基于进化神经网络的压气机结垢性能退化评估

采用进化神经网络方法,通过测量参数对压气机结垢性能退化模式进行了定量监控和评估。运用粒子群算法优化径向基函数(Radial Base Function,RBF)神经网络的初始权值,即由神经网络训练样本所得到的实际和期望的输出之间的误差平方和构造适应度函数,对RBF神经网络的隐层中心、半径以及输入输出权值进行全局寻优搜索,设计了进化RBF神经网络,并对模拟得到的压气机结垢的样本进行训练和测试。结果表明:进化RBF神经网络的模式识别能力比普通RBF神经网络的要强,对燃气轮机性能退化评估和健康管理具有重要理论意义和应用价值。

Performance Deterioration Evaluation of Compressor Fouling Based on Evolving Neural Network

Based on evolving neural network method, the performance deterioration mode of compressor fouling was quantitative supervised and evaluated by measurement parameters. The initial weighted values of Radial Base Function （RBF） neural network were optimized by particle swarm algorithm, which was the sum of error squares obtained by the actual and expectation output of neural network training samples as the fitness function. The global optimization search of hidden center, radius, input and output weighted values for RBF neural network was conducted. The evolving RBF neural network was designed, the simulated samples of compressor fouling was trained and tested. The results show that the mode identification capability of evolving RBF neural network is better than simple RBF artificial neural network. It is important theoretical and application value for gas turbine performance deterioration evaluation and health management.