基于深度学习的地址分割算法在电力业务辅助录入系统上的应用研究

Research on the Application of Address Seg-mentation Algorithm Based on Deep Learning in Power Service Auxiliary Input System

现代电力行业中的信息系统通常需要对地址信息进行特定的格式处理以提高数据的统计与分析能力。本文提出一种基于深度学习的地址分割算法,对地址信息按行政区级进行分割。采用ELU激活层对LSTM网络进行优化以提高网络的整体性能,另外采用GAN网络对数据集进行了增强,进一步降低了训练过程中过拟合情况的发生。实验结果表明,算法对于地址数据的分割平均准确率达到99%,平均运算时间为0.1秒,满足辅助录入系统需求。算法利用分割地址信息关联对应的供电局(所),有效提升了电力业务的办理效率,具有较好的应用前景。

Information systems in the modern power industry often require specific formatting of address information to improve data analysis and statistics capabilities. This paper proposes a deep learning-based address segmentation algorithm that separates address information into administrative divisions. The LSTM network is optimized using the ELU activation layer to enhance overall network performance. Additionally, GAN networks are used to augment the dataset, further reducing over-fitting during the training process. Experimental results show that the algorithm achieves an average accuracy of 99% in address segmentation with an average computation time of 0.1 seconds, meeting the requirements of auxiliary input systems. The algorithm utilizes the segmentation of address information to associate it with the corresponding power supply bureaus, thereby further enhancing the efficiency of electricity service processing. It has great potential for application in the power industry.