﻿ 基于BP神经网络和SVM的电厂粉尘浓度在线监测 Online Monitoring of Dust Concentration in Power Plant Based on BP Neural Network and SVM

Advances in Energy and Power Engineering
Vol.04 No.04(2016), Article ID:18304,8 pages
10.12677/AEPE.2016.44013

Online Monitoring of Dust Concentration in Power Plant Based on BP Neural Network and SVM

Yifan Zhao, Zhongguang Fu

Key Laboratory of Condition Monitoring and Control for Power Plant Equipment, North China Electric Power University, Beijing

Received: Jul. 21st, 2016; accepted: Aug. 14th, 2016; published: Aug. 17th, 2016

ABSTRACT

For the purpose of achieving online monitoring of dust concentration, the online monitoring parameters in DCS system are adopted to analyze the factors which influence the concentration of smoke dust, and the BP neural network and support vector machine are used to propose an on-line monitoring method for dust concentration proposed. Simulation and prediction are based on the operating data of a power plant 600 MW unit. The simulation results show that the prediction accuracy of the two models is both more than 96%, and the prediction error of BP model is less than 4%, while the error of SVM model is even less than 2.5%. On the whole, these two models are ideal for dust concentration monitoring, but the accuracy of the SVM model is relatively higher, and it has higher generalization ability, and is more stable. Therefore, it can be a kind of effective method for on-line monitoring.

Keywords:Dust Concentration, On-Line Monitoring, BP Neural Network, Support Vector Machine (SVM)

1. 引言

2. 基于BP神经网络的粉尘浓度在线监测模型

2.1. BP神经网络

BP神经网络做为一种机器学习方法，内嵌了很多数学思想和学习准则，基本思想是将输入数据经过多次迭代运算得到输出值。正向运算过程：已知量由输入层输入，经过隐含层的学习，由输出层得到预测值。反向判断过程：将预测值与期望值对比，误差反向传播，对各系数进行修改，重复迭代以期获得较高精度 [5] 。BP神经网络算法流程分为神经网络的构建、训练和预测预测三步，如图1所示。

BP神经网络的拓扑结构如图2所示。

2.2. 训练集与测试集

Figure 1. The algorithm flow of BP neural network

Figure 2. BP neural network topology map

2.3. 模型构建与拟合

3. 基于SVM的粉尘浓度在线监测模型

3.1. 支持向量机(SVM)

3.2. 模型构建与拟合

Figure 3. Forecast output of BP neutral network

Figure 4. The prediction error of BP neutral network

Figure 5. Forecast output of SVM

Figure 6. The Prediction error of SVM

Table 1. Model test results

Table 2. Comparison of modeling results

4. 模型结果分析及比较

5. 结论

BP神经网络结构简单，参数设定方便，针对多数案例都有较好的处理效果，应用广泛 [12] 。支持向量机(SVM)结构稳定，引入了核函数，将变量映射到高维空间，优化了求解过程 [13] [14] 。本文用两种方法分别建立了燃煤电厂排烟粉尘浓度在线监测模型。两种模型的预测精度高于96%，符合期望，对于工程实践有很大的指导意义。

Online Monitoring of Dust Concentration in Power Plant Based on BP Neural Network and SVM[J]. 电力与能源进展, 2016, 04(04): 95-102. http://dx.doi.org/10.12677/AEPE.2016.44013

1. 1. 唐娟. 粉尘浓度在线监测技术的现状及发展趋势[J]. 矿业安全与环保, 2009, 36(5): 69-71, 74.

2. 2. Yang, S.H., Wang, X.Z., Mcgreav, Y.C., et al. (1998) Soft-Sensor Based Predictive Control of Industrial Fluid Catalytic Cracking Processes. Institution of Chemical Engineerings Trans. IchemE (S0263-8762), 76, 499-508.

3. 3. 段中兴, 嵇启春. 催化剂粉尘浓度软测量建模研究与应用[J]. 系统仿真学报, 2008, 20(14): 3899-3902, 3906.

4. 4. 靳涛. 火电机组反向建模方法的研究[D]: [博士学士学位]. 北京: 华北电力大学(北京), 2011: 20-27.

5. 5. Matlab中文论坛. Matlab 神经网络30个案例分析[M]. 北京: 北京航空航天大学出版社, 2010: 12-21.

6. 6. 孙久春. 烟尘(粉尘)浓度在线监测仪的研制[D]: [硕士学士学位]. 沈阳: 东北大学, 2004.

7. 7. 丁世飞, 齐丙娟, 谭红艳. 支持向量机理论与算法研究综述[J]. 电子科技大学学报, 2011, 40(1): 2-10.

8. 8. 牛东晓, 谷志红, 邢棉, 王会青. 基于数据挖掘的SVM短期负荷预测方法研究[J]. 中国电机工程学报, 2006, 26(18): 6-12.

9. 9. 曹宏芳, 付忠广, 齐敏芳. PSO-SVM软测量方法在火电厂煤质发热量测量中的应用[J]. 热能动力工程, 2014(6): 731-735, 764-765.

10. 10. 蔡杰进, 马晓茜. 基于SVM的燃煤电站锅炉飞灰含碳量预测[J]. 燃烧科学与技术, 2006, 12(4): 312-317.

11. 11. 孙俊, 王艳, 金夏明, 毛罕平. 基于MSCPSO混合核SVM参数优化的生菜品质检测[J]. 农业机械学报, 2013, 44(9): 209-213, 218.

12. 12. 秦富童, 岳丽华, 万寿红. 应用BP神经网络的目标识别效果评估[J]. 计算机工程与应用, 2010, 46(5): 148-150, 156.

13. 13. 王恒, 花国然, 贾民平, 陈左亮. 基于LS-SVM和GM的球磨机料位动态软测量[J]. 热力发电, 2015(1): 77-81.

14. 14. 宫唤春. 基于模糊支持向量机的凝汽器故障诊断[J]. 热力发电, 2015(6): 98-101.

15. 15. 李勇, 王建君, 曹丽华. 汽轮机主蒸汽流量在线监测方法研究[J]. 热力发电, 2011, 40(4): 33-36, 40.