基于深度学习和图像识别的变电站虚端子连接校核技术

Verification Technology for Virtual Terminal Connection Based on Deep Learning and Attention Image Recognition

针对变电站配置描述(substation configuration description, SCD)校验工作量庞大以及人工校验效率低、准确率低的问题,提出一种基于深度学习和图像识别的虚端子连接智能校核技术。首先利用Tensorflow框架的Faster R-CNN模型,对SCD虚连接图中的保护、测控智能电子设备(intelligent electronic device, IED)进行智能辨识,将端列端识别的Attention-OCR方法和“Advanced EAST+Tesseract OCR”的两阶段识别方法进行对比,选择准确率较高的Attention-OCR端到端识别方法实现SCD虚连接图中的文字、数字识别;然后,利用图像处理技术实现SCD虚连接图连接关系识别,数据整理后得到SCD侧虚连接表。最后,开展SCD侧与Excel侧虚连接表正向与反向2个方向的比对,形成核对报告,筛选出不对应的信息,有效地提高了SCD校验的效率和准确率。

In response to the heavy workload of substation configuration description(SCD)verification and current low efficiency and accuracy of manual verification,this paper proposes a virtual terminal connection intelligent verification technology based on deep learning and image recognition.Firstly,it uses Faster R-CNN model of Tensorflow framework to intelligently identify the protection and measurement and control intelligent electronic device(IED)in the SCD virtual connection graph.By comparing the two-stage recognition methods of Advanced EAST and Tesseract OCR with the end-to-end recognition methods of Attention-OCR,it selects the end-to-end Attention OCR method with higher accuracy ultimately to achieve text and number recognition in the SCD virtual connection graph.Then,it uses image processing technology to identify the connection relationship of the SCD virtual connection graph,and organizes the data to obtain the SCD side virtual connection table.Through the forward and reverse comparison of the virtual connection tables on the SCD side and Excel side,a verification report is finally formed to screen out non corresponding information,effectively solving the current problems of low efficiency and accuracy in manual comparison.