Research On Running State Recognition Method Of Hydraulic Turbine Based On Fruit Fly Optimization Algorithm Improved Probabilistic Neural Network

Due to the adjustment of energy structure in China,the application of unstable energy such as wind power,hydropower and so on is more and more extensive.With the rapid development of such energy,both the power generation and the complexity of power grid operation increase,so it is particularly important to ensure the safe and stable operation of power grid.The turbine generator unit has excellent functions of peak load regulation,frequency modulation and fault standby.Taking the full use of the turbine unit function can effectively reduce the pressure of the power grid,improve the quality of power,alleviate the peak valley contradiction and other problems that are easy to occur in the power grid,guarantee stable operation of the power grid.With the structure expanding of the water turbine,so the energy conversion relationship of each part is becoming more and more complex.In the operation process,if a part of the unit fails will greatly affect the operation of the whole water turbine unit,resulting in huge economic losses or even a serious disaster.Therefore,to ensure the hydraulic turbine units safe operate,timely diagnosing the faults and predicting the state of hydraulic turbine units are the major practical problems that need to be solved urgently.The paper introduces the self-information theory to explore the contribution of different parameters in the state classification of the hydraulic turbine,and extracts characteristic parameters of pressure fluctuation signal At the same time,the smooth factor in the Probabilistic Neural Network(PNN)is optimized by the Fruit Fly Optimization Algorithm(FOA),and the classification model of turbine operation state based on FOA-PNN is established to realize the real-timemonitoring of turbine fault.First,in order to reveal the relationship between turbine running state and energy characteristics,wavelet analysis were used to study the frequency distribution and energy characteristics of pressure pulsation collected at the tail pipe of the turbine.The results show that with the increase of internal cavitation,the ratio of low frequency band energy decreases,the signal energy of high frequency band increases,and the energy intensifies heavily.Therefore,the study of the spectrum distribution and energy characteristics of the pressure pulsation signal is of reference significance for judging the operation failure of the turbine.Secondly,according to the complex characteristics of the hydraulic turbine structure,the contribution degree of the working condition variables to the pressure fluctuation signal of the hydraulic turbine is calculated by using the mutual information theory,and the unit speed,the guide vane opening,the unit flow and the working head are obtained as the main relevant parameters of the pressure fluctuation signal.The vibration trend prediction model of the wavelet neural network is built to intuitively show the nonlinear correlation between the pressure fluctuation signal and the main working condition parameters.Predicting the vibration trend of hydraulic turbine.Using Support vector machine(SVM)model to predict the operation state of hydraulic turbine.Finally,in order to realize the real-time state prediction of hydraulic turbine,the PNN is used to learn the simple,fast and powerful classification ability,train the vibration characteristic vector of hydraulic turbine,build the PNN model,and realize the operation state prediction of hydraulic turbine;meanwhile,in view of the impact of the only parameter,smoothing factor ?,on the classification accuracy of PNN network,this paper uses the FOA to optimize The model of FOA-PNN is constructed to verify the effectiveness of the model for the classification of unit operation state.The accuracy of FOA-PNN model is verified by comparing with the prediction accuracy of SVM and PNN network model.