Research On Pattern Recognition Algorithm And Its Application For Multi-source Features

When we study the related equipments or waveforms in power system,the characteristics of a single information source are not enough to fully describe it.In recent years,with the advent of the era of big data.Internet of things and cloud computing,the amount of power grid data is growing and the types are increasing,which provides multi-source information support for the identification of various modes such as equipment status diagnosis.But it is difficult to guarantee the classification effect only by traditional single-kernel classification due to the increase of feature sources,large distribution differences and complex relationship between information.Therefore,it is necessary to study the effective analysis and processing of high-dimensional features and to maintain the recognition stability of the classifier.In order to improve the performance of pattern recognition algorithm based on multi-source features,this paper introduces fusion algorithm in feature extraction,pattern classification and decision-making,and then constructs the framework of multi-source feature recognition algorithm to meet the application needs of different scenarios.In order to achieve the fusion of extraction links,the paper considers the influence of the joint action of paired features on the correlation between features and categories and the standardization of measurement scale,and proposes a feature selection algorithm based on the max-relevance and min-redundancy criterion for joint interaction-redundancy to improve the max-relevance and min-redundancy criterion.When completing the fusion of pattern classification,support vector machine is selected as classifier to overcome the problem of small sample size from some sources.And because the learning performance of different kernel functions or different parameters of the same kernel function is quite different,the multi-source features are mapped to different high-dimensional spaces by combining multiple kernel functions.Due to the sample space of multi-source features at this time,the paper proposes a multi-kernel SVM algorithm integrating radius information.This algorithm introduces the radius information into the multi-kernel SVM model,and jointly determines the kernel function weight with interval.By analyzing the fusion of decision-making links,it is found that it can’t improve the the final identification accuracy when the amount of data from a certain source is small.Therefore,the structure of the integrated classifier is used to build the neural network corresponding to each source feature.In order to make the neural network corresponding to the small sample feature set learn and express its typical features better,the initial training and secondary training are carried out with multi-source feature samples and small sample features to modify the weight coefficient of neural network,so as to promote the action of sensitive neurons and restrain the action of insensitive neurons.Power quality disturbance identification is of great significance to the analysis and the selection of control measures of power quality disturbances.In the past,most researches have focused on the identification of single power quality disturbance,but there are many kinds of power quality disturbances in the field.At this time,the feature space is more complex and the feature boundary is more fuzzy.It is impossible to identify all kinds of disturbances only by single extraction method.By analyzing the characteristics of the feature curve obtained by various feature extraction methods,the paper further explains the necessity of introducing multi-source features to the identification of mixed power quality disturbances,and then the three methods proposed in this paper are applied to the identification of mixed power quality disturbances.Through the analysis of the simulation data,the effectiveness of the multi-source features in the identification of mixed power quality disturbances and the effectiveness of the three methods proposed in this paper are verified.Power transformer is the key link of power system.It is directly related to the stability of power supply whether the potential faults of transformer can be found as soon as possible to ensure the normal operation of transformer.Because structure of power transformer is complex,the operation environment is uncertain,and the relationship between fault symptoms and fault occurrence mechanism is diverse and fuzzy,so it has great limitations to diagnose transformer fault only with a single information.And on the basis of analyzing the characteristics of various transformer characteristics,it further explains the necessity of introducing oil gas and electrical characteristics simultaneously and then the three methods proposed in this paper are applied to transformer fault diagnosis.Through the analysis of field data,the effectiveness of gas and electrical characteristics in transformer fault diagnosis and the effectiveness of the three methods proposed in this paper is verified.The simulation data and case analysis verify the effectiveness of the three methods proposed in this paper.Through the evaluation of the feature selection results based on the maximum correlation minimum redundancy criterion of joint interaction redundancy,it shows that the method can select the key feature subsets from a large number of features,and ensure the identification accuracy in the case of few features;through the analysis of the identification results of multi-kernel SVM integrating radius information,it shows that the method can effectively combine information from different sources,with high identification accuracy and good robustness;Through the analysis of the identification results of the integrated classifier algorithm,it shows that the information from other sources can provide supplementary information for the small sample feature set,and improve the final identification accuracy.