基于SCADA数据的风电机组变桨系统运行状态评估与预测研究

发布时间：2018-04-23 16:29

本文选题：变桨系统 + SCADA数据　；参考：《北京交通大学》2017年硕士论文

【摘要】：风能作为一种洁净、无污染的绿色能源,已成为可再生能源利用的主力军。然而,高达25%～30%的运行和维护成本严重制约着风电行业的发展。同时,随着我国近10年风电技术的迅速发展,大型风场的SCADA(Supervisory Control And Data Acquisition,数据采集与监视控制)系统积累了大量历史数据,且运行中的风电机组也每时每刻都在产生实时数据。这些数据蕴含着与风机运行和设备状态有关的丰富信息。如何利用这些闲置的SCADA数据进行风机运行状态的评估与预测,无疑具有重要的现实意义和学术价值。因此,本文针对这一问题展开如下研究:1.基于SCADA数据的变桨系统状态特征参数提取方法研究。在深入研究变桨系统结构和故障机理的基础上,结合某风场级综合监测系统,采用ReliefF算法和互信息技术挖掘其SCADA数据中与变桨系统运行状态相关的特征参数。定性和定量仿真测试表明,ReliefF算法与采用互信息技术及不进行数据预处理方法(即将所有故障参数都作为模型输入)相比,训练速度更快,故障分类准确率更高。2.基于ReliefF算法提取的变桨系统预处理数据,探讨支持向量回归预测(SVR)模型参数寻优及应用于变桨系统的状态评估及预测问题。首先采用交叉验证法和网格搜索法(CV)进行SVR模型参数的寻优。然后,将功率输出作为SVR模型的决策参数,建立SVR回归预测模型。针对随机干扰因素,采用滑动窗口残差估计方法分析预测模型的残差趋势,进而评估变桨系统运行状态。与神经网络的对比实验表明,本文所建SVR模型具有较好的状态评估与预测能力。3.基于小世界邻域粒子群优化支持向量机(SWPSO-SVR)的变桨系统状态评估与预测模型研究。由于CV参数寻优方法存在计算量大,效率和精度低的缺点,在数据量大的情况下,所建SVR模型不利于实际应用。因此本文进一步探讨SVR参数寻优问题,提出SWPSO-SVR改进算法。对比CV-SVR算法可知,SWPSO-SVR算法具有精度高、收敛速度快、不易陷入局部极小值等优点。结合某风场SCADA数据测试分析,验证了改进模型用于变桨系统状态评估与预测的有效性和实用性。4.变桨系统运行状态监测方法工程应用系统研究。应用Matlab软件中的GUI开发可与综合监控系统连接的交互式平台,以拓展综合监控系统变桨系统故障预测与报警的功能,为变桨系统的安全运行维护提供决策支持。综上,风场SCADA数据蕴含着丰富的运行状态信息,通过分析其数据变化规律,可评估和预测变桨系统的健康运行状态。本文研究具有一定的经济性和实用性,可实现变桨系统在线早期故障预警的目的。
[Abstract]:As a clean and clean green energy, wind energy has become the main force of renewable energy utilization. However, up to 25% of the operating and maintenance costs seriously restrict the development of wind power industry. At the same time, with the rapid development of wind power technology in China in the past 10 years, the SCADA(Supervisory Control And Data requirement (data acquisition and Monitoring Control) system of large wind field has accumulated a large amount of historical data, and the running wind turbine is producing real-time data every moment. These data contain rich information about fan operation and equipment status. How to use these idle SCADA data to evaluate and predict the running state of fan is undoubtedly of great practical significance and academic value. Therefore, this paper focuses on this problem as follows: 1. Research on extraction method of State characteristic parameters of variable Propeller system based on SCADA data. On the basis of deep research on the structure and fault mechanism of the variable propeller system, the ReliefF algorithm and mutual information technology are used to mine the characteristic parameters related to the running state of the variable propeller system in its SCADA data, combined with the integrated monitoring system of a certain wind field. The qualitative and quantitative simulation results show that the algorithm has higher training speed and higher fault classification accuracy than using mutual information technology and data preprocessing method (that is, using all fault parameters as model input). Based on the preprocessing data of variable propeller system extracted by ReliefF algorithm, the parameter optimization of support vector regression prediction (SVR) model and its application to the condition evaluation and prediction of variable propeller system are discussed. First, cross validation method and grid search method are used to optimize the parameters of SVR model. Then, the power output is taken as the decision parameter of the SVR model, and the SVR regression prediction model is established. Aiming at the random disturbance factors, the residual trend of the prediction model is analyzed by using the sliding window residual estimation method, and then the running state of the propeller system is evaluated. The comparison experiment with neural network shows that the proposed SVR model has a good capability of state evaluation and prediction. Research on the State Evaluation and Prediction Model of variable Propeller system based on Small-world Neighbourhood Particle Swarm Optimization support Vector Machine (SWPSO-SVR). Because the CV parameter optimization method has the disadvantages of large calculation, low efficiency and precision, the established SVR model is unfavorable for practical application in the case of large amount of data. Therefore, this paper further discusses the optimization problem of SVR parameters, and proposes an improved SWPSO-SVR algorithm. Compared with the CV-SVR algorithm, the SWPSO-SVR algorithm has the advantages of high accuracy, fast convergence speed and difficulty in falling into local minima. The effectiveness and practicability of the improved model for the state evaluation and prediction of a variable propeller system are verified by combining the SCADA data of a certain wind field. Research on Engineering Application system of variable Propeller system running condition Monitoring method. The interactive platform which can be connected with the integrated monitoring system is developed by using the GUI of Matlab software to expand the function of fault prediction and alarm of the integrated monitoring system and to provide decision support for the safe operation and maintenance of the variable propeller system. In a word, the wind field SCADA data contain abundant information of running state. By analyzing the variation law of the data, the healthy running state of the variable propeller system can be evaluated and predicted. The research in this paper has certain economy and practicability, and can realize the purpose of on-line early fault warning of the variable propeller system.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM315

【参考文献】