Research On Comprehensive Fault Diagnosis Technology Of Weather Radar System

The new generation weather radar system of the domestic network has all been working for more than ten years,some of them have been in use for nearly 20 years.Some key components are aging,and the probability of fault is on the rise.Currently,The radar maintenance and troubleshooting mainly rely on radar manufacturers and experts with rich maintenance experience,and it is difficult to guarantee the timeliness of radar fault detection.Actually,fault diagnosis has a variety of methods and algorithms,such as neural network,fuzzy logic,support vector machine,genetic algorithm,etc.,but many methods are difficult to apply in weather radar fault diagnosis system due to the small number of fault samples.In 2017,Chengdu University of Information Technology developed a real-time acquisition system for indicating radar working status and performance—The Weather Radar Standard Output Controller(WRSOC).It can monitor 87 radar status parameters,such as key technical parameters,performance indicators and adaptation parameters of weather radar subsystems,and the amount of data is sufficient to provide a data base.Based on the fault alarm information and the monitoring of key parameters of the WRSOC in Ji'an,Jiangxi Province,this paper discusses and studies the fault diagnosis technology of weather radar using Logic Regression,Neural Network,Support Vector Machine(SVM),Anomaly Detection and artificial Synthetic Minority Oversampling Technique(SMOTE).The main contents are as follows:(1)Currently,in allusion to the fault tree expert system used by meteorological radar,a shallow network classification model based on knowledge and self-learning is proposed to classify faults.The principles and characteristics of logistic regression,neural network and SVM classification algorithm are studied.Practical application in diagnosis.(2)Due to the difficulty in obtaining fault sample labels in the meteorological radar business process,the high cost of accurate label labeling,and the imbalance of data between different fault categories,the anomaly detection algorithm with unsupervised learning ability and the artificial oversampling technology of synthesis of minority samples are studied.(3)For the fusion of these three methods,in the context of meteorological radar fault diagnosis,a SVM semi-supervised integrated fault diagnosis method based on artificial data synthesis of SMOTE and anomaly detection is proposed.At the same time,the implementation and analysis of the method are completed.The efficiency of the proposed method is verified by the comparison experiments.The results show that the semi-supervised fault diagnosis method of SVM integrated with SMOTE and anomaly detection can effectively improve the accuracy of fault diagnosis under the condition of unbalanced fault sample data.Due to the limited accuracy of fault sample labels and the expensive cost of expert labeling,the effect of the method proposed in this paper is not very obvious.However,through analysis and comparison experiments,the superiority of the proposed method which is instructive and meaningful can still be verified,and it can improve the fault diagnosis technology to a new level.