风机长期运行、部件结构复杂以及所处环境恶劣，导致其部件故障率居高不下，但是部件在发生故障之前，该部件的参数会发生变化。针对目前风机故障部件参数预测精度不高，本文以提高风机故障部件参数预测精确度度为研究目标，从特征选择、预测模型网络结构和预测模型参数优化三个方面对风机故障部件参数预测方法进行研究，主要研究内容如下： 

首先，提出了一种基于专家经验和改进Relief-F的两阶风机故障部件参数特征选择方法。对河北多个风场中风机部件故障统计和故障原因分析，选定故障频率较高的三桨叶位置不同步故障和发电机轴承温度超限故障作为研究对象。针对传统特征选择算法计算效率低、冗余信息确定困难等问题，提出了两阶段故障参数特征选择方法，通过第一阶段专家经验筛选和第二阶段改进Relief-F算法筛选，剔除特征量的冗余无效信息，筛选出与故障参数相关性强的特征变量，据此筛选出三桨叶位置不同步故障强相关的8个特征变量和发电机轴承温度超限故障强相关的8个特征变量。

其次，构建了基于LSTM-Attention的风机故障部件参数预测模型，并据此进行故障部件参数预测与验证分析。针对长短期记忆网路（Long Short-Term Memory, LSTM）处理部件故障参数数据出现信息丢失的问题，通过在LSTM网络隐藏层与全连接层之间构建一个Attention层，利用Attention机制来重新分配LSTM隐藏层的输出比重，增强LSTM模型对故障参数特征关键信息的敏感度，有效解决了LSTM处理长时间序列特征遗忘的缺陷。最后对桨距角和发电机轴承温度进行预测，通过实验仿真进行模型验证并对比其它的预测方法，结果表明，采用LSTM-Attention模型的预测精度更高，验证了所提模型的有效性。

最后，再针对LSTM- Attention神经网络参数选择困难、秃鹰搜索算法(Bald Eagle Search, BES)存在陷入局部最优和收敛精度低等问题，提出一种改进秃鹰搜索算法(Ct-GBES)，使用该算法对预测模型参数进行寻优，将优化结果最优参数赋给LSTM- Attention模型，最终构建出Ct-GBES-LSTM-Attention预测模型。采用河北风电场监测的有关数据对风机故障部件参数进行预测并与其它方法进行对比，证明所提预测模型的有效性，并利用所提预测模型对选定故障进行预警验证。结果表明：采用Ct-GBES-LSTM-Attention预测模型相比其它优化算法模型的预测精度更高，并且两种部件在发生故障前均能发出异常状态信号并及时报警，验证了所提模型的有效性和实用性。

The long-term operation, complex component structure, and harsh environment of the fan result in a high failure rate of its components. However, the parameters of the component will change before it fails. In response to the current low accuracy of parameter prediction for wind turbine faulty components, this paper aims to improve the accuracy of parameter prediction for wind turbine faulty components. The research focuses on three aspects: feature selection, prediction model network structure, and optimization of prediction model parameters. The main research contents are as follows:

Firstly, a two-stage wind turbine fault component parameter feature selection method based on expert experience and improved Relief-F is proposed. Statistical analysis and root cause analysis of wind turbine component failures in multiple wind farms in Hebei Province were conducted. Three blade position asynchronous faults and generator bearing temperature exceeding faults with high failure frequencies were selected as the research objects. Aiming at the problems of low computational efficiency and difficulty in determining redundant information in traditional feature selection algorithms, a two-stage fault parameter feature selection method is proposed. Through the first stage of expert experience screening and the second stage of improved Relief-F algorithm screening, redundant and invalid information of feature quantities is removed, and feature variables with strong correlation with fault parameters are selected. Based on this, 8 feature variables strongly related to asynchronous three blade position faults and 8 feature variables strongly related to generator bearing temperature exceeding faults are selected.

Secondly, a wind turbine fault component parameter prediction model based on LSTM- Attention was constructed, and based on this, fault component parameter prediction and verification analysis were carried out. In response to the problem of information loss in processing component fault parameter data using Long Short Term Memory networks, an Attention layer is constructed between the hidden layer and fully connected layer of the LSTM network. The Attention mechanism is used to redistribute the output weight of the LSTM hidden layer, enhancing the sensitivity of the LSTM model to key information of fault parameter features and effectively solving the problem of feature forgetting in LSTM processing long time series. Finally, the pitch angle and generator bearing temperature were predicted, and the model was validated through experimental simulation and compared with other prediction methods. The results showed that the LSTM Attention model had higher prediction accuracy, verifying the effectiveness of the proposed model.

