【作者】 刘国栋；

【导师】 隋涛；

【作者基本信息】 山东科技大学 ， 电力系统及其自动化， 2023， 硕士

【摘要】 随着清洁能源的使用普及以及双碳目标的提出,风力发电因其可再生、无污染、装机灵活、建设周期短等优势逐渐得到广泛的研究和应用。我国风力资源开发前景广阔,年新增和累计并网风电装机容量也在不断增加,但是由于风的随机性与间歇性使得风能具有不确定性,这种不确定性会引起风电输出功率波动,给电力系统的调控带来巨大的挑战。因此,准确预测风电场输出功率对于电力系统调度及经济运行具有十分重要的参考意义。为此,本文以提高功率预测准确度、降低整体预测误差为目标,提出一种基于混合策略优化海鸥算法下的Elman风电功率短期预测模型,具体的研究内容如下:首先,对风电输出功率预测的方法及现状进行了简要的概述,从风机发电原理出发,分析了影响风机风电功率输出的因素,以甘肃天水某风电场风机运行数据集为样本,对输出功率影响因素进行相关性分析,选定预测模型的输入特征;在此基础上,考虑到风电功率异常数据对预测精度的影响,分析了异常数据产生的原因和分布特点,确定了针对不同分布特征异常数据的识别剔除方法。其次,选用三种基础网络模型分别进行建模与仿真分析,通过对比预测效果,分析各模型的评价指标,选用Elman网络作为基础预测模型建立Elman短期功率预测模型,并分析了不同激活函数组合、隐含层个数对Elman模型预测效果的影响;针对Elman预测模型易陷入局部最优的问题,利用海鸥算法对Elman网络权值寻优,得到了海鸥算法优化Elman网络的短期功率预测模型,经过仿真分析验证了模型的预测准确度,并通过细化预测步长和增加预测时点,验证了网络模型的泛化能力。然后,为了进一步缩小模型预测整体误差,在海鸥算法优化Elman网络的短期功率预测模型基础上,针对海鸥算法在模型寻优过程中种群更新容错率过低的问题,通过对比几种不同的混沌映射,选用初始分布特性较好的Circle混沌映射进行改进,利用改进后的Circle混沌映射优化算法初始化过程,使种群在空间中的分布更加均匀;针对海鸥算法在迭代过程中局部搜索能力匮乏的问题,采用正余弦算子改进海鸥算法的迭代方式,以平衡全局搜索能力和局部搜索能力。最后,建立了混合策略优化海鸥算法下的Elman短期功率预测模型,在所用数据集上通过仿真验证并分析模型预测性能评价指标,结果表明,相较于几个基础模型,混合策略优化海鸥算法下的Elman短期预测模型在MSE、MAPE、RMSE指标上都有所下降,能有效缩小预测误差、降低模型预测离散程度,整体上具有较好的预测效果。更多还原

【Abstract】 With the popularization of the use of clean energy and the introduction of the dual carbon goal,wind power generation has gradually been widely studied and applied because of its advantages such as renewable,non-polluting,flexible installation and short construction period.China’s wind power resources are promising and the annual and cumulative installed wind power capacity is increasing,but the random and intermittent nature of wind makes wind power uncertain,and this uncertainty can cause fluctuations in wind power output,bringing huge challenges to the regulation of the power system.Therefore,accurate prediction of wind farm output power is a very significant reference for power system dispatch and economic operation.To this purpose,with the objective of improving the power prediction accuracy and reducing the overall prediction error,this thesis proposes an Elman wind power short-term prediction model based on the hybrid strategy optimisation seagull algorithm.The specific research contents are as follows:Firstly,a brief overview of the methods and current status of wind power output prediction is given.From the principle of wind turbine power generation,the factors affecting the wind power output of wind turbines are analysed,and the correlation analysis of the factors affecting the output power is carried out with a wind turbine operation data set in Tianshui,Gansu Province as a sample,and the input features of the prediction model are selected;on this basis,considering the impact of wind power anomalous data on the prediction accuracy,the causes and distribution characteristics of the anomalous data are analysed,and the identification and rejection methods for the anomalous data with different distribution characteristics are determined.Secondly,three base network models are selected for modeling and simulation analysis,and the evaluation indexes of each model are analyzed by comparing the prediction effect.The Elman network is selected as the base prediction model to establish the Elman short-term power prediction model,and the effects of different combinations of activation functions and the number of implied layers on the prediction effect of the Elman model are analyzed.To address the problem that the Elman prediction model tends to fall into local optimum,the Seagull algorithm is used to find the optimal weights of the Elman network,and the short-term power prediction model of the Seagull algorithm optimized Elman network is obtained,and the prediction accuracy of the model is verified through simulation analysis,and the generalization ability of the network model is verified by refining the prediction step and increasing the prediction time point.Then,In order to further reduce the overall error of model prediction,based on the short-term power prediction model of Elman network optimized by the seagull algorithm,the problem of low error tolerance of population update in the model seeking process of the seagull algorithm is addressed by comparing several different chaotic mappings,and the Circle chaotic mapping with better initial distribution characteristics is selected for improvement,and the initialization process of the algorithm is optimized by using the improved Circle chaotic mapping to make the population more uniformly distributed in space.To address the lack of local search capability in the iterative process of the seagull algorithm,the positive cosine operator is used to improve the iteration of the seagull algorithm in order to balance the global search capability and local search capability.Finally,the Elman short-term power prediction model under the hybrid strategy optimization seagull algorithm is established,and the model prediction performance evaluation indexes are verified and analyzed by simulation on the used data set.The results show that compared with several basic models,the Elman short-term prediction model under the hybrid strategy optimization seagull algorithm decreases in MSE,MAPE,and RMSE indexes,which can effectively reduce the prediction error and the model prediction dispersion degree,and has a better prediction effect on the whole.更多还原