CAJ | 학술논문

为提高基于模型综合特性曲线的混流式水轮机非线性模型精度，提出了一种适合描述水轮机综合特性的Adaboost _LMBP集成神经网络模型构建方法。该方法首先分别根据混流式水轮机单位力矩、单位流量特性数据，利用Levenberg －Marquardt算法反复训练出若干个BP神经网络弱学习器，然后采用Adaboost集成学习算法对全部BP神经网络弱学习器进行组合，最终构建出水轮机单位力矩、单位流量特性的Adaboost_LMBP集成神经网络模型。计算结果表明，相较于一般的神经网络模型，混流式水轮机综合特性的Adaboost_LMBP集成神经网络模型具有更高的拟合精度，更好的泛化性能，能够有效提升基于模型综合特性曲线的混流式水轮机非线性模型的计算精度。
위제고기우모형종합특성곡선적혼류식수륜궤비선성모형정도，제출료일충괄합묘술수륜궤종합특성적Adaboost _LMBP집성신경망락모형구건방법。해방법수선분별근거혼류식수륜궤단위력구、단위류량특성수거，이용Levenberg －Marquardt산법반복훈련출약간개BP신경망락약학습기，연후채용Adaboost집성학습산법대전부BP신경망락약학습기진행조합，최종구건출수륜궤단위력구、단위류량특성적Adaboost_LMBP집성신경망락모형。계산결과표명，상교우일반적신경망락모형，혼류식수륜궤종합특성적Adaboost_LMBP집성신경망락모형구유경고적의합정도，경호적범화성능，능구유효제승기우모형종합특성곡선적혼류식수륜궤비선성모형적계산정도。
In order to improve the accuracy of Francis turbine nonlinear model based on model synthetic characteristics curve ,this pa-per proposes a design method of Adaboost_LMBP ensemble neural network model which is suitable to describe synthetic characteris-tics for Francis turbines .Firstly ,this method uses Levenberg-Marquardt algorithm to train some BP neural network weak learners according to turbine unit torque and flow characteristic data .Then Adaboost ensemble learning algorithm is used to assemble all BP neural network weak learners .Finally ,turbine unit torque and flow characteristic Adaboost_LMBP ensemble neural network model is completed .The calculation results show that Adaboost_LMBP ensemble neural network model of synthetic characteristics for Francis turbines has better fitting accuracy and generalization performance than general neural network models and it can improve the accuracy of Francis turbine nonlinear models based on model synthetic characteristics curve effectively .