Abstract:

Turbine steam flow is an important parameter for analyzing turbine operating efficiency. In order to solve such problems as lack of detection information, poor reliability of traditional measurement method, high cost and difficult installation of non-contact measurement method etc., a soft-sensing model based on the cooperative quantum genetic algorithm particle swarm optimization (CQGAPSO) and a well-defined part-connected neural network (NN) was proposed. It could eliminate some adverse effects of approximation ability caused by redundant structure of NN. The amplitude-based coding method and cooperation mechanism improved the learning efficiency, approximation precision and generalization of NN. The soft-sensing experimental results of turbine steam flow showed that the soft-sensing model of turbine steam flow could obtain higher prediction accuracy and greater robustness than the all-connection NN and other models.