风电场短期功率组合预测方法和评价研究

发布时间：2019-03-15 15:43

【摘要】：风能即是清洁能源又是可再生能源,且取之不尽用之不竭,大力开发风力发电产业,将成为未来电力战略部署工作的重点。在实际应用中,由于风的不确定性、随机性、间歇性等特点,给风电竞价上网和运行调度带来了不便。功率预测技术的出现,使这一问题得以解决。国内外关于风电功率预测方面的研究成果较多,均表明,不同的预测方法可对原始数据特征有着不同的体现,组合在一起才能够全面、合理的利用信息来建立具有较高预测质量的模型。本文将基于内蒙古赤峰地区赛罕坝风电场的实测数据来建立短期功率组合预测模型,实现未来一天的风功率预测,具体如下:(1)由于历史数据中包含大量的统计特征。因此本文对历史风速、功率数据进行具体分析,得到风速序列的统计特性、功率与风速的关系以及影响风功率大小的其他因素,为后续建模时特征向量的选取奠定基础。(2)支持向量机作为化繁为简的小样本学习方法,在面临复杂的样本空间时具有一定优势。本文运用最小二乘支持向量机方法来进行风速预测,在参数确定方面,采用粒子群优化算法寻优,使得传统的依据经验来确定模型参数的方法得以改善,并对标准粒子群算法加以改进,以避免粒子因早熟收敛而陷入局部最优。通过对所建模型的误差评价值指标进行统计分析,评价该模型的好坏。(3)预测方法各有所长,因此本文运用不同的功率预测方法,结合风速预测的输出,对未来一天的功率进行预测。通过对模型误差评价指标的分析,选取较为互补的两种方法作为功率组合预测模型的元素,即基于同一组数据的预测误差曲线走势相反。(4)将两种单项预测方法进行组合,采用熵权法确定组合模型权值,将同样的输入数据送入组合模型进行功率预测,对运行结果进行对比分析。误差评价指标除了平均绝对误差和平均绝对百分比误差以外,又加入绝对误差指标来进一步约束。结果表明组合模型比任一单项预测模型的效果都要好;再进一步缩短数据采样时间间隔,运用组合模型重新预测,由于数据特征更加充实,模型的预测精度又得以提升。为证明模型的泛化特性,本文对多组数据进行测试、检验,均得出较好效果,表明该预测模型适合当地风电场使用。
[Abstract]:Wind energy is not only clean energy but also renewable energy, and it is inexhaustible. The development of wind power industry will be the focus of future power strategic deployment. In practical application, due to the characteristics of wind uncertainty, randomness and intermittence, wind power bidding has brought inconvenience to online access and operation scheduling. With the emergence of power prediction technology, this problem can be solved. There are many research results on wind power prediction at home and abroad, all of which show that different forecasting methods can reflect the characteristics of the original data differently, and only when combined, can they be fully integrated. Reasonable use of information to establish a model with high prediction quality. Based on the measured data of Saihanba wind farm in Chifeng area of Inner Mongolia, a short-term combined forecasting model of wind power is established in this paper. The results are as follows: (1) due to the large number of statistical characteristics contained in the historical data, the short-term wind power prediction model will be realized in the future. In this paper, the historical wind speed and power data are analyzed in detail, and the statistical characteristics of wind speed series, the relationship between power and wind speed, and other factors affecting wind power are obtained. (2) support vector machine (SVM), as a small sample learning method for simplifying complexity, has some advantages in the face of complex sample space. In this paper, the least square support vector machine method is used to predict the wind speed. In the aspect of parameter determination, the particle swarm optimization algorithm is used to optimize the model parameters, which improves the traditional method to determine the model parameters according to the experience. The standard particle swarm optimization algorithm is improved to prevent particles from falling into local optimization due to premature convergence. Through the statistical analysis of the error evaluation value index of the model, the quality of the model is evaluated. (3) the forecasting methods have their own strong points, so this paper uses different power forecasting methods, combined with the output of wind speed prediction, Forecast the power of the next day. Through the analysis of model error evaluation index, two complementary methods are selected as the elements of power combination prediction model, that is, the prediction error curve based on the same set of data has the opposite trend. (4) the two single prediction methods are combined. The weight of the combined model is determined by entropy method, and the same input data is put into the combined model for power prediction, and the running results are compared and analyzed. In addition to the average absolute error and the average absolute percentage error, the index of error evaluation is further constrained by adding the index of absolute error. The results show that the combined model has better effect than any single forecasting model, and further shortens the time interval of data sampling and uses the combined model to re-forecast, because the data features are more abundant, the prediction accuracy of the model can be improved again. In order to prove the generalization characteristics of the model, this paper tests and tests many sets of data, and gets a good result, which shows that the prediction model is suitable for local wind farms.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM614

【参考文献】