Vibration Response Prediction Research Of Overflow Powerhouse Structure

Hydropower station is a complex structure and has a particularly large number ofvibration sources. These characteristics make vibration problem of hydropowerstructure become very common during runtime.Hydropower structure vibration hassignificant influence on instruments and equipments,staff health and stability ofbuilding operation.Using limited monitoring data to get a purpose of comprehensivegrasp of hydropower station become a new research subject.This study based on measured data without considering precise mathematicalmodel. Through data of unit and tail water pulsation monitoring, nonlinear mappingof vibration characteristics, to predict vibration of powerhouse structure.Used flies optimization algorithm (FOA) to optimize the spread value ofgeneralized regression neural network (GRNN). And combined with other neuralnetwork prediction model (BP, ELMAN) to make a contrast research on deformationprediction problem about structure vibration response of hydropower station. Get afinally conclusion that FOA-GRNN is superior to BP and ELMAN on predictionability and learning speed.It illustrates that FOA-GRNN on vibration prediction isfeasible and superior.Particle swarm optimization algorithm (PSO) is easy to fall into local extremumand premature convergence. To make up the defect of PSO，put forward a new kind ofPSO which based on survival of the fittest and step by step selection(referred to asSSPSO). Strong optimization ability of SSPSO has been proved through typical testfunctions. Then use SSPSO to optimize smoothness value of GRNN. It makes full usethe advantages of optimization ability of SSPSO and fewer parameters of GRNN.Then set up vibration response prediction model which based on research data.Predicted results show that search capability of SSPSO has been greatly improvedcompared with PSO. At the same time，prediction accuracy、convergence performanceand generalization ability of SSPSO-GRNN is better than others.Combines particle swarm optimization algorithm (PSO),genetic algorithm (GA)and fruit flies optimization algorithm (FOA) with radial basis neural network (RBF)separately, to optimize the spreading parameter of RBF. Then build PSO-RBF,GA-RBF and FOA-RBF neural network and andlysis vibration of hydraulic structure.The study leads to the conclusions that PSO-RBF, GA-RBF and FOA-RBF have agood predictive effect. They are appropriate for prediction research of vibration response of powerhouse structures. Whereby， FOA-RBF is the most steady andaccurate.To sum up, the hybrid models of intelligent algorithm with neural network arenot only easy to be understood and grasped, but also have a high precision.It is verysuitable for vibration response prediction of hydropower station. New methods andnew ideas are provided for other power station vibration research and will enhanceintelligent monitoring of hydraulic structure.