Finally, in response to the difficulty in selecting parameters for LSTM Attention neural networks and the problems of local optima and low convergence accuracy in Bald Eagle Search  algorithm, an improved Bald Eagle Search algorithm  is proposed. This algorithm is used to optimize the parameters of the prediction model, assign the optimal parameters to the LSTM Attention model, and ultimately construct the Ct-GBES-LSTM-Attention prediction model. Using relevant data monitored by Hebei wind farm to predict the parameters of wind turbine faulty components and comparing them with other methods, the effectiveness of the proposed prediction model is demonstrated, and the selected faults are verified through early warning using the proposed prediction model. The results showed that the Ct-GBES-LSTM-Attention prediction model had higher prediction accuracy compared to other optimization algorithm models, and both components could emit abnormal state signals and give timely alarms before faults occurred, verifying the effectiveness and practicality of the proposed model.

[1]孙毅, 谷家训, 郑顺林, 等. 考虑广义储能和LCA碳排放的综合能源系统低碳优化运行策略[J]. 上海交通大学学报, 2023, 46(12): 18-24.

[2]和萍, 祁盼, 申润杰, 等. 计及风电和燃料电池的综合能源系统阻尼特性分析[J]. 电力科学与技术学报, 2020, 35(01): 14-23.

[3]李强年, 赵巧妮, 聂龑. 基于层次分析法(AHP)的风电产业发展影响因素探析—以甘肃省酒泉风力发电基地为例[J]. 电网与清洁能源, 2019,35(10): 75-81.

[4]刘吉成, 何丹丹, 龙腾. 适应能源互联网需求的风力发电数据集成研究[J]. 电网技术, 2017, 41(03): 978-984.

[5]林楚. 《全球风能报告2024》发布风电行业进入加速增长新时代[N]. 机电商报, 2024-04-22(A07).

[6]周浩豪. 基于数据驱动的火电厂风机故障预警研究[D]. 上海：上海电力大学, 2023.

[7]孟召潮. 基于SCADA数据的风机传动系统健康度评估及故障类型预测[D]. 河北： 河北工业大学, 2021.

[8]Quan H, Srinivasan D, Khosravi A. Short-term load and wind power forecasting using neural network-based forecasting intervals[J]. IEEE Transactions on Neural Networks & Learning Systems 2017; 25(12): 303-315.

[9]Wu Y K , Ching-Ying L , Shao-Hong T , et al. Actual experience on the short-term wind power forecasting at Penghu — From an island perspective[C]// 2010 International Conference on Power System Technology. Proceedings of Science and Technology(02). Shanghai: Shanghai University Press, 2010: 10-15.

[10]Alpaydin E. Japanese Journal of Medical Physics An Official Journal of Japan Society of Medical Physics[J]. Introduction to Machine Learning, 2004, 96(15): 183-192.

[11]Liu Z P, Zhang L, Carrasco J. Vibration anslysis for large-scale wind turbine blade bearing fault detection with an empirical wavelet theesholding method[J]. Renewable Energy, 2020, 146(12): 99-110.

[12]Huang D J, Zhang W A, Guo F H, et al. Wavelet packet decomposition-based multiscale CNN for fault diagnosis of wind turbine gearbox[J]. IEEE transactions on cybernetics, 2021, 11(10): 3123667.

[13]郑小霞, 李美娜. 基于小波包和并行隐马尔科夫的风力机易损部件健康状态评价[J].太阳能学报, 2019, 40(02): 370-379.

[14]张振海, 王维庆, 王海云, 等. 基于改进小波包的风电机组齿轮箱复合故障特征提取研究[J]. 太阳能学报, 2022, 43(09): 331-336.

[15]齐咏生, 单成成, 高胜利, 等. 基于AEWT-KELM的风电机组轴承故障诊断策略[J].太阳能学报, 2022, 43(08): 281-291.

[16]Pang B, Tian T, Tang G J. Fault state recognition of wind turbine gearbox based on generalized multi-scale dynamic time warping[J]. Structural Health Monitoring, 2021, 20(06): 3007-3023.

[17]Cho S, Choi M, Gao Z, et al. Fault detection and diagnosis of a blade pitch system in a floating wind turbine based on Kalman filters and artificial neural networks[J]. Renewable Energy, 2021, 169(09): 1-13.

