摘要
首次采用自组织特征映射(SOM)网络结合BP神经网络方法建立了汽轮机功率模型,利用SOM网络的聚类功能,解决了传统样本提取方法正交性和完备性差的局限性。在合肥电厂125MW机组实际运行数据的基础上进行了仿真,仿真结果表明该模型的预测计算结果与实际数据误差在1.5%之内,大部分误差不超过1%。利用该模型和建立的循环水系统功耗神经网络模型可确定不同工况下的真空运行最优值(基准值),为凝汽器真空运行最优值的确定提供了一个全新的方法。同时利用建立的神经网络模型可计算热力系统几个主要参数偏离基准值的能损偏差,与传统的运行参数基准值模型和能损偏差分析方法相比,该模型具有明确的设备针对性。
The Artificial Neural Network （ANN） model of Self Organization Feature Map （SOM） and BP network were introduced to set up the model of steam turbine power output, with the aid of SOM＇s clustering ability, the limitation of conventional collection of training samples were addressed. Based on the actual operating data in 125 MW unit of Hefei Power Plant, with the trained model, the error of the power output predicted by this model and the actual power generated by steam turbine is less 1.5%, most of which is no more than 1%. With the established ANN models of steam turbine power output and the power needed in circulating water system, the optimum vacuum value （reference value） of condenser as well as the energy loss deviation of several main operation parameters from reference can be determined. Compared with conventional model, the, model established in this paper is of good pertinence to equipment as a resuh of practical sampling data collection on running equipment.
出处
《中国电力》
CSCD
北大核心
2005年第12期47-50,共4页
Electric Power
