Objective Extracting the operating characteristic information of hydropower units and achieving multi-type fault diagnosis are key issues in the intelligent operation and maintenance of hydropower stations. This paper proposes a combination of a feature extraction method with good stability and strong noise resistance and a classifier. In addition, a multi-strategy whale optimization algorithm (MSWOA) is proposed to optimize the kernel-based extreme learning machine (KELM) classifier. Methods First, to address the problem that the length of the time series decreases as the segmentation scale increases, which may lead to the loss of critical data information and reduce the stability of multiscale approximate entropy (MAE), a new feature extraction method, termed time-shift multiscale approximate entropy (TSMAE), is proposed based on the concept of fractal theory. The stability of TSMAE is then verified using pink noise and white Gaussian noise (WGN) with seven different sequence lengths. The noise resistance of TSMAE is evaluated using simulated low-frequency eddy current signals with different noise levels. Furthermore, the feature extraction capability of TSMAE is verified by adding noise with different signal-to-noise ratios (SNRs, SNR) to rotor vibration test bench data and extracting features using TSMAE, assisted by T-distributed stochastic neighbor embedding (TSNE) visualization results. Second, to solve the problems that the whale optimization algorithm is highly dependent on the initial solution and easily falls into local optima during the optimization process, PWLCM chaotic mapping, a group individual memory strategy, and an adaptive T-distribution strategy are integrated to improve the algorithm, resulting in the MSWOA. Benchmark functions are then used to evaluate its performance. Next, because the hyperparameters of the KELM are difficult to tune, the MSWOA is employed to optimize its two important parameters and construct the classifier. The extracted features are subsequently identified and classified using the optimized classifier, and the diagnostic model is established through feature recognition and classification. Finally, the proposed method is verified using different datasets. Results and Discussion The performance verification results of TSMAE show that, with an increase in the scale factor, the MAE distributions under the two types of noise differ considerably. In particular, under WGN, MAE exhibits strong fluctuations. By contrast, the TSMAE values tend to decrease under both types of noise. Meanwhile, the TSMAE curves remain smooth and exhibit only slight fluctuations under different SNR conditions. Taking SNR = 10 dB as an example, the MAE curve differs considerably from those under other SNR conditions and shows stronger fluctuations. Therefore, TSMAE demonstrates superior overall performance. To evaluate the basic performance and convergence speed of the MSWOA, it was compared with particle swarm optimization (PSO), sine cosine algorithm (SCA), Harris hawks optimization (HHO), WOA, and grey wolf optimizer (GWO). Because the superiority of an optimization algorithm is difficult to prove solely through theoretical reasoning, the CEC 2005 benchmark functions were selected to compare the performance of the different algorithms. To reduce the influence of randomness on the experimental results, 50 repeated experiments were conducted, and the mean values and standard deviations were evaluated. The results show that, compared with the theoretical optimal value, the MSWOA achieves the best performance on the test functions and exhibits the smallest standard deviations. Therefore, the MSWOA demonstrates superior convergence accuracy and a stronger ability to escape local optima. In the rotor vibration test section, to further highlight the advantages of TSMAE in feature extraction, multiscale permutation entropy (MPE), time-shift multiscale permutation entropy (TSMPE), multiscale dispersion entropy (MDE), time-shift multiscale dispersion entropy (TSMDE), MAE, and TSMAE were compared and visualized using TSNE. The visualization results indicate that MPE performs poorly, with scattered characteristic signals that make the unit states difficult to distinguish. Although TSMPE can completely distinguish the Rub state and produces relatively compact features, some Nor and Imb states remain aliased. By contrast, the Nor, Mis, and Imb states extracted by MDE cannot be distinguished completely, and substantial overlap exists among the features. The performance of TSMDE is clearly better than that of MDE, and the characteristic signals are more compact; however, a small amount of overlap between Nor and Imb still exists. The performance of MAE is better than that of MPE and MDE, but the characteristic signal distribution is insufficiently compact, and the four states cannot be completely distinguished. Compared with TSMPE and TSMDE, TSMAE can effectively distinguish different unit states, except for slight overlap between individual Mis and Imb signals. Overall, TSMAE exhibits the best feature extraction performance among the compared methods. To more intuitively demonstrate the diagnostic performance of different methods, the MSWOA was used to optimize the KELM for classification of the above feature extraction results. As MPE and MDE exhibit poor feature extraction performance, only MAE, TSMPE, TSMDE, and TSMAE were compared. The results show that the TSMAE‒MSWOA‒KELM model achieved a 100.00% diagnostic rate, with no misclassification among different unit states. The classification accuracy based on MAE feature extraction was 97.2%, with four misclassifications occurring during the classification process. The diagnostic rates of TSMPE and TSMDE were 98.6% and 99.3%, respectively. These results indicate that TSMAE provides the best feature extraction performance, while the proposed TSMAE‒MSWOA‒KELM model achieves the best diagnostic performance. Furthermore, 10-fold cross-validation method was used to evaluate the model. Under the MSWOA‒KELM classifier, the mean diagnostic accuracies of MAE, TSMPE, and TSMAE were 98.54%, 99.58%, 99.44%, and 100.00%, respectively. TSMAE achieved the best diagnostic performance, as well as the best algorithm stability and generalization capability, thereby demonstrating the superiority of time-shift multiscale entropy. For the measured data of the hydropower unit, two early-warning-state samples classified by MAE classification were misjudged as normal states, resulting in a classification accuracy of 97.14%. By contrast, TSMAE produced no misclassification, and the fault recognition rate remained 100.00%. The results indicate that the proposed method improves diagnostic accuracy by 1.46% and 2.86% on different datasets, further verifying the superiority of the proposed approach. Conclusions The results demonstrate the proposed method is suitable for the fault diagnosis of hydropower units. Compared with traditional methods, the proposed method exhibits better feature extraction capability and stability and demonstrates significant advantages over other entropy-based feature extraction and diagnostic methods, thereby providing a new approach for hydropower unit fault diagnosis.
