基于改进深度强化学习算法的电网缺陷文本挖掘模型研究被引量：6

Research on Grid Defect Text Mining Model Based on Improved Deep Reinforcement Learning Algorithm

在线阅读下载PDF

导出

摘要为提高电网缺陷文本的感知深度与泛在性,改善典型仿生智能算法处理低价值密度、多维波动非线性、动态迭代更新属性下的电网海量文本缺陷特征感知问题时存在的异构泛在性较差、动态感知策略同步效能滞后、缺陷文本划分密集且生成困难、样本效率与迭代次数失衡、异构数据输入下的模型性能下降严重等固有弊端,提出了一种基于改进深度强化学习算法的电网缺陷文本挖掘模型。借助国家电网公司全局数据中心,构建非结构化的电网缺陷文本数据池,引入深度学习多维感知电网缺陷文本特征,实现缺陷文本的语义框架自构建,引入强化学习实现缺陷文本自主识别决策,并把当前收益(语义槽)和未来收益反馈给环境(隐性知识)模拟策略网络,在有限马尔科夫决策过程中引入多重Q网络机制实现知识地图的自生成,进而实现本体字典自动完善。以南方电网贵州电网有限责任公司数据管控中心为效能评价载体,基于谷歌的Tensorflow 1.2.1和Open AI的Gym 0.9.2环境开发了可视化验证环境并对模型进行了实证分析,仿真验证结果表明本文所提模型可以在较短的时间内处理低价值密度、多维波动非线性、动态迭代更新属性下的电网海量文本缺陷特征感知问题,在深度泛在性、感知自主性、决策准确性、异构数据输入下的模型容错性等方面具有明显优势。 In order to improve the perception depth and ubiquity of power grid defect text,improve the heterogeneous ubiquity of typical bionic intelligent algorithm in dealing with power grid massive text defect feature perception problems under the attributes of low value density,multi-dimensional fluctuation nonlinearity and dynamic iterative updating,the synchronization efficiency of dynamic perception strategy lags behind,defect text is divided intensively and hard to generate,sample efficiency and iteration times This paper proposes a text mining model of power grid defects based on improved deep reinforcement learning algorithm.With the help of the global data center of State Grid Corporation of China,an unstructured text data pool for power grid defects is constructed.The features of power grid defects are perceived in-depth learning,the semantic framework of defect text is self constructed,the decision-making of defect text is realized through reinforcement learning,and the current income(semantic slot)and future income are fed back to the environment(tacit knowledge)simulation strategy Network,in the process of finite Markov decision-making,multi-Q network mechanism is introduced to realize the self generation of knowledge map and the automatic improvement of ontology dictionary.Taking the data management and control center of China Southern Power Grid Guizhou Power Grid Co.,Ltd.as the performance evaluation carrier,based on tensorflow 1.2.1 of Google and gym of openai 0.9.2 environment developed a visual verification environment and made an empirical analysis of the model.The simulation results show that the model proposed in this paper can deal with the feature perception of massive text defects under the attributes of low value density,multi-dimensional fluctuation nonlinearity,dynamic iterative updating in a short time,under the conditions of depth universality,perception autonomy,decision-making accuracy and heterogeneous data input It has obvious advantages in model fault tolerance.

作者吴漾王鹏宇缪新萍柳林溪田钺 Wu Yang;Wang Pengyu;Miu Xinping;Liu Linxi;Tian Yue(China Southern Power Grid Guizhou Power Grid Co.,Ltd,Guizhou Guiyang 550003,China;Information Center of China Southern Power Grid Guizhou Power Grid Co.,Ltd,Guizhou Guiyang 550003,China;Measurement center of Guizhou Power Grid Co.,Ltd.of China Southern Power Grid,Ltd,Guizhou Guiyang 550003,China)

机构地区南方电网贵州电网有限责任公司南方电网贵州电网有限责任公司信息中心南方电网贵州电网有限责任公司计量中心

出处《科技通报》 2021年第2期47-55,共9页 Bulletin of Science and Technology

基金贵州电网有限责任公司科技项目(066700KK52170012) 中国南方电网有限责任公司科技项目(GZHKJXM20160055)

关键词电网缺陷文本深度强化学习文本深度挖掘隐形知识模型仿真验证 power grid defect text deep reinforcement learning text deep mining hidden knowledge model simulation verification

分类号 TM73 [电气工程—电力系统及自动化] TP18 [自动化与计算机技术—控制理论与控制工程]

作者简介吴漾(1984-),男,汉族,贵州毕节人,中级工程师,硕士,主要研究方向:电网数据智能监测及信息化支持、文本语义解析理论与人工智能,E-mail:gzdwtg2@163.com。