecewise linear
steam turbine valve flow
作者简介
路宽(1983),男,高级工程师,主要研究方向为网源协调技术研究与人工智能大数据分析。
