基于神经网络的离心泵性能预测研究与实现
发布时间:2019-03-25 07:27
【摘要】:泵的性能预测研究就是根据泵的叶轮、蜗壳、导叶等过流部件的几何参数,分析内部流动特征,以此预测泵的性能,是在泵产品设计中必不可少的重要环节,具有缩短研发周期、降低开发成本和提高产品设计质量等优点。因此开展泵的性能预测研究一直都是从事泵领域研究专家学者一个十分重要的课题。但是,到目前为止,泵的性能预测研究的结果还存在精度不高、不能满足工程实际需要的不足。因此对离心泵的性能研究不但具有重要的学术价值和社会效益,而且对研究其他泵的性能提供了可资借鉴的依据。本文的主要研究内容和成果有:(1)详细阐述了目前国内外针对泵性能预测研究的现状,重点描述了水力损失法、流场计算法及神经网络法这三种泵性能预测的方法,根据这三种方法的优缺点,确定本文里研究内容。(2)通过对离心泵性能分析,阐述各参数对离心泵性能的影响,为后面的研究打下基础;(3)介绍了离心泵性能预测的相关技术,即通过分析Matlab和VC6.0软件及其神经网络工具在进行数据处理、非线性拟合、动态仿真等的巨大优势,因此本文利用神经网络控制作为本课题的研究方向;(4)采用神经网络的两种算法:使用VC6.0实现了贝叶斯BP神经网络算法,使用Matlab实现了GA-RBF算法,并分别对实现离心泵性能预测进行设计;(5)利用GA-RBF神经网络建立离心泵性能预测模型来实现离心泵的性能预测,具体阐述了其实现过程,并根据选取的57组单级单吸离心泵的设计参数和试验参数进行仿真验证,结果表明离心泵性能GA-RBF神经网络预测模型与原有的离心泵性能RBF神经网络预测模型同样有效,并且设置参数更简单、更方便;(6)利用贝叶斯BP神经网络建立离心泵性能预测模型来实现离心泵的性能预测,具体阐述实现过程,并根据选取的57组单级单吸离心泵的设计参数和试验参数进行仿真验证,结果表明离心泵性能BRBP神经网络预测模型与原有的离心泵性能LMBP神经网络预测模型和离心泵性能RBF神经网络预测模型同样有效,并且设置参数更简单、更方便,是一种比较有前途的离心泵性能预测方法。
[Abstract]:The prediction of pump performance is based on the geometric parameters of the impeller, volute, guide vane and other flow components, and analyzes the internal flow characteristics to predict the pump performance, which is an essential and important link in the design of pump products. It has the advantages of shortening R & D cycle, reducing development cost and improving product design quality. Therefore, the research of pump performance prediction has always been a very important topic in the field of pump research experts and scholars. However, up to now, the results of pump performance prediction are still not high precision and can not meet the practical needs of the project. Therefore, the research on the performance of centrifugal pump not only has important academic value and social benefit, but also can be used for reference to study the performance of other pumps. The main contents and achievements of this paper are as follows: (1) the present situation of pump performance prediction at home and abroad is described in detail, and the hydraulic loss method, flow calculation method and neural network method are described in detail. According to the advantages and disadvantages of these three methods, the research contents in this paper are determined. (2) through the analysis of the performance of centrifugal pump, the influence of each parameter on the performance of centrifugal pump is expounded, which lays a foundation for further research; (3) the related technology of centrifugal pump performance prediction is introduced, that is, by analyzing the great advantages of Matlab and VC6.0 software and their neural network tools in data processing, nonlinear fitting, dynamic simulation, etc. Therefore, this paper uses neural network control as the research direction of this subject. (4) two kinds of neural network algorithms are used: Bayesian BP neural network algorithm is implemented by using VC6.0, GA-RBF algorithm is implemented by Matlab, and the performance prediction of centrifugal pump is designed. (5) the performance prediction model of centrifugal pump is established by using GA-RBF neural network to predict the performance of centrifugal pump, and the realization process of centrifugal pump is described in detail. According to the design parameters and test parameters of 57 single-stage and single-suction centrifugal pumps, the simulation results show that the GA-RBF neural network prediction model of centrifugal pump performance is as effective as the original RBF neural network prediction model of centrifugal pump performance. And setting parameters is simpler and more convenient; (6) the performance prediction model of centrifugal pump is established by using Bayesian BP neural network to predict the performance of centrifugal pump, and the realization process is described in detail. The simulation results are verified according to the design parameters and test parameters of 57 single-stage and single-suction centrifugal pumps. The results show that the BRBP neural network prediction model of centrifugal pump performance is as effective as the original LMBP neural network prediction model of centrifugal pump performance and the RBF neural network prediction model of centrifugal pump performance, and setting parameters is simpler and more convenient. It is a promising method to predict the performance of centrifugal pump.
