基于Transformer和ARMA双数据驱动模型的抽水蓄能机组劣化趋势集成预测  

Integrated Prediction of Deterioration Trends for Pumped Storage Units Based on Transformer and ARMA Dual Data-Driven Models

作　　者：钟子威 祝令凯 郭俊山 郑威 巩志强 商攀峰 ZHONG Zi-wei;ZHU Ling-kai;GUO Jun-shan;ZHENG Wei;GONG Zhi-qiang;SHANG Pan-feng(State Grid Shandong Electric Power Research Institute,Jinan 250003,China;Shandong Smart Grid Technology Innovation Center,Jinan 250003,China)

机构地区：[1]国网山东省电力公司电力科学研究院,山东济南250003 [2]山东省智能电网技术创新中心,山东济南250003

出　　处：《水电能源科学》2025年第3期191-195,共5页Water Resources and Power

基　　金：国网山东省电力公司电力科学研究院自主研发项目(ZY-2024-11)。

摘　　要：为更精准地预测抽水蓄能机组劣化趋势,提出了一种基于Transformer和自回归滑动平均(ARMA)双数据驱动模型的抽水蓄能机组劣化趋势集成预测方法。该方法先利用完全自适应噪声集成经验模态分解对CatBoost模型构建的劣化序列进行分解,再根据分解所得分量的不同时间尺度特性,利用Transformer模型对非线性分量进行预测,利用ARMA模型对线性分量进行预测,最后将预测值叠加得到最终预测结果。利用某抽水蓄能机组监测数据进行试验,结果表明,所提方法具有较好的预测性能,能够有效提高抽水蓄能机组劣化趋势预测准确性。To predict the deterioration trend of pumped storage units more accurately,a dual data-driven integrated prediction method was proposed.This method utilized both Transformer and autoregressive moving average(ARMA)models to obtain the prediction process based on dual data sources.Firstly,the degradation sequences that constructed by the CatBoost model were decomposed using complete ensemble empirical mode decomposition with adaptive noise.Subsequently,considering the different temporal characteristics of the decomposition components,the Transformer model was employed to predict the nonlinear parts,while the ARMA model was used for the linear components.Finally,the predicted values were aggregated to obtain the ultimate forecast.Experimental validation using monitoring data from a specific pumped storage unit demonstrates the effectiveness of the proposed method,indicating improved prediction performance and enhanced accuracy in forecasting the degradation trends of pumped storage units.

关 键 词：劣化趋势预测 完全自适应噪声集成经验模态分解 TRANSFORMER 自回归滑动平均 

分 类 号：TV734.21[水利工程—水利水电工程] TK730[交通运输工程—轮机工程]