基於粒子群演算法結合S轉換及關聯規則特徵篩選之軸心偏離分析

Rotor Eccentricity Analysis Based on Particle Swarm Optimization Combined with S-transform and Association Rule Feature Selection

林承儒(Cheng-Ju Lin)

指導教授 ： 李俊耀

中原大學/電機資訊學院/電機工程研究所/碩士(2017年)

https://doi.org/10.6840/cycu201700610

摘要

本研究旨於提出粒子群模型最佳化S轉換參數分析(Particle swarm optimization-based S-transform, PST)結合關聯規則特徵篩選，針對風力發電機軸心偏離故障，進行訊號分析能力提升及特徵篩選，並以倒傳遞類神經網路與k最近鄰分類法進行故障檢測。
首先，本研究使用粒子群演算法(Particle swarm optimization, PSO)結合S轉換(S-transform, ST)，構成粒子群模型最佳化S轉換參數分析，其目的為得到擴張函數 內最佳化之參數 ，其次，應用關聯規則特徵篩選(Association rule feature selection, ARFS)，在訊號完成分析後擷取其特徵，再使用Apriori演算法找出特徵之關聯規則(Association rule)，並進行特徵篩選，減少劣質特徵數量，以提升辨識率。
結果顯示，對於訊號分析問題，以本研究提出之粒子群模型最佳化S轉換參數分析，相較於小波多重解析分析、希爾伯特-黃轉換及S轉換等方法，對於軸心偏離故障檢測具有較佳準確率，而對於特徵篩選問題，本研究再應用關聯規則特徵篩選，有效減少劣質特徵，進一步提升倒傳遞類神經網路分類器辨識準確率。

關鍵字

粒子群演算法 ； S轉換 ； 風力發電機 ； 特徵篩選 ； Apriori演算法 ； 關聯規則 ； 倒傳遞類神經網路

並列摘要

This study proposes a particle swarm optimization-based S-transform (PST) combined with association rule feature selection (ARFS) approach with the aims to enhance the detection accuracy of the rotor eccentricity of wind turbines. Back propagation neural network (BPNN) and k-nearest neighbor algorithm (k-NN) are applied to the features to recognize the rotor eccentricity faults.
First, this study combines a particle swarm optimization (PSO) algorithm with the S-transform to construct PST. The purpose of this method is to obtain the optimized parameter in the dilation to improve the detection accuracy.
Second, this study proposes a new feature selection scheme. This study extracts the features of current signals in signal analysis method, and inputs these features to the Apriori algorithm to find the association rules and selects the effective features subset of features set. The method can reduce the negative features as well as improving the BPNN classifier more accurately.
For the rotor eccentricity faults, the results indicate that the PST has better detection accuracy than wavelet transform (WT), Hilbert Huang transform (HHT) and S-transform (ST). Moreover, for the feature selection problem, the results show that the classification accuracies and the number of poor features of classifiers by using to the back propagation neural network (BPNN) is superior.

並列關鍵字

PSO ； ST ； wind turbines ； feature selection ； Apriori algorithm ； association rules ； BPNN

參考文獻

[1] M. F. Hsieh and Y. H. Yeh, “Rotor eccentricity effect on cogging torque of PM generators for small wind turbines,” IEEE Trans. on Magnetics, vol. 49, no. 5, pp. 1897-1900, May. 2013.

[2] I. Tabatabaei, J. Faiz, H. Lesani, and M. T. Nabavi-Razavi, “Modeling and simulation of a salient-pole synchronous generator with dynamic eccentricity using modified winding function theory,” IEEE Trans. on Magnetics, vol. 40, no. 3, pp. 1550-1555, May. 2004.

[3] D. F. Shi, P. J. Unsworth, and R. X. Gao, “Sensorless speed measurement of induction motor using Hilbert transform and interpolated fast Fourier transform,” IEEE Trans. on Instrumentation and Measurement, vol. 55, no.1, pp. 290-299, 2006.

[4] M. A. Ehteram, M. Sadighi, and H. B. Tabrizi, “Analytical solution for thermal stresses of laminated hollow cylinders under transient nonuniform thermal loading,” Mechanika, pp. 30-37, 2011.

[5] E. Strabgas, S. Aviyente, and S. Zaidi, “Time-frequency analysis for efficient fault diagnosis and failure prognosis for interior permanent ac motor,” IEEE Trans. on Industry Electronics, vol. 55, no. 12, pp. 4191-4199, Dec. 2008.