基于循环BP模型的变压器状态数据清洗方法

Online Detection Method of Transformer Capacity Based on Cyclic BP Model

受运行环境、运行状态或采集装置的干扰,变压器在线检测状态数据中存在大量的噪声,会造成短路阻抗的在线计算出现波动,影响变压器容量的检测结果。在综合分析当前数据清洗方法的基础上,提出一种以BP神经网络为核心的循环清洗方法,对历史数据无监督清洗并同步训练出清洗模型,利用所训练出的清洗模型实现对检测数据的在线清洗。使用沈阳地区某企业配电变压器运行数据进行了仿真试验,搭建循环BP模型的清洗框架并训练清洗模型,其数据在无监督清洗后进一步做短路阻抗的拟合计算,以验证所提方法的有效性。结果表明,基于循环BP模型的变压器容量在线检测方法所计算的短路阻抗结果趋于稳定,计算误差维持在5%以下,可有效提高短路阻抗在线拟合的精度。总之,循环BP模型是一种无监督、智能化的数据清洗方法,可以有效改善噪声数据对检测结果的影响,提高变压器容量检测的精度,更具智能性。

Due to the interference of operating environment,operating state or acquisition device,there is a lot of noise in the on-line detection state data of transformer,which makes the on-line calculation of short-circuit impedance fluctuate and affects the detection results of transformer capacity.Based on the comprehensive analy⁃sis of the current data cleaning methods,a circulating cleaning method with BP neural network as the core was proposed.The historical data was cleaned unsupervised and the cleaning model was trained synchronously.The online cleaning of the test data was realized through the trained cleaning model.The simulation experiment is carried out by using the operation data of the distribution transformer of an enterprise in Shenyang.The clean⁃ing framework of the cyclic BP model is built and the cleaning model is trained.The data is further fitted by the short-circuit impedance after unsupervised cleaning to verify the effectiveness of the proposed method.The re⁃sults show that the short-circuit impedance calculated by the on-line detection method of transformer capacity based on cyclic BP model tends to be stable,and the calculation error is maintained below 5%,which can effec⁃tively improve the accuracy of on-line fitting of short-circuit impedance.In short,the cyclic BP model is an un⁃supervised and intelligent data cleaning method,which can effectively improve the influence of noise data on the detection results,improve the accuracy of transformer capacity detection,and be more intelligent.