【学位授予单位】:电子科技大学
【学位级别】:硕士
【学位授予年份】:2014
【分类号】:TH311
[Abstract]:The prediction of pump performance is based on the geometric parameters of the impeller, volute, guide vane and other flow components, and analyzes the internal flow characteristics to predict the pump performance, which is an essential and important link in the design of pump products. It has the advantages of shortening R & D cycle, reducing development cost and improving product design quality. Therefore, the research of pump performance prediction has always been a very important topic in the field of pump research experts and scholars. However, up to now, the results of pump performance prediction are still not high precision and can not meet the practical needs of the project. Therefore, the research on the performance of centrifugal pump not only has important academic value and social benefit, but also can be used for reference to study the performance of other pumps. The main contents and achievements of this paper are as follows: (1) the present situation of pump performance prediction at home and abroad is described in detail, and the hydraulic loss method, flow calculation method and neural network method are described in detail. According to the advantages and disadvantages of these three methods, the research contents in this paper are determined. (2) through the analysis of the performance of centrifugal pump, the influence of each parameter on the performance of centrifugal pump is expounded, which lays a foundation for further research; (3) the related technology of centrifugal pump performance prediction is introduced, that is, by analyzing the great advantages of Matlab and VC6.0 software and their neural network tools in data processing, nonlinear fitting, dynamic simulation, etc. Therefore, this paper uses neural network control as the research direction of this subject. (4) two kinds of neural network algorithms are used: Bayesian BP neural network algorithm is implemented by using VC6.0, GA-RBF algorithm is implemented by Matlab, and the performance prediction of centrifugal pump is designed. (5) the performance prediction model of centrifugal pump is established by using GA-RBF neural network to predict the performance of centrifugal pump, and the realization process of centrifugal pump is described in detail. According to the design parameters and test parameters of 57 single-stage and single-suction centrifugal pumps, the simulation results show that the GA-RBF neural network prediction model of centrifugal pump performance is as effective as the original RBF neural network prediction model of centrifugal pump performance. And setting parameters is simpler and more convenient; (6) the performance prediction model of centrifugal pump is established by using Bayesian BP neural network to predict the performance of centrifugal pump, and the realization process is described in detail. The simulation results are verified according to the design parameters and test parameters of 57 single-stage and single-suction centrifugal pumps. The results show that the BRBP neural network prediction model of centrifugal pump performance is as effective as the original LMBP neural network prediction model of centrifugal pump performance and the RBF neural network prediction model of centrifugal pump performance, and setting parameters is simpler and more convenient. It is a promising method to predict the performance of centrifugal pump.
【学位授予单位】:电子科技大学
【学位级别】:硕士
【学位授予年份】:2014
【分类号】:TH311
【参考文献】
相关期刊论文 前3条
1 宫赤坤,闫雪;基于RBF神经网络的预测控制[J];上海理工大学学报;2005年05期
2 朱海燕;朱晓莲;黄,
本文编号:2446769
本文链接:https://www.wllwen.com/kejilunwen/jixiegongcheng/2446769.html