基于PSO-ELM的变压器油纸绝缘状态无损评估方法研究

油浸式电力变压器作为电网中的重要组成部分，其可靠运行至关重要。针对变压器长期运行后无法定量评估其绝缘状态的问题，本文开展了油纸绝缘模型的加速老化及受潮试验，探明了油纸绝缘老化及受潮程度对其回复电压曲线的影响规律，并提出采用粒子群优化（PSO）改进极限学习机(ELM)的参数预测方法，实现了基于回复电压曲线特征参量的油纸绝缘老化与受潮状态量化评估。通过对油纸绝缘模型理化性能分析的对比结果可知，基于PSO-ELM方法的预测值精度远高于传统ELM方法，油纸绝缘内含水率及纸板聚合度预测的绝对误差范围分别小于±0.4%、±30。

Study on the assessment method of transformer oil-paper insulation state based on PSO-ELM

Oil-immersed power transformer is an important part of power grid, that reliable operation is very important. Aiming at the problem that the insulation state of transformer cannot be assessed quantitatively after long-term operation, the accelerated aging and damp tests of oil-paper insulation model were carried out in this paper. The influence of aging and damp of oil-paper insulation on its recovery voltage curves were explored. The particle swarm optimization (PSO) was used to improve the parameter prediction method of Extreme Learning Machine (ELM), which realized the quantitative assessment of aging and moisture of oil-paper insulation based on the characteristic parameters of the recovery voltage curve. By comparing the physical and chemical performance analysis of oil-paper insulation models, it can be seen that the prediction accuracy based on PSO-ELM method is much higher than that of traditional ELM method. The absolute error range for predicting the moisture content of oil-paper insulation and the DP of pressboard is less than ± 0.4% and ± 30 respectively.