Research On Type Recognition Of Cavitation And Early Warning Of Incipient Cavitation Of Hydraulic Turbine

Hydro power has many advantages,such as pollution-free and high security.It has been used by so many countries in the world,and China is a big country of hydroelectric power generation.Hydraulic turbine is the core component of hydro power station.Cavitation has always been one of the main threats to the safe operation of hydraulic turbine units.This paper focuses on the recognition and classification of the hydraulic turbines cavitation signals characteristics,and the early warning of the initial cavitation,then uses digital signal processing,deep learning and other research methods.This paper mainly completes the following parts:(1)Aiming at the characteristics of cavitation signals in different locations of hydraulic turbines are difference,a cavitation feature extraction method based on improved sign dynamic entropy(IMSDE)is proposed.Then the performance of the algorithm is analyzed and tested by using a segment of cavitation signal.It is concluded that the IMSDE algorithm can better reflect the signal difference and better suppress environmental noise.Then,experiments are carried out to verify the performance by using the cavitation signals under four different working conditions.The features extracted by the IMSDE algorithm are optimized by Laplace Score(LS).Finally,the optimized features are classified and recognized by Least Square Support Vector Machine(LSSVM).It is verified that the IMSDE algorithm has better feature extraction ability and can complete the classification and identification work of water turbine cavitation occurring at different positions.(2)Airfoil cavitation of hydro-turbine is one of the most common and main phenomena that endanger turbine blades.Aiming at the characteristics of different intensities cavitation signals are difference,a extracting cavitation characteristics method based on stack sparse self-coding(SSAE)is proposed.An improved random forest algorithm(OPRF)based on Out-of-bag probability and posterior probability is proposed to deal with the problem that all generated decision trees are treated equally in the original random forest(RF)classifier.Then,cavitation signals of four different working conditions of water turbine are used for experimental verification.As a result,OPRF classifier is superior to RF classifier,and the proposed SSAE-OPRF algorithm can more accurately complete the classification and identification of hydraulic turbines cavitation signals of different intensification.(3)Aiming at the problem that the results of the feature classification algorithm model is unsatisfactory when the amplitude domain of the cavitation signal is directly used for deep learning,the WT-SSAE-OPRF and PSD-SSAE-OPRF algorithm models are proposed.Then,using the cavitation signals of four different working conditions to verify the two new algorithm models.The results show that overall classification accuracy and simulation efficiency have excellent performance,which can meet the accuracy and real-time requirements of turbine cavitation intensity recognition.A multi-sensor fusion model based on D-S evidence theory is proposed to solve the problem that the cavitation characteristics of hydraulic turbine collected by a single sensor may be incomplete.Through the fusion of the three kinds of sensors' characteristic decision-making levels,it becomes a reality to extract the cavitation characteristics of hydraulic turbine from many-side.Finally,the experimental data verify that the integrated method can effectively improve the overall classification accuracy of hydraulic turbine cavitation intensity.The SSAE-OPRF algorithm is optimized by the above two methods.(4)Finally,in order to complete the early warning of cavitation of hydroturbine,the noise signal of hydroturbine is calculated by kurtosis and root-mean-square calculation.Then based on the SDF of symbolic dynamics filtering,an improved symbolic dynamics filtering(SMDF)algorithm is proposed,and the monitoring index of primary cavitation M is given.Then,experimental simulation verifies that the newly proposed monitoring index is better than kurtosis monitoring index and root mean square(RMS)monitoring index,and can better give early warning of incipient cavitation of hydraulic turbine.