Forecasting Of The Feature-Based Hydroelectric Unit State

With the development of the mechanical industry, hydroelectric unit becomes more and more large, automatic, unit faults have great effect to the production. So it's more important to identify and predict the operation state and faults of the hydroelectric unit. To assure that the hydroelectric unit can keep safe, steady and long periodic operation, it's necessary to monitor the running status and predict the trend of long periodic development, and carry out the leading management and condition-based maintenance in response to the unit status. So reducing sudden fault and maintenance expense will bring economic performance to enterprises and society. First, this article analyses the vibration characteristics of the hydroelectric unit, and how to identify and diagnose the vibration faults. On the basis of it the turbine's operation state is divided into small-duty, vortex strip, steady-duty etc according to the change in load. Second, the eigenvalues of different states are extracted by using the wavelet packet decomposition principle and are putted into a Back Propagation network as a vector. So a rule of correspondence between the eigenvalues and the turbine's operation state is established. Finally, according to the historical and present vibration data, a prediction model is created. With the variation trend the unit state(normal, exception or fault) in the future can be predicted. This result is compared with vibration evaluation criterion and reflects the variation rule of unit state. So whether the vibration of the unit is in control range or out of control can be checked out, and whether the unit is abnormal can be found early. Then resolution should be adopted and it's beneficial for maintenance decision and process control.