Research On The Identification Of Drainage Pipe Blockage Based On Imbalanced Data Classification Method

Machine learning is widely used in the research of the recognition of the state of drainage and drainage pipeline insertion,but the data of the research on the recognition of the state of drainage pipeline interruption has obvious distribution imbalance characteristics.Traditional machine learning methods have great difficulties in the classification of imbalanced data,because There is a cumulative time course for pipeline blockage,and there is a certain lag and incompatibility in the detection of replacement.If there is a problem in the recognition of the status of the drainage pipeline blockage,it is easy to cause the pipeline to be judged as a normal pipeline and thus form a potential safety hazard over time.Therefore,it is necessary to accurately identify the operating state of the drainage pipe and timely diagnose the degree of insertion of the drainage pipe.Taking the replacement of drainage pipes as the application background here,for the problem of the decline in the accuracy of the operating state recognition caused by the imbalance in the data acquisition of the normal and stop failure states of the urban drainage pipes,based on the resampling algorithm and cost sensitivity,respectively A research method of pipeline interchange status recognition,replacing it with pipeline interchange status recognition,using laboratory simulation data to conduct simulation experiments,said.The specific work that the algorithm proposed in this article is feasible is as follows(1)Presents the research status of unbalanced data classification and pipeline insertion recognition,analyzes related theories of unbalanced data classification,and introduces typical methods and performance evaluation indicators of unbalanced data classification.Using the data obtained by active acoustic detection,the acoustic signals are analyzed by signal processing methods,and common state recognition methods are analyzed(2)Research on pipeline insertion state recognition based on unbalanced data resampling.Aiming at the problem of the reduction of the accuracy of the operating state recognition caused by the imbalance in the data acquisition of the normal and inserted fault states of the urban drainage pipeline,the wavelet packet decomposition algorithm is first used to decompose the acoustic response signal in 3 layers,and different resolution signals will be obtained As a characteristic component signal Obtain the energy entropy of the characteristic component signal,and then use the under-sampling method of K-means for the unbalanced data set,and perform the improved oversampling method and the mixed sampling method of the two sampling methods to obtain the balanced data set,and finally the four types of data Set for classification identification and index evaluation,this method can effectively improve the classification accuracy on the data scale(3)Research on pipeline reset state recognition based on FOA optimized CS-SVM.According to the unbalanced data sets collected under various operating conditions inside the drainage pipeline,first,the original unbalanced data set is subjected to wavelet packet decomposition and reconstruction algorithms to obtain decomposition coefficients of different amplitudes.Second,the energy entropy of each decomposition coefficient is extracted,and the approximate entropy Indicators are used to construct feature vector sets;Drosophila optimization algorithm(FOA)is used to optimize the selection of correction factors Cm and kernel function parameters g for different types of samples,that is,to optimize the cost-sensitive support model machine(CS-SVM)model and input feature sets into optimization In the later CS-SVM model,for the recognition of the normal and switching state of the drainage pipeline,this method improves the recognition accuracy in the decision method,reduces the human intervention errors generated when the data size is solved,overfitting,and the data sample is redundant Repeat and other issues.