油色谱气体含量预测模型研究

油色谱气体含量预测模型研究致谢此次论文撰写的整个过程都得到了孙劲光教授的指导和帮助，因此在此必须对他表示感谢。这一年多来，从选题开始孙劲光教授就用他渊博的知识、敏捷的思维、严谨的治学态度和高度的敬业精神深深鼓舞着我，到后来确定了题目开始着手研究，这中间孙劲光教授总会百忙之中抽出时间为我指导研究方向，询问学习进度，并推荐值得借鉴的相关书籍和文章，最后能够完成此文章的写作，孙老师是我一生不可多得的学习榜样，我将永远铭记在心。摘要电流互感器和变压器等电力设备是电力系统中最重要的设备，其稳定运行直接影响整个电力系统的安全性和可靠性，而这类电力设备经常容易出现潜伏性故障，要想解决这些故障，油中溶解气体分析技术，即DGA技术能够发挥极大作用。设备故障多长时间后将发生或者在将来某个时间设备是否安全，需要利用预测模型进行预测。为了提高预测的准确度，优化现有预测模型，本文通过研究分析，提出一种基于油色谱气体建立气体含量预测模型，建立受限玻尔兹曼机(RBM)二层神经网络预训练得到初始化权重参数，在此基础上构建BP神经网络模型对气体含量进行预测，经实验验证，本文提出的模型较传统B神经网络预测模型在精度和收敛速度上有较大优势。该论文有图9幅，表4个，参考文献29篇。关键词：油色谱气体含量预测；受限玻尔兹曼机：BP神经网络ABSTRACTCurrent transformer and transformer are the most important equipment in thepower system.Its stable operation directly affects the safety and reliability of thewhole power system.Dissolved gasanalysis (DGA)is the prevailing effective methodto find latent faults of current transformers and transformers,especially after longdevelopment process.To resolve how long the failure of the equipment will occur orwhether the equipment is safe at a certain time in the future,a prediction modelneeds to be used.In order to improve the accuracy of prediction and optimize theexisting prediction models,a gas content prediction model based on oil chromatogramgas is proposed through research and analysis,The two-layer neural network ofrestricted Boltzmann machine (RBM)is established to obtain the initialization weightparameters.On this basis,the BP neural network model is constructed to predict thegas content.The experimental results show that the model proposed in this paper hasgreater advantages in accuracy and convergence speed than the traditional BP neuralnetwork prediction model.Keywords Fault prediction of DGA;RestrictedBoltzmann machine;BP neuralnetwork