Clean Coal Technology

2022, v.28;No.142(06) 199-205

[Print This Page] [Close]
Current Issue | Past Issue | Advanced Search

Performance prediction of circulating fluidized bed unit based on machine learning

HAN Yi;ZHANG Qiyue;DUAN Lunbo;WANG Yankai;YU Yingli;FU Xuchen;RONG Jun;SUN Shichao;Inner Mongolia Electric Power Research Institute Branch,Inner Monglia Electric Power (Group)Co.,Ltd.;Key Laboratory of Energy Thermal Conversion and Control,Ministry of Education,Southeast University;

Abstract:

Coal power is an important supporting and regulating power supply in the clean and low-carbon transformation of power system. However, the technical output of thermal power units is hindered due to factors such as low-quality coal combustion, which seriously affects the safe operation of power grid and new energy power consumption. In view of this, a projection model building method based on the integration of mechanism simulation and data drive was presented in this paper. The sample space of boiler thermal system was constructed by mechanism simulation, and the unit output prediction was carried out based on mathematical projection. Considering the theoretical accuracy of mechanism simulation and the strong generalization of mathematical projection, the dynamic boundary output prediction of circulating fluidized bed units and the analysis of output blocking factors were realized under the condition of multi-factor coupling. The test results of 300 MWe demonstration unit shows that: considering the three influencing factors of auxiliary machine limitation, heating surface parameter overrun and key parameter overrun, the alarm values for exceeding the limit of operating parameters such as coal feeder, induced draft fan, slag cooler, bed temperature, screen wall temperature and fluidization wind speed are set respectively. The maximum deviation of mechanism simulation is 3 ℃,and the error rate is less than 2%. The BP neural network model with 7 inputs and 1 outputs is screened and designed based on the principal component analysis method. After network optimization by genetic algorithm, the network training and prediction are carried out by using 32 training samples and 5 test samples. The relative error of model training is within ±1.2%,the relative error of model prediction is within ±1.5%,ind icating that there is high accuracy, generalization ability, and worth reference.

Key Words: data drive;output prediction;circulating fluidized bed boiler;mathematical model;principal component analysis method;genetic algorithm

Abstract:

Keywords:

Foundation: 国家重点研发计划资助项目(2019YFE0100100-08)

Authors: HAN Yi;ZHANG Qiyue;DUAN Lunbo;WANG Yankai;YU Yingli;FU Xuchen;RONG Jun;SUN Shichao;Inner Mongolia Electric Power Research Institute Branch,Inner Monglia Electric Power (Group)Co.,Ltd.;Key Laboratory of Energy Thermal Conversion and Control,Ministry of Education,Southeast University;

DOI: 10.13226/j.issn.1006-6772.21061701

References:

Accessibility
Information
Service
Key Words
The author of this article
Cnki
Share