摘要
随着人工智能技术的发展和进步,我们日常生活的各个方面现在都发生了巨大的变革,尤其是卷积神经网络在输电线路中电力导线故障检测中的应用,极大程度上消除了输电安全隐患,保障了人民生活用电,但是常规卷积神经网络应用于视觉任务需要大量的训练数据,而输电线路中某些故障极为稀缺,收集与标记这些训练数据要消耗巨大的人力、物力。基于此本文提出利用自监督表示学习算法应用于输电线路中电力导线的故障分类识别任务,以缓解数据采集与标注困难的问题。自监督表示学习算法可从未标记样本中进行学习,不需要负采样,有更高的训练效率,在实验中,将自监督表示学习算法与其他基线方法进行比较,其表现优异,在电线损坏分类与识别任务中,能够达到0.87的平均精度,表明了该算法的有效性与实用性。
With the development and progress of artificial intelligence technology, all aspects of our daily life have undergone tremendous changes, especially the application of convolutional neural networks in the fault detection of power conductors in power transmission lines, which has greatly eliminated power transmission. Potential safety hazards ensure people’s electricity consumption, but the application of conventional convolutional neural networks to visual tasks requires a lot of training data, and certain defects in the transmission line are extremely scarce. Collecting and marking these training data consumes huge manpower and material resources. Based on this, this paper proposes to apply self-supervised representation learning algorithm to the fault classification and recognition task of power conductors in transmission lines to alleviate the problem of data labeling difficulties. The self-supervised representation learning algorithm can learn from unlabeled samples and does not require negative sampling. It has higher training efficiency. In the experiment, the self-supervised representation learning algorithm is compared with other baseline methods and its performance is excellent. In the task of classification and identification of wire damage, the average accuracy can reach 0.87, which shows the effectiveness and practicability of the algorithm.
