摘要
电力系统的运行状态具有动态变化特性,导致基于静态数据训练的模型难以适应实际场景的时变需求,无法实现高效、准确的实时优化。为此,提出一种基于大规模生成模型的电力业务视觉场景优化方法。首先,采集电力业务相关的视觉数据并进行预处理;随后,利用深度学习技术从预处理后的配电场景数据中提取特征,以表征初始配电场景的数据属性。在此基础上,构建基于卷积神经网络的电力视觉大模型,并利用清洗后的数据集及其标注信息对模型进行训练与深度优化。优化后的模型用于监测配电设备的运行状态,并通过确定可视化像素范围完成场景优化。所采用的大规模生成模型能够依据实时数据与反馈信息进行自适应调整,随着电力系统运行状态的动态演化持续学习新数据并更新模型参数,从而有效适应实际场景的变化。实验结果表明,所提方法将处理帧率提升至115 FPS,较优化前提高45%;设备状态监测准确率达到90%,在多种环境条件下均表现出良好的泛化能力,适用于大规模配电设备状态监测任务,并实现了有效的场景优化效果。
The operating state of the power system is inherently dynamic,which makes it difficult for models trained on static data to adapt to changes in actual scenarios and optimize in real-time and accurately.Therefore,a visual scene optimization method for power business based on large-scale generative models is proposed.Firstly,visual data related to power business is collected and preprocessed;then,extract features from the preprocessed distribution scene data by deep learning technology,and obtain the data properties of the initial distribution scene.On this basis,a large-scale electric vision model based on convolutional neural network is developed and rigorously optimized using a cleaned dataset and its associated annotation information.The optimized model is deployed to monitor the operational status of distribution equipment within power distribution scenarios and performs scene optimization by determining an appropriate visual pixel range.The large scale generative model supports adaptive adjustment and continuous refinement through real-time data and feedback.With the dynamic changes in the power system,generative models can continuously learn new data and update model parameters to adapt to changes in actual scenarios.Experimental results demonstrate that the proposed method achieves a processing frame rate of 115 FPS—an improvement of 45%compared to the pre-optimization baseline—and attains an equipment status monitoring accuracy of 90%.The approach exhibits robust performance across diverse environmental conditions,proving suitable for large-scale distribution equipment monitoring tasks and delivering effective scene optimization outcomes.
作者
陶俊
郭庆
齐奕斐
薛濛
黄旭东
TAO Jun;GUO Qing;QI Yifei;XUE Meng;HUANG Xudong(Anhui Jiyuan Software Co.,Ltd.,Hefei 230094,China)
出处
《国外电子测量技术》
2025年第12期317-322,共6页
Foreign Electronic Measurement Technology
基金
面向电网差异化场景的人工智能共性技术及典型应用深化研究及应用(SGITG-JTZDYFGGZX-2025-01-03)。
关键词
电力业务
视觉场景
深度学习
卷积神经网络
power business
visual scene
deep learning
convolutional neural network
