基于工程经验和RBF神经网络的水轮机综合特性曲线拓展和重构  

Expansion and Reconstruction of Comprehensive Characteristic Curve for Water Turbine Based on Engineering Experience and RBF Neural Network

作　　者：吴嵌嵌 张雷克[2] 马震岳[3] WU Qianqian;ZHANG Leike;MA Zhenyue(Civil Engineering Institute,Sanjiang University,Nanjing,Jiangsu Province,210012,China;College of Water Resources Science and Engineering,Taiyuan University of Technology,Taiyuan,Shanxi Province,030024,China;School of Hydraulic Engineering,Faculty of Infrastructure Engineering,Dalian University of Technology,Dalian,Liaoning Province,116024,China)

机构地区：[1]三江学院土木工程学院,南京210012 [2]太原理工大学水利科学与工程学院,太原030024 [3]大连理工大学建设工程学部水利工程学院,大连116023

出　　处：《应用基础与工程科学学报》2019年第5期996-1007,共12页Journal of Basic Science and Engineering

基　　金：国家自然科学基金项目(51709196);国家重点研发计划项目(2016YFC0401905);江苏省高等学校自然科学研究面上项目(18KJB570002)

摘　　要：水轮机特性曲线是描述水轮机运行过程中运行参数变化及相互关系的重要数据图.采用径向基函数(Radial Basis function,RBF)神经网络拓展和拟合水轮机特性曲线的方法可避免用具体的数学表达式表示该特性曲线高度非线性的函数关系,而是通过对离散样本点的学习和训练来获得水轮机全特性曲线.在使用RBF方法进行拟合数据和重构曲面之前,本文利用边界条件和工程经验将已知工况区域的特性曲线向小开度区域和低效率区域进行了有效拓展,大幅增加了RBF神经网络的学习样本数量,从而进一步提高了RBF方法拟合水轮机全特性曲线的精度和可靠度.以HL180水轮机为例,对拓展和拟合水轮机综合特性曲线的具体过程进行了详细介绍.计算结果表明,同仅采用RBF神经网络方法相比,采用本文方法得到的水轮机全特性曲线具有更高的精度和可靠度.The comprehensive characteristic curve is a graph with some data groups,which is of great importance and used to describe the nonlinear relationship among the variables during the operation process of water turbine.Taking use of Radial Basis Function(RBF) neural network to expand and fit characteristic curve of water turbine,can be achieved easily through the learning and training of discrete sample points,instead of utilizing exact mathematical expressions to represent the highly nonlinear relationships for corresponding curve.Before fitting and recomposing the curve with help of RBF method,an effective expansion to the small opening gate and low efficiency area for the known region is done,based on boundary conditions of characteristic curve and engineering experience.The preliminary work dramatically increases the quantity of learning sample points for RBF neural network,furthermore,which improves the precision and reliability of the fitting curve by employing RBF method.Take HL180 water turbine as an example(Francis type,specific speed is 180 m·kW),the detailed process for curve expansion and reconstruction is introduced.The calculation results reveal that compared with the individual RBF method,the expanded and fitted comprehensive characteristic curve obtained by the way used in this paper is more precise and reliable.

关 键 词：水轮机特性曲线 拓展 拟合 径向基函数神经网络 

分 类 号：TK730.2[交通运输工程—轮机工程] TV724.1[动力工程及工程热物理—流体机械及工程]