ZhangTing, WuYunna.Collaborative allocation model and balanced interaction strategy of multi flexible resources in the new power system based on Stackelberg game theory[J].Renewable Energy,2024,220:119714. doi:10.1016/j.renene.2023.119714
[2]
YangXiuyuan, FanXindong, ChenQiyu.Control strategy of hydropower plant considering vibration region and its application in coordinated hydro and wind power generation[J].Proceedings of the CSEE,2019,39(18):5433‒5441. doi:10.13334/j.0258-8013.pcsee.190365
WangWeiyu, HeKuidong, JinYan,et al.Feature extraction of vibration signal of hydropower unit based on CEEMDAN sample entropy and PSO‒SVM[J].Engineering Journal of Wuhan University,2022,55(11):1167‒1175.
ChenFei, WangBin, LiuTing,et al.Fault diagnosis of hydropower units based on time-shifted multiscale attention entropy and random forest[J].Journal of Hydraulic Engineering,2022,53(3):358‒368. doi:10.13243/j.cnki.slxb.20210941
TongYu, PangXinyu, WeiZihan.Fault diagnosis method of rolling bearing based on GADF‒CNN[J].Journal of Vibration and Shock,2021,40(5):247‒253. doi:10.13465/j.cnki.jvs.2021.05.032
XuZhexi, LiuTing, RenShengmin,et al.Fault diagnosis method of hydropower units based on time-shifted multiscale fluctuation dispersion en-tropy and improved kernel extreme learning machine[J].Advanced Engineering Sciences,2024,56(3):41‒51.
HoudekV, SmolíkL, KubínZ.Impact point localization in three-dimensional structures using wavelet transform[J].Mechanical Systems and Signal Processing,2022,179:109365. doi:10.1016/j.ymssp.2022.109365
[13]
LiuDong, XiaoZhihuai, HuXiao,et al.Feature extraction of rotor fault based on EEMD and curve code[J].Measurement,2019,135:712‒724. doi:10.1016/j.measurement.2018.12.009
[14]
ZhaoZeni, YunSining, JiaLingyun,et al.Hybrid VMD‒CNN‒GRU-based model for short-term forecasting of wind power considering spatio-temporal features[J].Engineering Applications of Artificial Intelligence,2023,121:105982. doi:10.1016/j.engappai.2023.105982
[15]
TanHongchuang, XieSuchao, ZhouHui,et al.Sensible multiscale symbol dynamic entropy for fault diagnosis of bearing[J].International Journal of Mechanical Sciences,2023,256:108509. doi:10.1016/j.ijmecsci.2023.108509
[16]
ZhouFuming, YangXiaoqiang, ShenJinxing,et al.Modified multivariate hierarchical fluctuation dispersion entropy and its application to the fault diagnosis of rolling bearings[J].Journal of Vibration and Shock,2021,40(22):167‒174. doi:10.13465/j.cnki.jvs.2021.22.023
ChenFei, ZhangLiyao, LiuWenshen,et al.A fault diagnosis method of rotating machinery based on improved multiscale attention entropy and random forests[J].Nonlinear Dynamics,2024,112(2):1191‒1220. doi:10.1007/s11071-023-09126-x
JiangWenjun, HuXiao, ZhangPei,et al.Vibration feature extraction for hydropower units based on ensemble empirical mode decomposition and approximate entropy[J].Journal of Hydroelectric Engineering,2020,39(6):18‒27. doi:10.11660/slfdxb.20200602
SunLiting, WangXiang, YangAfeng,et al.Radio frequency fingerprint extraction based on multi-dimension approximate entropy[J].IEEE Signal Processing Letters,2020,27:471‒475. doi:10.1109/LSP.2020.2978333
[23]
ZhaoZhigao, ChenFei, HeXianghui,et al.A universal hydraulic-mechanical diagnostic framework based on feature extraction of abnormal on-field measurements:Application in micro pumped storage system[J].Applied Energy,2024,357:122478. doi:10.1016/j.apenergy.2023.122478
[24]
ZhangWenjing, GaoYuanchen, PengShizhe,et al.Fault diagnosis of hydroelectric units based on a novel multiscale fractional-order weighted permutation entropy[J].Fractal and Fractional,2022,6(10):588. doi:10.3390/fractalfract6100588
[25]
KokluM, UnlersenM F, Ali OzkanI,et al.A CNN‒SVM study based on selected deep features for grapevine leaves classification[J].Measurement,2022,188:110425. doi:10.1016/j.measurement.2021.110425
[26]
LuanXiaochi, ZhangXi, ShaYundong,et al.Method on inter-shaft bearing fault diagnosis based on extreme learning machine optimized by gray wolf optimizer[J].Journal of Propulsion Technology,2024,45(4):190‒200.
