基于极端学习机的短期风电功率预测研究

发布时间：2018-05-25 22:43

本文选题：风力发电 + 功率预测　；参考：《华北电力大学》2017年硕士论文

【摘要】：风电出力具有波动性和间歇性,大规模风电接入系统会对电网的电能质量带来不良影响,对电网的安全、稳定运行带来严峻挑战,精确的风电场输出功率预测是应对大规模风电并网问题,提高风电比例的有效手段之一。当前,国内对风电功率预测还处于理论研究阶段,开发并应用于实际的成熟可靠的预测系统较少,实践经验缺乏。可见,对风电输出功率预测方法进行研究具有重要的理论意义和现实意义,本文选择风电场输出功率短期预测方法进行研究。本文针对风电功率具有复杂的非线性、非平稳的特性,提出了一种基于经验模态分解和极端学习机相结合的风电功率预测方法。该方法首先运用经验模态分解法对原始风电功率时间序列进行分解处理,然后根据各个分量的特点分别建立合适的极端学习机预测模型,最后将各分量的预测值叠加得到最终预测值。结果表明,EMD处理降低了建模和预测的难度,提高了风电功率预测精度,并且极端学习机在学习速度和泛化性能上比传统BP神经网络具有更大的优势。为了改善极端学习机算法随机选取隐含层参数造成的模型不稳定问题,本文提出了一种基于经验模态分解和核极端学习机的预测方法。仿真结果表明,核极端学习机算法引入核函数映射代替极端学习机算法的隐含层映射,预测模型在稳定性和预测精度上都有了较大改善。为了进一步提高模型的预测精度,结合多核学习算法,本文提出了一种基于经验模态分解和多核极端学习机的功率预测方法。多核函数集合了多种基础核函数的特点,能更好的提取数据样本间的特征信息,具有更强的学习能力。仿真结果表明,该方法的风电功率预测效果得到有效提高,模型的预测精度更高,泛化性能更强,验证了该方法在风电功率预测中的有效性。
[Abstract]:Wind power output is volatile and intermittent. Large-scale wind power access system will bring adverse impact on power quality of power grid, and bring severe challenges to the security and stability of power grid. Accurate prediction of wind farm output power is one of the effective methods to solve the problem of large-scale wind power grid connection and improve wind power ratio. At present, wind power prediction in China is still in the stage of theoretical research, the development and application of practical mature and reliable forecasting system is less, and practical experience is lacking. Therefore, it is of great theoretical and practical significance to study the prediction method of wind power output power. In this paper, the short-term prediction method of wind farm output power is chosen to study. Aiming at the complex nonlinear and non-stationary characteristics of wind power, a wind power prediction method based on empirical mode decomposition and extreme learning machine is proposed in this paper. In this method, the original wind power time series is decomposed by empirical mode decomposition method, and then an appropriate extreme learning machine prediction model is established according to the characteristics of each component. Finally, the final prediction value is obtained by superposing the predicted values of each component. The results show that the EMD processing reduces the difficulty of modeling and prediction, improves the precision of wind power prediction, and the extreme learning machine has more advantages than the traditional BP neural network in learning speed and generalization performance. In order to improve the model instability caused by random selection of hidden layer parameters in extreme learning machine, a prediction method based on empirical mode decomposition and kernel extreme learning machine is proposed in this paper. The simulation results show that the kernel function mapping is introduced into the kernel extreme learning machine algorithm instead of the hidden layer mapping of the extreme learning machine algorithm, and the stability and prediction accuracy of the prediction model are greatly improved. In order to further improve the prediction accuracy of the model, a power prediction method based on empirical mode decomposition and multi-core extreme learning machine is proposed. The multi-kernel functions have the characteristics of many basic kernel functions, which can extract the feature information between the data samples better, and have a stronger learning ability. The simulation results show that the prediction effect of the proposed method is improved effectively, the prediction accuracy of the model is higher and the generalization performance is better. The effectiveness of the proposed method in wind power prediction is verified.
【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;TP181

【参考文献】