水轮机综合特性曲线BP神经网络拟合方法研究

Study on synthetic characteristic curve processing of Francis turbine combined with BP neural network

针对目前水轮机综合特性曲线处理中存在的问题,介绍了一种采用BP神经网络的解决方法,其核心思想是将水轮机非线性特性转换为可用于实时仿真的基于神经网络的力矩和流量特性。论文首先介绍了该方法的整体设计思路和求解过程;随后详细说明了其具体细节和实现步骤:样本数据读取和延拓方法以及神经网络的选取原则、训练过程;最后结合实际水轮机给出了处理结果,并分析了误差产生的原因。本文的研究成果为水轮机综合特性曲线拟合提供了一种新的途径。

This paper describes a new BP neural network method that expresses turbine nonlinear characteristics with torque and flow neural networks in numerical simulation to solve the existing problems in processing the synthetic characteristic curve of hydraulic turbines. The overall design idea and the solving process are demonstrated, and followed by an analysis of the specific details and implementation steps, i.e. how to retrieve and extend sample data and how to select a neural network and its training process. A case analysis on a practical curve and the likely causes of its calculation errors are included. This work provides a new perspective for synthetic characteristic curve fitting.