摘要
为了提升电力工程项目数据的审核效率,对评审文本信息的结构化提取识别方法进行研究。通过使用玻尔兹曼机替代传统的神经元,建立深度置信网络(DBN)来传递文本。在训练过程中,使用贪婪训练法逐步增加网络隐藏层的数量,以提升网络的非线性拟合能力。针对DBN在迭代后期网络收敛速度减缓的问题,引入Q强化学习算法,并采用奖励累计机制将DBN错分类的样本进行拷贝和旋转处理,进而输入训练集以增加样本的权重。利用ε贪婪选择法对网络误差函数的偏导数求解方法加以优化,进一步提升网络的训练速度。仿真结果表明,所提算法可以准确地提取设计资料中的项目名称、电压等级、工程类别等信息,且相较于传统DBN,所提方法在准确率、召回率和F_1值上分别提升了4.8%、5.3%、5.0%。
In order to improve the efficiency of power engineering project data review,this paper studies the structured extraction and recognition method of review text information.In this method,Boltzmann machine is used to replace traditional neurons,and a deep belief network(DBN)is established for text transmission.In the training process,the greedy training method is used to gradually increase the number of hidden layers of the network to improve the nonlinear fitting ability of the network.In addition,in view of the problem that the network convergence speed of DBN slows down in the later stage of iteration,the Q reinforcement learning algorithm is introduced.Through the reward accumulation mechanism,the samples misclassified by DBN are copied and rotated,and then input into the training set to improve the weight of these samples.Theε-greedy selection method is use to optimize the partial derivative solution method of the network error function,which further improves the training speed of the network.The simulation results show that the proposed algorithm can accurately extract information such as project name,voltage level and project category from the design data.Compared with the traditional DBN,the accuracy rate,recall rate and F 1 value are increased by 4.8%,5.3%and 5.0%,respectively.
作者
蒋哲
齐增清
JIANG Zhe;QI Zengqing(Hunan Jingyan Electric Power Design Co.,Ltd.,Changsha 410007,China)
出处
《微型电脑应用》
2025年第3期28-31,共4页
Microcomputer Applications
基金
湖南省科技计划项目(S2022CXCPB0559)
湖南经研电力设计有限公司自筹项目(Heid2021002)。
关键词
深度学习
强化学习
数据识别
文本提取
Q强化学习算法
deep learning
reinforcement learning
data recognition
text extraction
Q reinforcement learning algorithm
