基于深度信念网络的变压器油中溶解气体浓度预测方法  

Concentration Prediction of Dissolved Gases in Transformer Oil Based on Deep Belief Networks

作　　者：代杰杰 宋辉[1] 杨祎 陈玉峰 盛戈皞[1] 江秀臣[1] 

机构地区：[1]上海交通大学电气工程系,上海市闵行区200240 [2]国网山东省电力公司电力科学研究院,山东省济南市250002

出　　处：《电网技术》2017年第8期2737-2742,共6页Power System Technology

基　　金：国家自然科学基金项目(51477100);国家863高技术基金项目(2015AA050204);国家电网公司科技项目(520626150032)~~

摘　　要：对油中溶解气体浓度发展趋势进行预测,可为变压器状态评估提供重要依据。提出一种基于深度信念网络的变压器油中溶解气体浓度预测模型。该模型以7种特征气体浓度、环境温度、变压器油温为可视输入,通过对基于受限玻尔兹曼机的多隐层机器学习模型训练,可自动提取气体浓度自身发展规律,逐层激活各气体组分之间及温度对气体浓度影响的强相关性,抑制、弱化无关和冗余信息。该模型具有较高预测精度,克服了传统单一变量预测方法稳定性差的问题,同时避免了人工干预过程。通过算例分析,验证了该方法的有效性。Prediction of development trend of gas concentration dissolved in transformer oil can provide important basis for transformer condition assessment. A new prediction model based on deep belief networks is proposed. Seven types of characteristic gas concentration combined with environment temperature and transformer oil temperature are fed to input layer. The model can automatically extract regulation of gas concentration development trend through training a multi-bidden-layer machine learning model based on restricted Boltzmann machine. Correlation between different types of gases and influence of temperatures is activated layer by layer. Irrelevant and redundant information is inhibited by the model. The proposed method has higher prediction accuracy. It overcomes drawbacks of low stability in traditional methods and shortcoming of considering only one characteristic gas. In addition, it avoids manual intervention in calculation process. Finally, case analysis verifies effectiveness and superiority of the proposed model.

关 键 词：变压器 油中溶解气体 深度信念网络 相关性 预测 

分 类 号：TM855[电气工程—高电压与绝缘技术]