大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【计算机科学】【2006.01】【部分源码】火电厂监控与性能分析的人工神经网络模型

已有 174 次阅读 2020-4-4 19:20 |系统分类:科研笔记|关键词:学者|文章来源:转载

本文为瑞典隆德大学(作者:Mehrzad Kaiadi)的硕士论文,共103页。

 

厄勒森德斯卡夫特公司拥有许多工厂,主要向赫尔辛基市及其周边地区提供电力和地区供热。Västhamnsverket是该公司的主要工厂,装机容量为126兆瓦电和186兆瓦热Västhamnsverket是一个混合热电厂。为了证明一种有效的控制监测模型,针对Västhamnsverket的不同部分建立了若干神经网络模型。

 

本硕士论文是在Västhamnsverket应用的连续人工神经网络研究的一部分。这项研究的总体目标是为Västhamnsverket的整个蒸汽过程开发一个神经网络模拟器。为了对整个蒸汽过程建立一个更好的神经网络模型,该过程分为两个子模块,即案例1和案例2。因此,本文分别研究了两种情况,建立了两种不同的神经网络模型,然后将它们相互连接。每种情况的输入和输出参数都是根据后面讨论的适当标准来确定的。一旦输入和输出参数决定,现有数据集是从工厂获得的。为了使数据集尽可能的可靠,对神经网络训练前的数据进行了预处理。然后,经过适当筛选的数据集被用来训练具有交叉验证和测试功能的神经网络,这里使用的商业神经程序称为NeuroSolutions。通过跟踪和误差分析,获得了令人满意的精度。利用训练好的神经网络的权值矩阵,在Excel环境下生成函数,用公式生成器程序直接从输入参数预测输出参数。这些功能已经在一个用户友好的界面中使用,以便在一个简单而实时的模拟器中直接从输入参数获取输出参数。该人工神经网络模拟器开发的每个步骤的细节在后面的章节中都有描述。关于ANNs的工作,专门用于神经网络实现的先进商业软件为NeuroSolutions

 

Öresundskraft AB has a number of plantsthat produces mainly electricity and district heat to the city of Helsingborgand its surroundings. Västhamnsverket is the main plant of the company, with aninstalled capacity of 126 MW electricity and 186 MW heat. Västhamnsverket is ahybrid combined heat and power plant. In order to demonstrate an efficientcontrol-monitoring model, several ANN models have been developed for differentparts of Västhamnsverket. This master thesis is a part of a continuing ANNstudy applied at Västhamnsverket. The overall aim of this study is to developan ANN simulator for the whole steam process at Västhamnsverket. In order todevelop a better ANN model for the entire steam process, this process isdivided into two sub-modules, called case 1 and case 2. Consequently two caseshave been studied separately and two different ANN models have been developed.These are then linked to each other. The input and output parameters of each ofthese cases have been decided based on appropriate criteria as discussed later.Once the input and output parameters were decided, the dataset from theexisting plant were obtained from the plant. Data preprocessing before using itfor ANN training has been done in order to make the dataset as reliable aspossible. Then properly screened dataset has been used for training of ANNswith cross validation and testing using a commercial neural program calledNeuroSolutions. A satisfactory accuracy has been obtained by trail and error.Some functions have been generated in Excel environment using the weight matrixof the trained ANNs in order to predict output parameters directly from inputparameters by using a program called Formula Generator. These functions havebeen used in a user-friendly interface to obtain output parameters directlyfrom input parameters in an easy and real time simulator. The detail of eachsteps of the development of this ANN simulator has been described in subsequentsections. Regarding the work on ANNs, advanced commercial software specific forneural network implementations, viz. NeuroSolutions has been used.

 

1. 引言

2. 人工神经网络

3. NEUROSOLUTIONS

4. ÖRESUNDS AB

5. 数据处理

6. 案例研究

7. 接口

8. 结论

9. 未来工作展望


更多精彩文章请关注公众号:qrcode_for_gh_60b944f6c215_258.jpg



http://blog.sciencenet.cn/blog-69686-1226861.html

上一篇:[转载]【信息技术】【2017.02】基于运动车辆摄像记录的车牌位置检测
下一篇:[转载]【电信学】【2016】智能家居环境下物联网的安全机制

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...
扫一扫,分享此博文

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2020-5-30 05:18

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部