Semi-automatic fault database establishment method for power transmission and transformation line equipment

Abstract

The invention discloses a semi-automatic fault database establishment method for power transmission and transformation line equipment, and the method comprises the following steps: data collection: employing a data collection device to collect the related data of the power transmission and transformation line equipment, carrying out the screening of the data, and building an unclassified database.According to the semi-automatic fault database establishment method for the power transmission and transformation line equipment, firstly, the problems of automatic classification, detection and storage of the power transmission and transformation line equipment and a fault database are solved, semi-automatic establishment work of the power transmission and transformation line equipment and the fault database is completed, and secondly, model training-intelligent classification-intelligent test-intelligent optimization are integrated to form a complete closed loop, an autonomously optimized database system is constructed, power transmission and transformation line key equipment and a fault data set suitable for deep learning can be rapidly and efficiently established, the problem that thedata set is difficult to construct is solved, and a better use prospect is brought.

Description

technical field [0001] The invention relates to the fields of power systems and computer vision, in particular to a method for semi-automatically establishing a fault database of power transmission and transformation line equipment. Background technique [0002] As an important infrastructure in the power industry, power transmission and transformation lines are an important part of the power grid. The key components of power transmission lines mainly include insulators, fittings, and pole towers. If there is a problem with the components, it will endanger the stability of the entire power grid. Since 2006, to some extent, it can be said to have led a technological revolution in the era of big data. In 2012, it has achieved very good results in image recognition and target detection. It is much better than traditional methods in terms of recognition and detection of specific types of targets, and the recognition effect is very good, but in specific engineering applications: ...

AI Technical Summary

Problems solved by technology

[0003] The fundamental reason is that there is a lack of key equipment and fault data sets for power transmission and transformation lines. At present, there are no publicly available data sets such as PASCAL VOC, ImageNet, MS-COCO, Open Images Dataset, and Sun. Faulty data sets, this problem has caused great difficulties to the research in this field, and in the process of making data sets, there are still the following two problems:

[0004] 1. Manual classification of data is time-consuming and labor-intensive. After data collection, the data format is usually photos or videos. During the collection process, the high quality of the shooting content cannot be guaranteed, and preliminary quality screening is required manually. Secondly, the data collection content is multiple Components, while manually searching for faults of single-target components, it is necessary to manually classify all data first, and then perform fault finding work on specific target components. The whole process consumes a lot of manpower, material and financial resources, which seriously delays the work progress;

[0005] 2. The workload of labeling is heavy. For a single component or fault data set, there is no complete data analysis method, resulting in a large amount of data in the data set that has little impact on the model, which leads to a large number of pictures that need to be marked and increases the labor of labeling work. The workload, prolonging the data set preparation time, and wasting a lot of manpower;

[0006] 3. For model optimization work, there is still the problem of difficulty in using model test results. Model optimization work is the foundation of deep learning target detection. Start with model test results, analyze undetected images, and reconstruct two In order to complete the model optimization work, the model optimization work can only be completed by further advancing in this direction. In the process of data set reconstruction, it is necessary to analyze the undetected image features, find data from all data sets, and reconstruct the data set in proportion after preprocessing, so that Optimizing the model, this process requires in-depth analysis of image features, and repeated data search and labeling work, which increases the difficulty of model optimization

Images