【作者】 许涛；

【导师】 刘长良；

【作者基本信息】 华北电力大学 ， 控制科学与工程， 2021， 硕士

【摘要】 机组运行优化是火电厂节能减排的重要途径之一。常规的关键参数目标值计算方法有着精度低、实时性差的缺点,针对这一问题,提出了一种基于数据驱动的火电机组运行优化方法。研究了机组参数特征选择、火电厂数据处理和工况划分等相关算法;采集了现场机组数据进行实验分析;设计开发了一套关键参数目标值寻优软件,并在现场投入使用。具体内容分为以下几个方面:首先,采用基于互信息的特征选择方法,从火电机组众多的数据参数中选出重要的关键参数,以达到数据降维的目的;同时,设计了一种包含经济性、安全性、环保性要求的综合优化目标函数。然后,构建了数据驱动所依赖的历史工况库,主要涉及数据处理和工况划分两个方面。在数据处理方面,对采集的机组运行数据采用基于孤立森林的异常点去除算法进行数据清洗;对清洗后的数据使用经验小波变换算法进行趋势提取;对提取出的趋势信号采用R统计检验法进行稳态检测。在工况划分方面,对稳态工况数据采用多步K均值聚类算法进行工况划分,生成多个历史工况库;在每个历史工况库中,利用优化目标函数标定若干个最优历史工况点。此外,还设计了最优历史工况点迭代更新的功能:每次读入新数据时,使用优化目标函数计算得分,并与最接近的最优工况点进行比较,当新数据更优时,使用新数据替代原来的最优工况点,实现迭代更新。最后,基于上述研究内容,设计并开发了一套关键参数目标值寻优软件,该软件通过挖掘历史工况库,找到与当前运行状态类似的最优历史工况点,生成关键参数目标值来指导机组当前的运行。目前该软件已在国内某信息中心投入使用,运行良好。更多还原

【Abstract】 Unit operation optimization is one of the important ways to save energy and reduce emissions in thermal power plants.Conventional methods for calculating target values of key parameters have the disadvantages of low accuracy and poor real-time performance.To solve this problem,a data-driven optimization method for the operation of thermal power units is proposed.Related algorithms such as unit parameter feature selection,thermal power plant data processing and working condition division are studied;unit data is collected for experimental analysis;a set of optimization software for key parameter target values was designed and developed and put into use on site.The specific content is divided into the following aspects:First of all,the feature selection method based on mutual information is adopted to select important key parameters from the numerous data parameters of thermal power units;at the same time,a comprehensive optimization objective function including the requirements of economy,safety and environmental protection is designed.Then,a historical working condition database was constructed,which mainly involves two aspects: data processing and working condition division.In terms of data processing,anomalous point removal algorithm based on isolated forests is used for data cleaning;empirical wavelet transform algorithm is used for trend extraction;the R statistical test method was used for steady-state detection.In terms of working condition division,multi-step K-means clustering algorithm is used to divide the steady-state working condition data to generate multiple historical working condition libraries;In addition,the iterative update function of the optimal historical operating point is also designed.Finally,based on the above research content,a set of key parameter target value optimization software is designed and developed.The software guides the current operation of the unit by mining the historical working condition database to generate key parameter target values.At present,the software has been put into use in an information center in China and is running well.更多还原