1.华南理工大学电力学院,广东省广州市 510640;2.广东省电网智能量测与先进计量企业重点实验室,广东省广州市 510640
矢量化的电网厂站接线图需要调度运维人员参考设计图纸进行人工绘制及导入,工作量巨大且极易出错。针对电网厂站接线图中图元、文字以及接线关系3个核心的识别问题,提出一套基于人工智能的完整电网厂站接线图识别方法,显著提升识别的效率和准确性。通过结合分级预处理的重叠滑窗机制与YOLOv4算法,解决电网厂站接线图的“大图像小图元检测”问题;通过迁移学习与卷积循环神经网络结合,提升包含中文电气文本的识别率;通过嵌入电气领域知识与规则,提升接线识别的准确率。最后,根据获取的电气元件信息、文本信息、连接线信息,完成其他下游任务。采用实际调度系统导出的电网接线图数据集,设计了图元、文字、接线识别三方面的对比实验,验证了所提方法的有效性。
国家自然科学基金委员会-国家电网公司智能电网联合基金资助项目(U2066212);已申请国家发明专利(申请号:202111587313.8,202111587336.9,202111589662.3)。
王梓耀(1996—),男,博士研究生,主要研究方向:人工智能与优化在配电网规划与运行中的应用、配电网显式可靠性评估。E-mail:ziyaowang100@sina.com
罗庆全(1999—),男,硕士研究生,主要研究方向:电力指纹及电力人工智能相关应用。E-mail:571998939@qq.com
萧文聪(2000—),男,硕士研究生,主要研究方向:人工智能与优化在配电网规划与运行中的应用。E-mail:201830211110@mail.scut.edu.cn
余涛(1974—),男,通信作者,博士,教授,主要研究方向:复杂电力系统的非线性控制理论和仿真、优化及机器学习。E-mail:taoyu1@scut.edu.cn
1.College of Electric Power, South China University of Technology, Guangzhou 510640, China;2.Guangdong Provincial Key Laboratory of Intelligent Measurement and Advanced Metering of Power Grid, Guangzhou 510640, China
Vectorized wiring diagrams for power plants and substations need to be manually drawn and imported by dispatching operation and maintenance personnel with reference to design drawings, which is labor-intensive and error-prone. Aiming at the three core recognition problems, i.e., graphic element, text and wiring relationship of the wiring diagram for power plants and substations, a complete method of wiring diagram recognition for power plants and substations based on artificial intelligence is proposed to significantly improve the efficiency and accuracy of recognition. Through the combination of overlapping sliding window mechanism of hierarchical preprocessing and YOLOv4 algorithm, the problem of “large image and small graphic element detection” of the wiring diagram recognition for power plants and substations is solved. Through the combination of transfer learning and convolutional recurrent neural network, the recognition rate of Chinese electrical texts is improved. Embedded with the electrical domain knowledge and rules, the accuracy of wiring recognition is improved. Finally, according to the obtained electrical component information, text information, connection line information, other downstream tasks are completed. Using the data set of grid wiring diagram derived from the actual dispatching system, the comparison experiments of graphic element, text and wiring recognition are designed to verify the effectiveness of the proposed method.
[1] | 王梓耀,罗庆全,萧文聪,等.电网厂站接线图人工智能识别关键方法[J].电力系统自动化,2023,47(2):115-124. DOI:10.7500/AEPS20220223008. WANG Ziyao, LUO Qingquan, XIAO Wencong, et al. Key Method of Wiring Diagram Recognition with Artificial Intelligence for Power Plant and Substation[J]. Automation of Electric Power Systems, 2023, 47(2):115-124. DOI:10.7500/AEPS20220223008. |