[18]Qiu Y, Chen L, Feng Y, et al. An approach of quantifying gear fatigue life for wind turbine gearboxes using supervisory control and data acquisition data[J]. Energies, 2017, 10(08): 1084-1085.

[19]Zhang Y, Wang K, Qian X, et al. Robust fault-detection based on residual KL divergence for wind turbines[J]. IET Renewable Power Generation, 2019, 13(13): 2400-2408.

[20]王正, 王增全, 何洪. 随机载荷循环作用下的机械结构疲劳寿命预测模型[J]. 中国机械工程, 2012, 23(01): 98-101.

[21]张天一. 风电齿轮箱轴承疲劳寿命研究[D]. 吉林：吉林大学, 2012.

[22]林锋. 风电齿轮箱齿轮的疲劳寿命及点蚀下可靠度的计算[D]. 浙江：浙江理工大学, 2018.

[23]赵麦浪. 风力发电机组变桨轴承疲劳寿命影响因素探究[J]. 通讯世界,2019, 26(07): 231-232.

[24]罗静, 程家龙, 杨立波, 等. 基于非齐次泊松过程的数控系统可靠性建模[J]. 现代制造工程, 2019, 22(09): 137-140.

[25]徐宗贤, 戴建军. 风力发电机组齿轮箱弹性支撑橡胶疲劳寿命分析[J]. 机电工程, 2019, 36(11): 1177-1182.

[26]LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.

[27]Gers F A, Schmidhuber J, Cummins F. learning to forget: Continual preciction with LSTM[J]. Neural Conputation, 2000, 12(10): 2451-2471.

[28]Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507.

[29]Hinton G E, Osindero S, Teh Y W. A fast learning algoruthm for deep belief nets[J]. Neural Conputation, 2006, 18(7): 1527-1554.

[30]王东林, 吕丽霞, 王梓齐, 等. 基于GMM工况辨识和DAE的风机齿轮箱状态监测[J]. 中国测试, 2021,47(04): 89-95.

[31]Cheng X, Shi F, Zhao M, et al. Temporal attention convolutional neural network for estimation of icing probability on wind turbine blades[J]. IEEE Transactions on Industrial Electronics, 2021, 69(06): 6371-6380.

[32]Ojanssens O , Slavkovikj V, Vervisch B, et al. Convolutional neural network based fault detection for rotating machinery [J]. Journal of Sound and Vibration, 2016 20(08): 1563-1579.

[33]Babu G S, Zhao P , Li X L. Deep convolutional neural network based regression approach for estimation of remaining useful life[J]. International Conference on Database Systems for Advanced Applications, 2016, 21(07): 214-228.

[34]魏书荣, 张鑫, 符杨, 等. 基于GRA-LSTM-Stacking模型的海上双馈风力发电机早期故障预警与诊断[J]. 中国电机工程学报, 2021, 41(07): 2373-2383.

[35]黄荣舟, 汤宝平, 杨燕妮, 等. 基于长短时记忆网络融合SCADA数据的风电齿轮箱状态监测[J]. 太阳能学报, 2021, 42(01): 235-239.

[36]De Bruin T, Verbert K, Babuska R. Railway Track Circuit Fault Diagnosis Using Recurrent Neural Networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 13(05): 1-11.

[37]李佳慧. 基于循环神经网络的滚动轴承寿命预测[D]. 云南：昆明理工大学, 2023.

[38]曹渝昆, 巢俊乙, 王晓飞. 基于LSTM神经网络的风机齿轮带断裂故障预测[J]. 电气自动化, 2019, 41(04): 92-95.

[39]何群, 尹飞飞, 武鑫, 等. 基于长短期记忆网络的风电机组齿轮箱故障预测[J]. 计量学报, 2020, 41(10): 1284-1290.

[40]赵洪山, 刘辉海. 基于性能改善深度置信网络的风电机组主轴承状态分析[J]. 电力自动化设备, 2018, 38(02): 44-49.

[41]Shao H , Jiang H, Zhang H, et al. Electric locomotive beraing fault diagnosis using a novel convolutional deep belief network[J]. IEEE Transactions on Industrial Electronics, 2017, 65(03): 2727-2736.

[42]Teng W , Cheng H, Ding X, et al. DNN-based approach for fault detection in a direct drive wind turbine[J]. IET Renewable Power Generation, 2018, 12(10): 1164-1171.

