基于GMDH神经网络的超超临界机组过热蒸汽温度预测模型及仿真研究

GMDH neural network based predictive modeling and simulation for superheated steam temperature of 1000MW ultra-supercritical units

由于超超临界1 000MW机组过热蒸汽温度控制对象具有大滞后、非线性、动态参数随工况变化大等特点,使得传统的控制方法难以适应过热蒸汽温度的控制,出现过热蒸汽温度波动大,甚至超温等问题。对此,采用数据处理群集方法(GMDH)神经网络建立了过热蒸汽温度动态预测模型,以预测过热蒸汽温度的变化趋势。仿真结果表明,基于GMDH神经网络的过热蒸汽温度预测效果优于线性神经网络和BP神经网络,具有较好的移植性和实用性。

The superheated steam temperature （SST） of 1 000 MW ultra-supercritical units has the characteristics of large time-delay,nonlinear and dynamic parameters heavily relying on the load.So the SST has a large fluctuation under the conventional control scheme,which causes over-temperature phenomenon frequently.In order to monitor the SST,the group method of data handling （GMDH） based prediction model was applied to predict the SST.The dynamic prediction model was established by taking previous inlet temperature and ouJet temperature of the final superheater as input signal.The current outlet temperature was taken as expected signal,and the operational data was as training set.To verify the efficiency of the proposed prediction model,the line Network,BP and GMDH model were applied on the same operational data to get one-step-ahead prediction of the SST.The simulation results show that the GMDH model has a better predication than the others.