基于主成分与果蝇神经网络的酒泉基地短期风电功率预测

发布时间：2018-05-21 19:43

本文选题：Elman神经网络 + 主成分分析法　；参考：《兰州交通大学》2017年硕士论文

【摘要】：风力发电具有随机性、间歇性等特点,导致风电场输出电能产生较大的波动,直接接入电网,会严重威胁电网的稳定性、连续性和可调性。尤其是酒泉等超大规模风电基地采用集中式并网,进一步放大了风电波动性对电网造成的剧烈冲击,产生巨大的安全隐患。因此,针对风电出力的波动性,提出研究精确的风电功率预测方法,对实现风场发电量的高精度预测和安全、经济调度具有重要的实际价值。本文根据酒泉风电基地的对风电预报的现实需求,以酒泉风电基地内某风电场及周边测风塔的实测数据为基础,采用新型自适应果蝇算法优化改进的动态Elman神经网络对风电场进行未来24h内短期风电功率预测研究。具体工作如下:首先根据对风功率公式的分析确定本文采用的原始输入变量,针对原始数据的采集、传输以及存储的过程中导致的数据缺失、错误等问题,对数据进行完整性、合理性的检测,并对检测出的异常数据进行分类,并进行均值填补和非线性回归填补。根据对当前甘肃地区各风电场误差数据分布和误差产生原因的分析,选择合适的误差评价指标对本文的预测效果进行合理的评价。其次,针对Elman神经网络本身的梯度下降学习算法收敛速度慢、易陷入局部最优的劣势,提出通过具有良好全局寻优性能和计算性能的自适应果蝇优化算法(FOA)优化Elman神经网络模型权值、阈值,建立改进的FOA-Elman神经网络模型。最后,从提高功率预测精度的角度出发,考虑到影响预测精度的因素中除了模型选择、学习算法之外,输入数据的有效性也是至关重要的,所以采用主成分分析法(PCA)对短期风电功率预测模型的输入特征分析处理,经过对输入数据的主成分分析,得到四种互不相关的主成分,使用处理完的四种主成分作为输入变量代入改进的FOA-Elman神经网络模型,建立PCA-FOA-Elman神经网络模型通过测试数据进行预测分析。通过实验仿真,将PCA-FOA-Elman模型与基于自适应果蝇算法的FOA-Elman神经网络模型以及改进Elman神经网络模型的预测效果图和预测误差分别进行比较分析。结果显示:在短期风电功率预测中,建立的PCA-FOA-Elman模型的预测效果相对于Elman神经网络模型平均绝对误差降低了39.44%,均方根误差降低了36.45%,为提高短期风电功率预测精度提供了一种新思路,对实现风电的可测、可控、可调具有重要意义。
[Abstract]:Wind power generation has the characteristics of randomness and intermittency, which leads to large fluctuation of wind farm output energy and direct access to power grid, which will seriously threaten the stability, continuity and tunability of power grid. In particular, Jiuquan and other super-large scale wind power bases use centralized grid-connected, which further amplifies the severe impact of wind power fluctuation on the power grid, resulting in huge safety risks. Therefore, in view of the fluctuation of wind power output, an accurate wind power prediction method is proposed, which is of great practical value to achieve high precision prediction and safety of wind power generation and economic dispatch. According to the actual demand of wind power forecast in Jiuquan wind power base, this paper bases on the measured data of a wind farm and its surrounding wind tower in Jiuquan wind power base. A new adaptive Drosophila algorithm is used to optimize and improve the dynamic Elman neural network to predict the short-term wind power of wind farm in the next 24 hours. The specific work is as follows: firstly, according to the analysis of wind power formula, the original input variables are determined, and the data integrity is carried out in the process of collecting, transmitting and storing the original data. The rationality of the detection, and the detection of abnormal data classification, and the means of filling and nonlinear regression filling. Based on the analysis of the distribution of the error data and the causes of the errors in the current wind farms in Gansu Province, a reasonable evaluation of the prediction effect of this paper is carried out by selecting the appropriate error evaluation indexes. Secondly, for the gradient descent learning algorithm of Elman neural network itself, the convergence speed is slow, and it is easy to fall into the disadvantage of local optimum. An improved Elman neural network model based on adaptive Drosophila optimization algorithm (FOAA) with good global optimization performance and computational performance is proposed to optimize the weights and thresholds of the Elman neural network model. Finally, from the point of view of improving the accuracy of power prediction, considering the factors that affect the prediction accuracy, besides model selection and learning algorithm, the validity of input data is also very important. So we use the principal component analysis (PCA) to analyze the input characteristics of the short-term wind power prediction model, and through the principal component analysis of the input data, we get four independent principal components. The four principal components are used as input variables to replace the improved FOA-Elman neural network model, and the PCA-FOA-Elman neural network model is established to predict and analyze by testing data. Through experimental simulation, the prediction effect diagram and prediction error of PCA-FOA-Elman model, FOA-Elman neural network model based on adaptive Drosophila algorithm and improved Elman neural network model are compared and analyzed respectively. The results show that in the short-term wind power prediction, Compared with the average absolute error of Elman neural network model, the prediction effect of the established PCA-FOA-Elman model is reduced by 39.44 and the root mean square error is reduced by 36.45, which provides a new way to improve the prediction accuracy of short-term wind power, and can be used to realize the measurable and controllable wind power. Adjustable is of great significance.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM614

【参考文献】