作者：田宏伟¹, 李志鹏², 王煜伟¹, 孙钢虎², 杨沛豪²

作者单位：1. 国家能源集团谏壁发电厂，江苏 镇江 212006;
2. 西安热工研究院有限公司，陕西 西安 710000

关键词：振动;完备经验模态分解;极限学习机;预测

摘要：

在火电厂中，风机的故障通常会引起风机振动幅值异常，因此对风机振动趋势的准确预测可以有效降低风机故障发生概率。由于原始的风机振动数据具有较强的随机性和波动性，传统预测方法很难直接进行有效预测，因此需要对原始风机振动数据进行预处理，并应用先进的机器学习算法来进一步提高风机振动预测精度。该文采用完备经验模态分解（complete ensemble empirical mode decomposition adaptive noise, CEEMDAN）对原始数据进行预处理，将原始振动数据分解为一系列固态模量（intrinsic mode function，IMF），从而降低原始振动信号的非平稳性。其次使用经过鲸鱼算法（whale optimization algorithm, WOA）优化的极限学习机（extreme learning machine, ELM）来预测所有IMF序列。最后将所有IMF_S预测结果叠加得到最终预测值。为评估模型的预测性能，该研究采集某火电厂风机机组的振动数据进行多组对比试验。结果表明，该文提出的模型SSE平均降低39.58%，RMSE平均降低31.73%，验证CEEMDAN-WOA-ELM模型具有优越的数据处理和预测能力，适用于火电厂中风机振动的趋势预测。

Fan vibration trend prediction based on CEEMDAN-WOA-ELM model
TIAN Hongwei¹, LI Zhipeng², WANG Yuwei¹, SUN Ganghu², YANG Peihao²
1. National Energy Group Jianbi Power Plant, Zhenjiang 212006, China;
2. Xi’an Thermal Power Research Institute Co., Ltd., Xi’an 710000, China
Abstract: In the thermal power plant, the failure of the fan usually causes abnormal vibration amplitude of the fan. Therefore, accurate prediction of the trend of fan vibration can effectively reduce the probability of fan failure. Due to the strong randomness and fluctuation of the original fan vibration data, traditional prediction methods are difficult to directly perform effective predictions. Therefore, it is necessary to pre-process the original fan vibration data and apply advanced machine learning algorithms to further improve the prediction of fan vibration. In this paper, complete empirical mode decomposition adaptive noise (CEEMDAN) is used to preprocess the original data and decompose the original vibration data into a series of intrinsic mode functions (IMF), thereby reducing the non-stabilaty of original vibration signal. Non-stationarity. Secondly, an extreme learning machine (ELM) optimized by a whale optimization algorithm (WOA) is used to predict all IMF sequences. Finally, all IMF_S prediction results are superimposed to obtain the final prediction value. In order to evaluate the predictive performance of the model, this study collected vibration data from a fan plant in a thermal power plant and conducted multiple sets of comparative tests. The results show that the model proposed in this paper has an average reduction of 39.58% and an average reduction of RMSE of 31.73%, which verifies that the CEEMDAN-WOA-ELM model has superior data processing and prediction capabilities and is suitable for the prediction of fan vibration trends in thermal power plants.
Keywords: vibration;complete empirical mode decomposition;extreme learning machine;prediction
2020, 46(7):146-152  收稿日期: 2019-10-21;收到修改稿日期: 2019-12-06
基金项目: 国家能源江苏公司科技项目(JSKJ-2019-01)
作者简介: 田宏伟(1966-),男,江苏镇江市人,高级工程师,硕士,主要从事电厂生产技术管理工作
参考文献
[1] 张建伟, 江琦, 朱良欢, 等. 基于改进HHT的泵站管道工作模态辨识[J]. 农业工程学报, 2016, 32(2): 71-76
[2] 刘晓霞. 基于有限元分析双层海底管道断裂失效问题[J]. 管道技术与设备, 2016(5): 14-16
[3] 陈林. 泄洪闸闸墩原型振动测试、预测与安全评价[J]. 振动、测试与诊断, 2014, 34(5): 938-946
[4] 练继建, 张龑, 刘防, 等. 厂顶溢流式水电站振源特性研究[J]. 振动与冲击, 2013, 32(18): 8-14
[5] 王海军, 毛柳丹, 练继建. 基于RVM方法的水电站厂房结构振动预测研究[J]. 振动与冲击, 2015, 34(3): 23-27
[6] 张松兰. 支持向量机的算法及应用综述[J]. 江苏理工学院学报, 2016, 22(2): 14-17
[7] 黎慧. 基于EMD和逻辑回归的轴承性能退化评估与剩余寿命预测[D]. 南昌：华东交通大学, 2017.
[8] 剡昌锋, 易程, 吴黎晓,等. 基于EEMD和ARIMA模型的汽轮机故障趋势预测[J]. 甘肃科学学报, 2016, 28(4): 100-106
[9] ASMA C. New directional bat algorithm for cotinuous optimization problems[J]. Expert Syst Appl, 2016, 69: 159-175
[10] HUANG N, SHEN Z, LONG S, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J]. Proc R Soc A Math Phys Eng Sci, 1998, 454(995): 903
[11] TONG W, ZHANG M, YU Q, ZHANG H. Comparing the applications of EMD and EEMD on time-frequency analysis of seismic signal[J]. J Appl Geophys, 2012, 83: 29-34
[12] WU Z, HUANG N. Ensemble empirical mode decom-position: a noise-assisted data analysis method[J]. Adv Adapt Data Anal, 2009, 1: 6281-4
[13] SAADAT B, RAHMAT H, MOEIN P. Short term electric load forecasting by wavelet transform and grey model improved by PSO (particle swarm optimization) algorithm[J]. Energy, 2014, 72.