基于神经网络的轴流转桨式水轮机传递系数

建立水轮机数学模型是研究机组稳定性,优化机组运行的基础。工程实际中,对水轮机小扰动工况的研究多采用分段线性化模型,需要求取不同工况下水轮机的传递系数。然而,传统的求取方法计算量大,得到的传递系数数量有限且在轻载工况时误差较大。本文基于外特性法和神经网络求导,研究了轴流转桨式水轮机传递系数的计算方法。利用模型综合特性曲线、飞逸曲线以及边界条件,获取水轮机模型的原始数据,训练并得到反映水轮机流量特性和力矩特性的神经网络。对包含神经网络的数学表达式进行求导,推导传递系数的计算式,它是工况点的函数。最后,对比了该方法和曲线拟合法在一些工况点处求得的传递系数,分析了各自的特点。结果表明,本文提出的基于神经网络求导的水轮机传递系数求取方法精度较高,减少了计算量,且有利于全面的认识传递系数随工况的变化情况。

Transfer coefficients for Kaplan turbine based on neural network

Building the mathematic model of hydro-turbine is basic for the research on its stability and optimal operation. In engineering practices, piecewise-linear models are usually used to analyze the little disturbance of hydro-turbine,which requires the transfer coefficients in different working states. However,conventional methods have the shortcomings of more computing burden, limited number of transfer coefficients and larger errors in lightly loaded conditions. In this paper, a computation method of transfer coefficients for hydro-turbine is proposed based on characteristic curves and the derivation of neural networks. The original data of hydro-turbine model is acquired using model synthesis characteristic curves, runaway curves and boundary conditions,and the neural network is trained and obtained reflecting the characteristics of discharge and torque of hydro-turbine. The derivative of the mathematical expression that contains the neural network is computed to deduce the calculation formula of transfer coefficients, which should be the function of working states. Finally, the transfer coefficients calculated respectively by the proposed method and curve fitting are compared at some working states,and the characteristics of them are analyzed. The results indicate that the proposed method based on the derivative of neural networks has a higher accuracy,reduces the computation and is conducive to an overall recognition on the change of transfer coefficients along with working states.