Research On Vectorization Technology Of Electrical Drawings

Electrical drawings represent the location of electrical equipment and their wiring in the field of power grids.It can guide wiring,line maintenance and troubleshooting issues.With the increasing complexity of the power grid structure,it has become the norm to manage the power grid dispatching with the help of computers and automation technologies.To do this,a conversion job is required to change the drawings in the image format to those can be applied by the supervisory control system,which is called vectorization.At present,the vectorization of drawing is carried out manually by staff with the help of the drawing system,which caused huge manpower and time costs.Therefore,it is necessary to study the automatic vectorization method of electrical drawings.The process of drawing vectorization is to extract the position and wiring relationship of electrical equipment in the drawing,and generate a structured information document similar to XML file(called G file in the field of power grid),which can be applied by the power grid control system.Through in-depth analysis of the process of manual vectorized electrical drawings,this thesis constructs a technical solution for electrical drawing vectorization,which realizes the identification and positioning of electrical equipment in electrical drawings,the recognition of label texts,and the connection analysis between electrical elements.The main contents of this thesis are as follows:(1)Aiming at the problem of poor detection ability of symbols with large number,small size and dense distribution of electrical equipment in drawings,an electrical symbol detection model based on YOLO and channel attention mechanism SE-Net is proposed.It achieved an average detection accuracy of 91.4% for all classes of symbols.While ensuring accuracy,the model also meets the requirements of real-time detection,and the average detection time of each drawing is 0.34 seconds.(2)Aiming at the algorithm bottleneck problem that the existing algorithms are difficult to extract the annotation texts with different font sizes in the drawings,a text area detection algorithm based on multi-scale feature extraction is proposed,which realizes the precise positioning of the all annotation texts in the drawings.Besides,A CRNN-based annotated text recognition model is constructed,and the text area located in the previous stage is fed into the model to obtain the its content.(3)Aiming at the problem that there are too many electrical elements with complex relationships in the drawings,and it is difficult to accurately extract the relationships and reproduce them into vector diagram,a vectorization strategy based on template matching is proposed,which makes full use of the block characteristics of the electrical drawings and the results of the previous work.Taking the tuple group as a unit,the analysis problem of the electrical element association in the tuple group is converted into the identification problem of the tuple group template,and then the topological relationship analysis of the electrical elements of the whole drawing is completed.Converting the problem of analyzing relationship into the problem of template identification,and then the topological relationship analysis of the electric elements in the whole graph is completed.(4)The electrical drawing vectorization system was designed and implemented,and it has been deployed and tested in a production environment of North China Electric Power Company,and it performed well.