大温差对复合绝缘子硅橡胶老化特性及运行寿命预测研究

Study on ageing characteristics and operating life prediction of silicone rubber for composite insulators under large temperature difference

藏东南地区环境温差较大,对复合绝缘子的绝缘性能产生负面影响,从而影响西电东送规模持续可靠送出。本文在-20~150℃下对硅橡胶试样开展720 h的高低温循环老化试验,对不同循环次数的硅橡胶进行各项测试用于研究其大温差环境老化性能变化规律。对测试的12个特征量通过Fisher Score进行特征量选择,筛选出4种显著相关性特征量。以运行0~11年复合绝缘子为研究对象,基于相关性特征量对在役复合绝缘子进行性能测试和结果分析,并提出一种改进遗传算法优化BP(back propagation)神经网络的预测算法。该算法一方面改进最优保存策略选择算子,另一方面迭代过程动态调整变异概率和交叉概率。结果表明:4项老化特征量拉伸强度、介质损耗因数tanδ、TGA最终剩余比例和Si-OH透过率下降率具有显著的相关性;相比传统BP神经网络和GA-BP(genetic algorithm-back propagation)神经网络,改进GA-BP神经网络的非线性学习和全局寻优能力更强,网络收敛速度更快;以老化11年的1组试样进行误差检验,改进GA-BP神经网络的检验误差结果为2.33%,5组复合绝缘子的运行寿命预测值与实际年限之间的误差均在5%以内。

The temperature difference in the southeastern region of Tibet is relatively large,which has negative influence on the insulation performance of composite insulators,and affects the continuous and reliable transmission of West East power.Silicone rubber samples were conducted 720 hours of high and low temperature cyclic ageing tests at-20-150℃ in this paper,and the silicone rubber sampes with different cycles were conducted various tests to study the changes of ageing performance in large temperature difference environments.The 12 kinds of tested characteristic parameters were conducted selection of characteristic parameters using Fisher Score,and 4 kinds of significantly correlated characteristic parameters were selected.Taking composite insulators running for 0-11 years as the research object,we conducted performance testing and result analysis on in-service composite insulators based on correlated characteristic parameters,and proposed an improved genetic algorithm to optimize the BP(Back Propagation) neural network prediction algorithm.On the one hand,this algorithm improves the optimal preservation strategy selection operator,and on the other hand,this algorithm dynamically adjusts the mutation probability and crossover probability during the iterative process.The results show that the four ageing characteristic parameters,which are tensile strength,dielectric loss factor tanδ,TGA residual ratio,and Si-OH transmittance reduction rate,are significantly correlated.Compared with the traditional BP and GA-BP(genetic algorithm back propagation) neural networks,the improved GA-BP neural network has stronger nonlinear learning and global optimization capability,and faster network convergence speed.The test error result of the improved GA-BP neural network on a group of samples aged for 11 years is 2.33%.The error between the predicted operating life and the actual service life of five groups of composite insulators is within 5%.