1、华北电力大学本科毕业设计（论文） I 基于 GRNN 网络的风电功率预测研究 摘要 大规模风电并网对电力系统造成很多不利影响，风电功率预测是减轻这些影响的一个重要手段， 故对风功率预测方法的研究具有十分重要的意义。 本文采用 GRNN 神经网络法对风电功率预测进行研究。首先，对风电场 历史 风功率数据进行分析，截断处理，建立 时间 序列预测的 GRNN 网络模型；利用该模型对历史数据进行 超前一步预测 ，目的是为了找到最优的 SPREAD 值。 SPREAD 值是 GRNN 神经网络的重要参数，该参数的选择对模型的推广能力具有重要的意义。 论文从理论上研究 神经网络的泛化能力及 GRNN 网络
2、的设计要点，重点讨论 SPREAD 参数的 物理 本质，给出训练过程中选择 该 参数的 几种 方法。 其次 ， 以某风电场的风功率历史数据为样本 ， 讨论样本设计及网络训练， 运用 MATLABR2008a 平台 编程实现对 GRNN 神经网络系统的 建模设计 。 最后， 通过 对 模拟 仿真的 手段 设定获取最小泛化误差的目标函数，进而选出最优的 SPREAD 参数 ，检验预测 效果。 关键词： 风电功率预测；泛化能力； GRNN 神经网络； 扩展系数 华北电力大学本科毕业设计（论文） II Study on wind power forecasting in wind farms base
3、d on GRNN neural networks Abstract Large scale wind power grid for power system caused a lot of adverse effect, wind power forecasting is the important method to reduce the influence, so the research of wind power forecasting method has very vital significance. In this paper, we adopt the method of
4、neural networks to GRNN to research wind power forecasting. First, we study the history wind power data of one wind farm and analysis them, then truncate the wind power data and build the GRNN neural networks model with time series； based on the model for forecasting ahead one step with the historic
5、al data, the purpose is to find the optimal value of the SPREAD. The value of the SPREAD is an important parameter to the GRNN neural networks, and the choice of this parameter has very vital significance to the generalization of this model. From theory, this paper will research the generalization a
6、bility of the neural networks and the main points of designing GRNN neural networks. This paper focuses on the physical nature of SPREAD and gives several methods of the parameter selection in the training process. Second, we take the historical wind power data of one wind farm as sample, and discus
7、s the sample designing and network training. We use MATLABR2008a platform for programming to design the GRNN neural networks model of the system. Finally, we use the method of simulation to set the objective function with the minimal generalization error, and then choose the best SPREAD to inspect the prediction effect. Keywords: wind power foresting； generalization ability； generalize regression neural networks (GRNN)； expansion coefficient
