220kV及以上电力变压器可靠性风险评估研究.pdf

摘 要 摘 要 目前国内外在变压器状态评估时数据收集比较单一，且习惯上某些特征量主 要采用 “是非制”，就是只有 “合格”和 “不合格”两种状态，没有对指标进行 量化，不能真实有效反映变压器的状态，这样对变压器进行状态评估不够细致， 不利于掌握变压器的真实运行状态。 本文依托云南省电力科学研究院，对电力变压器状态评估开展了进一步的研 究工作，将可靠性评估方法应用于电力变压器的可靠性评估，利用故障模式及影 响分析法建立基于故障树的变压器故障模型，计算得到变压器的可靠度值。在此 基础上采用 BP 神经网络对可靠度值进行拟合，并利用遗传算法（GA ）进一步 对BP 神经网络的初始权值和阀值进行优化，提高可靠度拟合的速率和准确度。 将该方法与风险评估方法从理论上和实际运算结果进行了对比研究，找到各自的 优缺点，并对评估初始数据进行了分析，计算各个数据特征量对评估结果的贡献 度。通过以上研究内容，在进行变压器的状态评估时，依据特征量贡献度大小收 集数据，综合两种方法的评估结果得到变压器的运行工况，提高了评估速率和可 信度，为状态检修提供依据。 关键词：电力变压器；可靠性评估；风险评估；神经网络；贡献度 I Abstract Abstract The current transformer data collection for the transformer assessment at home and abroad is relatively simple, and some of the features are assessed by the use of the to be or not to be system, which means only two states, qualified and unqualified. There is no quantification of indicators, which could not reflect the real insulation condition effectively, so the transformer condition assessment is not detailed enough, which is not conducive to grasp the real operational status of the transformer. This paper carried out further research work for the power transformer condition assessment, relies on Yunnan Electric Power Research Institute, the reliability evaluation method is applied to assess the reliability of power transformers, and a transformer fault tree model is established by using failure modes and effects analysis, to calculate the reliability of the transformer. BP neural network is used to fit the reliability on the basis, using genetic algorithms (GA) to optimize initial weights and thresholds