基于ICEEMDAN-IPSO-ELM的硅油溶解气体浓度组合预测方法

Combined Prediction Method of Dissolved Gas Concentration of Silicone Oil Based on ICEEMDAN-IPSO-ELM

高压电缆充油终端作为电力系统中传输电能的重要设备,对充油电缆终端内填充的硅油溶解气体浓度进行可靠预测,可为硅油的故障诊断提供一定的支撑。因此,提出一种基于局部异常因子与ICEEMDAN-IPSO-ELM的硅油中溶解气体浓度预测模型。首先,搭建模拟电缆终端内部硅油老化实验平台,通过色谱分析获得硅油中溶解气体浓度序列,进而对硅油中溶解气体浓度时间序列进行数据清洗,采用局部离群因子检测方法判断异常值并进行合理的修正,进而采用改进自适应白噪声完全集合经验模态分解将修正后的硅油中溶解气体浓度序列进行分解,得到不同时间尺度的本征模态函数分量,可以有效降低高、低频分量间的相互影响;其次,针对具有不同特征的频率分量搭建极限学习机网络预测模型,针对极限学习机模型参数较难选取的问题,采用改进粒子群优化方法对模型的权值和阈值参数寻优求解,在一定程度上优化了粒子群方法的寻优能力,并提高了组合预测方法的可靠性;最后,将不同频率分量的计算结果加和,便可得到硅油中溶解气体浓度的预测含量。具体实例表明,与其他预测模型相比,该方法能够可靠预测出硅油中溶解气体含量的未来走势,为硅油故障诊断技术提供了有力的保障。

As an important device for transmitting electric energy in the power system,the oil filled terminal of high-voltage cable can reliably predict the concentration of dissolved gas of silicone oil filled in the oil filled terminal,which can provide some supports for the fault diagnosis of silicone oil.Therefore,this paper proposes a prediction model of dissolved gas concentration in silicone oil based on local outlier factor and ICEEMDAN-IPSO-ELM.Firstly,an experimental platform is built to simulate the aging of silicone oil inside the cable terminal,and the concentration sequence of dissolved gas in silicone oil is obtained through chromatographic analysis.Furthermore,the data of the concentration time series of dissolved gas in silicone oil are cleaned,the local outlier detection method is used to judge the abnormal value and make reasonable correction,and then the improved adaptive noise complete set empirical mode decomposition is used to decompose the corrected concentration sequence of dissolved gas in silicone oil.The eigenmode function components with different time scales can effectively reduce the interaction between high and low frequency components.Secondly,aimed at the frequency components with different characteristics,an extreme learning machine network prediction model is built.Aimed at the problem of difficult selection of extreme learning machine model parameters,an improved particle swarm optimization algorithm is used to optimize the weight and threshold parameters of the model,which optimizes the optimization ability of particle swarm optimization method to a certain extent and improves the reliability of combined prediction method.Finally,the predicted content of dissolved gas concentration in silicone oil can be obtained by adding the calculation results of different frequency components.Specific examples show that,compared with other prediction models,this method can be adopted to reliably predict the future trend of dissolved gas content in silicone oil,and provides a strong guarantee for silicone oil fault diagnosis technology.