提高汽轮机排汽焓在线预测精度的GA-Elman神经网络算法

A GA-Elman neural network algorithm which can improve online prediction accuracy of exhaust enthalpy for steam turbines

利用遗传算法(GA)的良好寻优能力对汽轮机排汽焓动态递归(Elman)神经网络进行了优化,建立了GA-Elman神经网络预测模型,并以某电厂350MW机组为例进行了汽轮机排汽焓的在线计算。结果表明:GA-Elman神经网络预测模型克服了传统Elman神经网络利用梯度下降法进行训练所具有的易陷入局部极小值、收敛速度慢、精度低等缺点,提高了预测精度和收敛速度,较适合现场应用。

By taking use of good optimizing ability of the genetic algorithm （GA）,the dynamic recurrent neural network （Elman）of steam turbine exhaust enthalpy was optimized and a GA-Elman neural network prediction model was established.Taking a 350 MW unit steam turbine as the example,online calculation for the turbine exhaust enthalpy was conducted by applying this model.The results show that：this GA-El-man neural network model overcomes such problems as easy to fall into local minimum,slow convergence speed and low precision that the conventional Elman neural network （which applies gradient descent meth-od to conduct training）has.So this model enhances the prediction accuracy and convergence speed,which is more suitable for field application.