发电技术 ›› 2024, Vol. 45 ›› Issue (6): 1105-1113.DOI: 10.12096/j.2096-4528.pgt.22151

• 发电及环境保护 • 上一篇    

考虑天气影响的火电机组空冷系统性能预测方法

董建宁, 安吉振, 陈衡, 潘佩媛, 徐钢, 王修彦   

  1. 华北电力大学能源动力与机械工程学院,北京市 昌平区 102206
  • 收稿日期:2023-09-06 修回日期:2023-10-26 出版日期:2024-12-31 发布日期:2024-12-30
  • 通讯作者: 陈衡
  • 作者简介:董建宁(1999),男,硕士,主要研究方向为电站大数据分析和智能预警,realdjn@126.com
    安吉振(1997),男,硕士,主要研究方向为电站空冷系统优化与大数据分析,120202202033@ncepu.edu.cn
    陈衡(1989),男,博士,副教授,主要研究方向为电站大数据分析和智能预警,本文通信作者,heng@ncepu.edu.cn
    潘佩媛(1992),女,博士,副教授,主要研究方向为烟气低温受热面积灰腐蚀耦合机理、高温受热面热腐蚀机理、多相烟气组分反应动力学机制,peiyuanpan@necpu.edu.cn
    徐钢(1978),男,博士,教授,主要研究方向为能源动力系统优化与节能、污染物控制及温室气体减排,xgncepu@163.com
  • 基金资助:
    国家自然科学基金项目(52106008)

Performance Prediction Method for Air Cooling System of Thermal Power Unit Considering Weather Effect

Jianning DONG, Jizhen AN, Heng CHEN, Peiyuan PAN, Gang XU, Xiuyan WANG   

  1. School of Energy Power and Mechanical Engineering, North China Electric Power University, Changping District, Beijing 102206, China
  • Received:2023-09-06 Revised:2023-10-26 Published:2024-12-31 Online:2024-12-30
  • Contact: Heng CHEN
  • Supported by:
    National Natural Science Foundation of China(52106008)

摘要:

目的 直接空冷机组是一部分缺水地区常用的火力发电形式,由于其采用空气作为冷却介质,因此运行受到诸多限制。为解决直接空冷机组受环境影响大和耗煤量高的问题,对空冷岛换热性能进行预测研究。 方法 基于河北省某超临界2×600 MW机组的历史运行数据,利用MATLAB软件计算历史空冷岛性能,将历史数据作为训练集和测试集,通过长短期记忆(long short-term memory,LSTM)神经网络机器学习算法,实现对未来一段时间内的空冷岛性能预测。在不改变模型参数的条件下,通过去除各项特征的方式确定特征重要性排名,基于此确定最佳特征选择策略,进一步优化模型。考虑到空冷岛性能受天气影响大,为提升模型在特殊天气下的预测能力,将原数据集结合历史天气预报数据,编写考虑天气因素的预测程序,以预测空冷岛未来一段时间性能,并对预测结果进行可视化分析。 结果 所采用的预测模型预测准确度明显高于传统自回归移动平均模型(autoregressive integrated moving average model,ARIMA),对未来1 h以内的直接空冷机组换热性能预测拟合优度均在0.90以上。 结论 模型所采用的数据特征及算法可以为直接空冷机组的稳定运行提供数据支撑,为智慧电厂的建设提供技术基础。

关键词: 火力发电, 火电机组, 空冷系统, 直接空冷机组, 长短期记忆(LSTM)神经网络, 性能预测, 特征重要性, 天气因素

Abstract:

Objectives Direct air-cooled unit is a common equipment of thermal power generation in some water-deficient areas. The operation is subject to many restrictions because it uses air as its cooling medium. Heat transfer performance of air-cooled island was studied to solve these problems that direct air-cooled units are greatly affected by the environment and have high coal consumption. Methods Based on history-data of a supercritical 2×600 MW unit in Hebei Province, the performance of its air-cooled island was calculated with MATLAB software, this study considered the acquired data as the training set and the test set,which were used to predict future performance in virtue of long short-term memory (LSTM) neural network machine learning algorithm. Under the condition that the model parameters were not changed, the feature importance ranking was determined by removing all features, based on which the best feature selection strategy was determined to further optimize the model. Considering the great impact from the weather, a prediction procedure, taking into account weather factors, was written to improve the accuracy of predicting air-cooled island performance, by combining the original data set with historical weather data. Accordingly prediction results were subjected to visualization and analyzation. Results The prediction accuracy of the adopted prediction model is significantly higher than that of the traditional autoregressive integrated moving average model (ARIMA), and the goodness of fit of the direct air-cooled unit heat transfer performance prediction within the next hour is above 0.90. Conclusions The data characteristics and algorithms used in the model can provide data support for the stable operation of the direct air-cooled unit and provide a technical basis for the construction of intelligent power plants.

Key words: thermal power generation, thermal power units, air cooling system, direct air cooling unit, long and short-term memory (LSTM) neural network, performance prediction, feature importance, weather factor

中图分类号: