燃煤电站锅炉在线多目标燃烧优化的研究
|
摘要
近年来,随着节能减排越来越受到关注,燃煤电站锅炉燃烧优化课题得到了广泛的研究。电站锅炉在线多目标燃烧优化问题是一个综合了测量装置、大型流体机械电耗、燃烧预测模型建模、可控参数寻优等多方面因素的问题,本文着重研究其中的建模部分,采用混合核函数构造支持向量机建立燃烧预测模型,通过遗传算法实现参数寻优,编写燃烧优化指导系统实现了相应的功能。
针对风、粉管内流场不均匀导致难以布置测点的问题提出了提高测量精度的措施;对风速、煤粉浓度、煤粉细度、煤质特性及送引风机的测量装置原理进行了介绍,分析了测量精度对燃烧优化的重要性。
采用最小二乘支持向量机建立燃烧预测模型。为了提高支持向量机回归模型的精度,对核函数及其参数的选择进行了研究,将混合核函数应用到了燃烧预测模型的建立中,研究发现对不同的预测目标和训练样本数目,都有各自最佳的核函数,总体上看,对于电站锅炉燃烧优化问题,采用径向基核与多项式核加法混合方式有较理想的拟合与预测精度;在参数选择方面,对于采用加法混合核的支持向量机,分别研究了正则化参数、径向基核宽度、多项式核阶次的作用,总结了参数选择的规律;此外还发现,训练样本应尽可能取到所有输入参数的上下限,否则可能出现较大的预测偏差。介绍了遗传算法的计算步骤,综合考虑了大型流体机械电耗、磨煤机电耗、锅炉热效率及NOx排放浓度等多方面因素,导出了适应度函数“折算供电煤耗”,并实现多目标优化。
编程实现了“300MW燃煤锅炉燃烧优化指导系统”。通过燃烧优化指导、切圆中心位置模拟、历史数据查询、模型在线更新等功能,为实现实时在线燃烧优化打下了基础。
Energy-saving and emission-reducing in recent years have been gaining increasing attention. As a consequence, the project of Combustion Optimization of Coal-fired utility boiler has been extensively studied. The research of online multi-objective Combustion Optimization of Coal-fired utility boiler is a comprehensive one, which includes the following elements: measuring device, the modeling of large-scaled fluid machinery electricity consumption and combustion prediction model, Controllable parameter optimization and so on. This paper focuses on the modeling part. The corresponding function is achieved by applying mixed kernel functions to establish combustion prediction model; realizing optimization through genetic algorithm, and then authoring combustion optimization guidance system.
As to the difficulty in locating the measuring spots which are caused by the ill-distributed state of the flow field inside primary air duct, the paper offers certain measures to enhance the measuring accuracy; it explains the principle of the measuring device of pulverized coal concentration, coal powder fineness, properties of coal quality and blower; and it also analyzes the importance of measuring accuracy to combustion optimization.
The least squares support vector machines are applied to establish the combustion prediction model. In order to enhance the accuracy of the regression model, kernel functions and the option of parameters are studied. The mixture of kernel function is applied in the establishing of combustion prediction model. The result is that there are corresponding optimized kernel functions as to different predicting aims and sample numbers.
Generally, the application of RBF and Polynominal and additive in optimizing the boiler combustion in power stations achieves the comparatively ideal prediction accuracy. As to the support vector machine in the perspective of the option of parameters, the function of regularization parameter, RBF width and polynominal degree are studied. The regular pattern of the option of parameters is concluded. Besides, the training samples should be widely spread within the upper and lower limits of the parameters, otherwise serious prediction error may appear.
This paper introduces the calculating steps of genetic algorithm. The convert net coal consumption of fitness function is also deduced on the basis of considering the following elements: Large-scale fluid machinery, electricity consumption of mills, heat effectiveness of boilers and the emission concentration of NOx, and so on. Multi-objective Combustion Optimization is achieved.
Eventually the combustion optimization guidance system of 300MW utility boiler is achieved by programming. The foundation is laid for realizing real-time online combustion optimization through the guidance of combustion optimization, simulation of the central location of the circle, the query of historical data, online updates of the model etc.
针对风、粉管内流场不均匀导致难以布置测点的问题提出了提高测量精度的措施;对风速、煤粉浓度、煤粉细度、煤质特性及送引风机的测量装置原理进行了介绍,分析了测量精度对燃烧优化的重要性。
采用最小二乘支持向量机建立燃烧预测模型。为了提高支持向量机回归模型的精度,对核函数及其参数的选择进行了研究,将混合核函数应用到了燃烧预测模型的建立中,研究发现对不同的预测目标和训练样本数目,都有各自最佳的核函数,总体上看,对于电站锅炉燃烧优化问题,采用径向基核与多项式核加法混合方式有较理想的拟合与预测精度;在参数选择方面,对于采用加法混合核的支持向量机,分别研究了正则化参数、径向基核宽度、多项式核阶次的作用,总结了参数选择的规律;此外还发现,训练样本应尽可能取到所有输入参数的上下限,否则可能出现较大的预测偏差。介绍了遗传算法的计算步骤,综合考虑了大型流体机械电耗、磨煤机电耗、锅炉热效率及NOx排放浓度等多方面因素,导出了适应度函数“折算供电煤耗”,并实现多目标优化。
编程实现了“300MW燃煤锅炉燃烧优化指导系统”。通过燃烧优化指导、切圆中心位置模拟、历史数据查询、模型在线更新等功能,为实现实时在线燃烧优化打下了基础。
Energy-saving and emission-reducing in recent years have been gaining increasing attention. As a consequence, the project of Combustion Optimization of Coal-fired utility boiler has been extensively studied. The research of online multi-objective Combustion Optimization of Coal-fired utility boiler is a comprehensive one, which includes the following elements: measuring device, the modeling of large-scaled fluid machinery electricity consumption and combustion prediction model, Controllable parameter optimization and so on. This paper focuses on the modeling part. The corresponding function is achieved by applying mixed kernel functions to establish combustion prediction model; realizing optimization through genetic algorithm, and then authoring combustion optimization guidance system.
As to the difficulty in locating the measuring spots which are caused by the ill-distributed state of the flow field inside primary air duct, the paper offers certain measures to enhance the measuring accuracy; it explains the principle of the measuring device of pulverized coal concentration, coal powder fineness, properties of coal quality and blower; and it also analyzes the importance of measuring accuracy to combustion optimization.
The least squares support vector machines are applied to establish the combustion prediction model. In order to enhance the accuracy of the regression model, kernel functions and the option of parameters are studied. The mixture of kernel function is applied in the establishing of combustion prediction model. The result is that there are corresponding optimized kernel functions as to different predicting aims and sample numbers.
Generally, the application of RBF and Polynominal and additive in optimizing the boiler combustion in power stations achieves the comparatively ideal prediction accuracy. As to the support vector machine in the perspective of the option of parameters, the function of regularization parameter, RBF width and polynominal degree are studied. The regular pattern of the option of parameters is concluded. Besides, the training samples should be widely spread within the upper and lower limits of the parameters, otherwise serious prediction error may appear.
This paper introduces the calculating steps of genetic algorithm. The convert net coal consumption of fitness function is also deduced on the basis of considering the following elements: Large-scale fluid machinery, electricity consumption of mills, heat effectiveness of boilers and the emission concentration of NOx, and so on. Multi-objective Combustion Optimization is achieved.
Eventually the combustion optimization guidance system of 300MW utility boiler is achieved by programming. The foundation is laid for realizing real-time online combustion optimization through the guidance of combustion optimization, simulation of the central location of the circle, the query of historical data, online updates of the model etc.
引文
[1]周国.燃煤电站锅炉中的低NOx燃烧技术[J].节能技术, 2005(1): 44-46
[2]阎维平.洁净煤发电技术[M].中国电力出版社, 2008: 56-70
[3]王庆. NeuSIGHT神经网络闭环燃烧优化控制系统在沙角B电厂应用前景分析[J].河南电力, 2005(8): 34-37
[4]牛拥军,闰国庆.优化燃烧技术在邹县电厂中的应用[J].发电设备, 2004增刊: 102-105
[5]王军,章正林,姜书敏. SmartProcess燃烧优化系统及其应用[J].电力设备, 2006, 7(2)
[6]吕剑虹,王建武,杨榕等.电厂锅炉燃烧控制系统优化[J].中国电力, 2001, 34(10): 44-47
[7]周昊,朱洪波.基于Internet/Intranet的火电厂锅炉低NOx燃烧优化指导系统[J]. 2003(2): 8-11
[8] Rudolph G. Convergence properties of canonical genetic algorithms [J]. IEEE Trans on Neural Net-works, 1994(5): 96-101
[9]王学年,王忠元.中速磨煤机一次风量指示误差原因分析[J].华东电力, 1999(4): 39-40
[10] Nil Kanth Singh, K. Ramamurthi. Formation of Coanda jet from sharp-edged swirl nozzle with base plate. Experimental Thermal and Fluid Science, 2009: 675-682
[11] Vittorio Betta a, FURIO Cascetta b, Marilena Musto a. Giuseppe Rotondo, Numerical study of the optimization of the pitch angle of an alternative jet fan in a longitudinal tunnel ventilation system. Tunnelling and Underground Space Technology, 2009(24): 164-172
[12] Koppernann C. A signal model for cross-correlation flow meters to analyse systematic errors[J]. Measurement, 1984(2): 129-133
[13] R W Ainsworth, S J Thorpe, and R J Manners. A new approach to flow-field measurement-A view of Doppler global velocimetry techniques, Heat and Fluid Flow, 1997: 116-130
[14]赵萍,刘洪利.基于快速互相关算法的测速系统的设计[J].信息系统工程, 2010.5: 26-27
[15]于树明.压差法测量一次风管送粉量[J].华中电力, 2000, vol13: 20-22
[16]王乃宁.颗粒粒径的光学测量技术及应用[M].北京:原子能出版社, 2000: 212-243
[17]蔡小舒,潘咏志,吴伟亮等.电厂煤粉粒径、浓度和速度的在线测量技术研究[J].动力工程, 1999(6): 466-470
[18] George F Luger.人工智能(第四版)[M].北京:机械工业出版社, 2004: 32-45
[19] Vladimir N. Vapnik. The Nature Of Statistical Learning Theory(Second Edition) [M]. New York Springer, 1999: 25-62
[20] Nello Cristianini, John Shawe-Taylor. An Introduction to Support Vector Machines and Other Kernel-based Learning Methods [M]. England, Cambridge University Press, 2000: 82-106
[21]王春林,周昊,周樟华等.基于支持向量机的大型电厂锅炉飞灰含碳量建模[J].中国电机工程学报,2005, 25(20): 72-77
[22] Johan A.K.Suykens, Carlos Alzate. Primal and dual model representations in kernel-based learning[J]. Statistics Surveys, 2010, Vol.4: 148-183
[23] Boser B, Guyon I, Vapnik V N. A training algorithm for optimal margin classifiers[J]. Fifth Annual Workshap on Computational Learning Theory, Pittsburgh ACM, 1992: 144-152
[24] Smola A J. Learning with kernels[D]. Belin:Informatik der Technischen Universitat, 1998
[25] G.F.Smits, E.M.Jordaan. Improved SVM Regression using Mixtures of Kernels[A].IEEE Proc of the 2002 Int Joint Conf on Neural Networks[C]. Honolulu, 2002(3): 2785-2790
[26]朱燕飞,伍建平,李琦等. MISO系统的混合核函数LS-SVM建模[J].控制与决策, 2005(4):417-425
[27] Liu Jingxu, LI Jin, TAN Yuejin. An empirical assessment on the robustness of support vector regression with different kernels[C]. International Conference on Machine Learning and Cybernetics. Guangzhou: Institute of Electrical and Electronics Engineers ComputerSociety, 2005: 4289-4294
[28] Keerthi S S, Lin CJ. Asymptotic behaviors of support vector machines with gaussian kernel[J]. Neural computation, 2003, 15(7): 1667-1689
[29] Jeng J T. Hybrid approach of selecting hyperparameters of support vectormachine for regression[J]. IEEE Trans. 2006, 36(3): 699-709
[30] Cherkassky V, MA Y. Practical selection of SVM parameters and noise estimation for SVM regression[J]. Neural Networks, 2004, 17(1): 113-126
[31]冯俊凯,沈幼庭.锅炉原理及计算.第三版[M].北京:科学出版社,2003: 563-587
[32]赵虹,沈利,杨建国等.利用煤的工业分析计算元素分析的DE-SVM模型[J],煤炭学报, 2010(10): 1721-1724
[33]贾鸿祥.制粉系统设计与运行.北京:水利电力出版社, 1995: 72-140
[34]周明,孙树栋.遗传算法原理及应用[M].北京:国防工业出版社. 1999: 18-32
[35] M Marseguerra, E.Zio, F. Cadini. Genetic algorithm optimization of a model-free fuzzy control system[J]. Annals of Nuclear Energy 2005, vol32:712-728
[36] Yun-Can Xue, Qi-Wen Yang, Zhi-Qian Mei. Combined algorithm for time-varying system based on the damped least-squares estimation algorithm andimproved genetic algorithm[J]. Machine Learning and Cybernetics. 2010, vol3: 619-622
[37] Yan Tai-shan. An Improved Genetic Algorithm and its Blending Application with Neural Network [J]. Intelligent Systems and Applications (ISA), 2010: 1-4
[38] Jiang Hua Wei, YANG Kai. Research of improved genetic algorithm for thresholding image segmentation based on maximum entropy [J]. Computer Science and Information Technology (ICCSIT). 2009: 117-120
[39] Eiben A E, Aarts E H, Van Hee K M. Global convergence of genetic algorithms: an infinite markov chain Analysis[C].Parallel Problem Solving from Na-ture.Berlin: Springer-Verlag, 1991: 4-12.
[40]毕惟红,任红民,吴庆标.一种新的遗传算法最优保存策略[J].浙江大学学报, 2006(1): 32-35
[2]阎维平.洁净煤发电技术[M].中国电力出版社, 2008: 56-70
[3]王庆. NeuSIGHT神经网络闭环燃烧优化控制系统在沙角B电厂应用前景分析[J].河南电力, 2005(8): 34-37
[4]牛拥军,闰国庆.优化燃烧技术在邹县电厂中的应用[J].发电设备, 2004增刊: 102-105
[5]王军,章正林,姜书敏. SmartProcess燃烧优化系统及其应用[J].电力设备, 2006, 7(2)
[6]吕剑虹,王建武,杨榕等.电厂锅炉燃烧控制系统优化[J].中国电力, 2001, 34(10): 44-47
[7]周昊,朱洪波.基于Internet/Intranet的火电厂锅炉低NOx燃烧优化指导系统[J]. 2003(2): 8-11
[8] Rudolph G. Convergence properties of canonical genetic algorithms [J]. IEEE Trans on Neural Net-works, 1994(5): 96-101
[9]王学年,王忠元.中速磨煤机一次风量指示误差原因分析[J].华东电力, 1999(4): 39-40
[10] Nil Kanth Singh, K. Ramamurthi. Formation of Coanda jet from sharp-edged swirl nozzle with base plate. Experimental Thermal and Fluid Science, 2009: 675-682
[11] Vittorio Betta a, FURIO Cascetta b, Marilena Musto a. Giuseppe Rotondo, Numerical study of the optimization of the pitch angle of an alternative jet fan in a longitudinal tunnel ventilation system. Tunnelling and Underground Space Technology, 2009(24): 164-172
[12] Koppernann C. A signal model for cross-correlation flow meters to analyse systematic errors[J]. Measurement, 1984(2): 129-133
[13] R W Ainsworth, S J Thorpe, and R J Manners. A new approach to flow-field measurement-A view of Doppler global velocimetry techniques, Heat and Fluid Flow, 1997: 116-130
[14]赵萍,刘洪利.基于快速互相关算法的测速系统的设计[J].信息系统工程, 2010.5: 26-27
[15]于树明.压差法测量一次风管送粉量[J].华中电力, 2000, vol13: 20-22
[16]王乃宁.颗粒粒径的光学测量技术及应用[M].北京:原子能出版社, 2000: 212-243
[17]蔡小舒,潘咏志,吴伟亮等.电厂煤粉粒径、浓度和速度的在线测量技术研究[J].动力工程, 1999(6): 466-470
[18] George F Luger.人工智能(第四版)[M].北京:机械工业出版社, 2004: 32-45
[19] Vladimir N. Vapnik. The Nature Of Statistical Learning Theory(Second Edition) [M]. New York Springer, 1999: 25-62
[20] Nello Cristianini, John Shawe-Taylor. An Introduction to Support Vector Machines and Other Kernel-based Learning Methods [M]. England, Cambridge University Press, 2000: 82-106
[21]王春林,周昊,周樟华等.基于支持向量机的大型电厂锅炉飞灰含碳量建模[J].中国电机工程学报,2005, 25(20): 72-77
[22] Johan A.K.Suykens, Carlos Alzate. Primal and dual model representations in kernel-based learning[J]. Statistics Surveys, 2010, Vol.4: 148-183
[23] Boser B, Guyon I, Vapnik V N. A training algorithm for optimal margin classifiers[J]. Fifth Annual Workshap on Computational Learning Theory, Pittsburgh ACM, 1992: 144-152
[24] Smola A J. Learning with kernels[D]. Belin:Informatik der Technischen Universitat, 1998
[25] G.F.Smits, E.M.Jordaan. Improved SVM Regression using Mixtures of Kernels[A].IEEE Proc of the 2002 Int Joint Conf on Neural Networks[C]. Honolulu, 2002(3): 2785-2790
[26]朱燕飞,伍建平,李琦等. MISO系统的混合核函数LS-SVM建模[J].控制与决策, 2005(4):417-425
[27] Liu Jingxu, LI Jin, TAN Yuejin. An empirical assessment on the robustness of support vector regression with different kernels[C]. International Conference on Machine Learning and Cybernetics. Guangzhou: Institute of Electrical and Electronics Engineers ComputerSociety, 2005: 4289-4294
[28] Keerthi S S, Lin CJ. Asymptotic behaviors of support vector machines with gaussian kernel[J]. Neural computation, 2003, 15(7): 1667-1689
[29] Jeng J T. Hybrid approach of selecting hyperparameters of support vectormachine for regression[J]. IEEE Trans. 2006, 36(3): 699-709
[30] Cherkassky V, MA Y. Practical selection of SVM parameters and noise estimation for SVM regression[J]. Neural Networks, 2004, 17(1): 113-126
[31]冯俊凯,沈幼庭.锅炉原理及计算.第三版[M].北京:科学出版社,2003: 563-587
[32]赵虹,沈利,杨建国等.利用煤的工业分析计算元素分析的DE-SVM模型[J],煤炭学报, 2010(10): 1721-1724
[33]贾鸿祥.制粉系统设计与运行.北京:水利电力出版社, 1995: 72-140
[34]周明,孙树栋.遗传算法原理及应用[M].北京:国防工业出版社. 1999: 18-32
[35] M Marseguerra, E.Zio, F. Cadini. Genetic algorithm optimization of a model-free fuzzy control system[J]. Annals of Nuclear Energy 2005, vol32:712-728
[36] Yun-Can Xue, Qi-Wen Yang, Zhi-Qian Mei. Combined algorithm for time-varying system based on the damped least-squares estimation algorithm andimproved genetic algorithm[J]. Machine Learning and Cybernetics. 2010, vol3: 619-622
[37] Yan Tai-shan. An Improved Genetic Algorithm and its Blending Application with Neural Network [J]. Intelligent Systems and Applications (ISA), 2010: 1-4
[38] Jiang Hua Wei, YANG Kai. Research of improved genetic algorithm for thresholding image segmentation based on maximum entropy [J]. Computer Science and Information Technology (ICCSIT). 2009: 117-120
[39] Eiben A E, Aarts E H, Van Hee K M. Global convergence of genetic algorithms: an infinite markov chain Analysis[C].Parallel Problem Solving from Na-ture.Berlin: Springer-Verlag, 1991: 4-12.
[40]毕惟红,任红民,吴庆标.一种新的遗传算法最优保存策略[J].浙江大学学报, 2006(1): 32-35