基于多物理场计算和模糊神经网络算法的变压器热点温度反演

Transformer Hot-spot Temperature Inversion Based on Multi-physics Calculation and Fuzzy Neural Network Algorithm

热点温度是影响油浸式变压器绝缘寿命的重要因素之一。针对分布式光纤传感器测温系统测量热点温度时造成的不良影响与安装的不便性。提出一种基于多物理场计算和模糊神经网络算法的油浸式变压器热点温度反演模型。以66 kV三相三柱油浸自冷式变压器为例,采用有限体积法耦合迭代求解变压器流体–温度场,确定热点温度范围区域。筛选出环境温度、变压器顶层油温、变压器外壳及油箱顶部等12个特征点的温度建立模糊子集,再通过T–S模糊神经网络算法预测变压器绕组热点温度值。计算结果表明其与光纤传感器实际测量绕组热点温度误差<4%,比传统方法具有更高的精度。该方法实现了特征量测量非植入式,为电力变压器在线监测、热点温度计算提供一种新的计算思路。

The hot-spot temperature of oil-immersed transformers is one of the important factors affecting the insulation life of power transformers. The adverse impact and installation inconvenience may be aroused when measuring the hot-spot temperature by the distributed optical fiber sensor temperature measurement system, therefore, we proposed an oil-immersed transformer hot spot temperature prediction model based on fuzzy neural network algorithm. Taking 66 kV three-phase three-column oil-immersed self-cooled transformer as an example, we adopted the finite volume method to iteratively solve the transformer fluid-temperature field and determine the hot spot temperature range. The fuzzy subsets of 12 characteristic points such as ambient temperature, transformer top oil temperature, transformer casing and fuel tank top were screened, and the temperature of the transformer winding hot spot was predicted by the T-S fuzzy neural network algorithm. The results show that the temperature error of the winding hot spot measured with the fiber optic sensor is within 4% and it has higher precision than traditional method. This method realizes the non-implantable characteristic measurement and provides a new method for the calculation of hot spot temperature in power transformer on-line monitoring.