基于现场数据的过热汽温动态建模

论文总字数：23705字

摘 要

电厂过热汽温控制对电厂的经济性有重要影响，因此建立精确的过热汽温系统动态模型就很重要。本文采用主元分析从影响过热汽温的主汽压力，有功功率，给水温度，减温水量等10个因素中确定减温水量为主要影响因素。采用平均值法对现场采集的数据进行去噪处理，减少了数据的波动，便于后续的最小二乘法和神经网络建模。在ELM神经网络建模前，由于采集的数据的单位可能不一致，因此需要对数据进行归一化处理，使得数据在[-1,1]之间。然后分别采用最小二乘法和ELM神经网络对过热汽温系统建模，并比较其精度。采用最小二乘法对过热汽温系统建模，需要测试不同的阶次和延迟最后建立最佳过热汽温模型，仿真结果表明模型输出与实际值相似度很高。同时每隔5s采集数据可以有效地减少数据波动对建模的影响，而且使用该数据所建立模型的精度更高。ELM神经网络带有反馈环节，其自身结构决定其可以捕捉到动态的输入输出关系，适合用于过热汽温动态建模，但是其需要选取合适的训练函数，而且ELM神经网络建模速度虽然比BP网络建模速度快，但是比最小二乘法建模速度慢。从建模的结果看，ELM神经网络所建立模型的精度比最小二乘法要高。

关键字：汽温系统；动态建模；现场数据；最小二乘法；ELM神经网络

The dynamic modeling of superheating steam temperature system based on field data

ABSTRACT

Since the controlling system in power plants has an important impact on the economical efficiency of power plants , the establishment of accurate modeling of superheating steam temperature controlling system is very important. In this paper, using PCA （principal component analysis）analyses the main factors that influence the superheating stream temperature. The average method is adopted to deal with the noise in the data from the field, in order to reduce the volatility of the data, and facilitate the superheating stream system modeling established by least-square method and neural network.  Before ELM neural network modeling, the data need to be normalized since the unit of data may be inconsistent. Then least-square method and ELM neural network are used to model the superheating steam temperature system and we will compare their precisions. The least-square method is used to model superheating stream system. By testing different orders and delays, we find the best model. The simulation results show that the similarity of the model output and the actual value is very high. Meanwhile using the data that are collected every 5 seconds models superheating stream system, which can effectively reduce the impact of data fluctuations in the modeling, and the new model has a higher accuracy . Then ELM neural network is used to model overheating stream system. ELM neural network has feedback link, which helps it capture the dynamics of the input-output data, so ELM neural network is suitable for modeling superheating steam system. But it is necessary to select the appropriate training function, and the modeling of ELM neural network is faster than BP Network, but it is still slower than the least-square method. The modeling results show that the accuracy of the model established by ELM neural network is higher than it established by least-square method.

KEYWORDS：Steam temperature system；Dynamic modeling；Field data； least-square method；Extreme Learning Machine

目录

第一章、绪论 1

1.1、背景及意义 1

1.2、课题研究现状 3

1.3、本文研究内容 4

第二章、过热汽温系统的主元分析6

剩余内容已隐藏，请支付后下载全文，论文总字数：23705字

相关图片展示：