MaYanli, ChengJunsheng, WangPing,et al.A new rotating machinery fault diagnosis method for different speeds based on improved multivariate multiscale fuzzy distribution entropy[J].Nonlinear Dynamics,2023,111(18):16895‒16919. doi:10.1007/s11071-023-08609-1
[29]
XiLei, HeMiao, ZhouBoqi,et al.Research on false data injection attack detection in power system based on improved multi layer extreme learning machine[J].Acta Automatica Sinica,2023,49(4):881‒890. doi:10.16383/j.aas.c211127
AnLiuming, ShaDesheng, ZhangQing,et al.Research on intelligent fault diagnosis of wind turbine based on WOA‒KELM algorithm[J].Thermal Power Generation,2023,52(12):131‒139.
ZahidM U, NisarM D, ShahM H,et al.Specific emitter identification based on multi-scale multi-dimensional approximate entropy[J].IEEE Signal Processing Letters,2024,31:850‒854. doi:10.1109/lsp.2024.3375264
[34]
TangShengxue, MaChenyang, GouZe.Early fault state identification method of the rod control system power equipment based on time-frequency characteristics fusion and GWO‒ELM[J].Chinese Journal of Scientific Instrument,2023,44(1):121‒130.
MirjaliliS, LewisA.The whale optimization algorithm[J].Advances in Engineering Software,2016,95:51‒67. doi:10.1016/j.advengsoft.2016.01.008
[37]
HuangQian, LiuSheng, LiMengmeng,et al.Multi-strategy chimp optimization algorithm and its application of engineering problem[J].Computer Engineering and Applicatio-ns,2022,58(19):174‒183. doi:10.3778/j.issn.1002-8331.2101-0520
ChenXuanru, WuLifei, YangXiaozhong.Fractional order PID parameter tuning based on improved sparrow search algorithm[J].Control and Decision,2024,39(4):1177‒1184. doi:10.13195/j.kzyjc.2022.1360
HuXiao, XiaoZhihuai, LiuDong,et al.Vibration fault identification method of hydropower unit based on EEMD‒SDCCⅠ‒HMM[J].Journal of Vibration and Shock,2022,41(3):165‒175.
KangYansong, ZangShunlai.Improved particle swarm optimization algorithm based on multiple strategies[J].Jo-urnal of Northeastern University(Natural Science),2023,44(8):1089‒1097.
MirjaliliS.SCA:A Sine Cosine Algorithm for solving optimization problems[J].Knowledge‒Based Systems,2016,96:120‒133. doi:10.1016/j.knosys.2015.12.022
[46]
HeidariA A, MirjaliliS, FarisH,et al.Harris Hawks optimization:Algorithm and applications[J].Future Generation Computer Systems,2019,97:849‒872. doi:10.1016/j.future.2019.02.028
[47]
LuoJinkun, HeFazhi, GaoXiaoxin.An enhanced grey wolf optimizer with fusion strategies for identifying the parameters of photovoltaic models[J].Integrated Computer‒Aided Engineering,2022,30(1):89‒104. doi:10.3233/ica-220693
[48]
ZhaoXin, WangDongli, PengHong,et al.Transformer fault diagnosis based on a multi-strategy improved dung beetle optimizer[J].Power System Protection and Control,2024,52(6):120‒130.
LiuDong, LaiXu, HuXiao,et al.Research on on-line evaluation method of state degradation of hydropower unit based on vibration signal[J].Journal of Hydraulic Engineering,2021,52(4):461‒473. doi:10.13243/j.cnki.slxb.20200298