[43]Shao H , Jiang H, Zhang H, et al. A novel deep autoencoder fauture learning method for rotating machinery fault diagnosis[J]. Mechanical Systems and Signal Processing, 2017, 95(12): 187-204.

[44]Wang L , Zhang Z, Xu J, et al. Wind turbine blade breakage monitoring with deep autoencoders[J]. IEEE Transactions on Smart Grid, 2016, 9(04): 2824-2833.

[45]Wang L , Zhang Z, Xu J, et al. Sparse dictionary learning based adversarial variational auto-encoders for fault identification of wind turbines[J]. Measurement, 2021, 183(04): 109810.

[46]Zhang G, Wu Y, Wong K P, et al. An advanced approach for construction of optimal wind power forecasting intervals[J]. IEEE Transactions on Power Systems 2015; 30(05): 2706-2715.

[47]Yan J, Li N, Liu Y Q, et al. Short-term uncertainty forecasting method for wind power based on real-time switching cloud mode[J]. Automation of Electric Power Systems 2019; 43(03): 357-375.

[48]Li D, Liu C, Gan W. A new cognitive model: cloud model [J]. International Journal of Intelligent Systems, 2009; 24(03): 357-375.

[49]程凤伟, 常浩. 面向非平衡数据的大间隔近邻Relief算法[J]. 山西大学学报(自然科学版), 2022, 45(04): 1014-1022.

[50]李丽君, 张海清, 李代伟, 等. 基于冗余性分析的改进ReliefF特征选择算法[J]. 软件工程, 2023,26(11): 48-51.

[51]H S B , M A A , Ella A H , et al. Retraction Note: Hybrid intelligent model for classifying chest X-ray images of COVID-19 patients using genetic algorithm and neutrosophic logic[J]. Soft computing, 2023, 27(13): 9221-9221.

[52]魏泰, 贺少雄, 胡子武, 等.基于改进孤立森林算法的风电机组异常数据清洗[J]. 科学技术与工程, 2024,24(09): 3691-3699.

[53]Zang H, Xu R, Cheng L, et al. Residential load forecasting based on LSTM fusing self-attention mechanism with pooling[J]. Energy, 2021, 229(15): 120682.

[54]Tang X, Chen H, Xiang W, et al. Short-term load forecasting using channel and temporal attention based temporal convolutional network[J]. Electric Power Systems Research, 2022, 205(13): 107761.

[55]刘园园.基于混沌映射的萤火虫算法研究[D].西安:西安工程大学,2018.

[56]王改革,郭立红,段红,等.基于萤火虫算法优化BP神经网络的目标威胁估计[J].吉林大学学报(工学版), 2013,43(04) :1064-1069.

[57]郭晨霞,刘佑祺,杨瑞峰. 基于萤火虫算法优化 BP 神经网络的光强度补偿方法[J].电子测量技术,2021,44 (13) :6-10.

[58]王陶. 神经网络和萤火虫算法在股票价格预测中的应用[D] . 湘潭:湘潭大学,2021.

[59]齐长兴,毕义明,李勇,等. 改进的萤火虫算法优化 BP 神经网络及应用[J] . 现代防御技术,2018,46( 5) :32- 38.

[60]李敬明,倪志伟,朱旭辉,等. 基于佳点萤火虫算法与BP神经网络并行集成学习的旱情预测模型[J] . 系统工程理论与实践,2018,38(05) :1343-1353.

[61]ALSATTAR H A, ZAIDAN A A, ZAIDAN B B. Novel meta-heuristic bald eagle optimization algorithm[J]. Artificial Intelligence Review, 2020, 53(03): 2237-2264.

[62]RASOL J , XU Y , ZHOU Q , et al. N-fold Bernoulli probability based adaptive fast-tracking algorithm and its application to autonomous aerial refuelling[J]. Chinese Journal of Aeronautics, 2023, 36(01): 356-368.

[63]Li S, Chen H , Wang M , et al. Slinme mould algorithm:a new method for stochastic optimization[J]. Future Generation Computer Systerms, 2020, 111(12): 300-323.

[64]TANYILDIZI E, DEMIR G. Golden sine alogorithm: a novel math-inspired alogorithm[J]. Advance in Electrical and Computer Engineering, 2017, 17(20): 71-78.

[65]李宪羽. 基于麻雀搜索算法优化支持向量机的风电机组故障预测[D]. 辽宁：大连交通大学,2022.