基于改进人工蜂群算法的分散式风功率预测方法研究

发布时间：2018-10-22 18:38

【摘要】：随着风力发电的快速发展,风电场并网容量不断递增。近几年来,大型风电场建设已基本完成,为增加新能源发电装机容量,我国开始挖掘分散式风电场的资源,大力发展分散式风电场。但由于风速的波动性和不确定性,风力发电产生的风功率大小也具有不确定性,对电网的扰动也越来越大。当前,许多分散式风电场并没有风功率短期预测设备,给当地调度和自身发展带来诸多难题。因此分散式风电场的风功率短期预测势在必行,准确的风功率预测为风电场竞价上网提供了可靠保证,对调度自动化实现和现场指导具有深远意义。当前,风功率的预测方法主要有时间序列法、BP神经网络法和支持向量机。但是,时间序列法存在定阶困难的问题,BP神经网络存在权值选择问题,支持向量机存在参数设置的问题。本文针对不同风电场的现实状况和预测需求,利用改进人工蜂群结合三种不同方法对上述风电场经网络法对0-4小时以内的风功率进行预测,结果发现,该方法适用于历史数据比较充裕,对预测精度要求比较高的场合。最后,针对BP神经网络需要数据量大、容易陷入局部最优和学习速度慢等问题,将改进人工蜂群的支持向量机法应用到风功率预测中来；该算法需求的数据样本少,预测准确度较高,避免了其它一些学习算法的复杂学习过程。事实证明,该算法适合于数据样本少、预测准确度要求比较高的场合。为了满足国家电网对分散式风电并网和风电场自身发展要求,本文还开发了一种分散式风电风功率预测软件。该软件基于VB.net语言,在处理复杂计算过程时,主要依赖于MATLAB的动态链接库文件；该软件可以基于实时数值天气预报和历史风资源数据,实现未来一段时间内的风功率预测功能,另外该软件还可以根据调度提供的数据进行实时监控预警功能。最后,利用该软件对本提到的各种算法进行验证比较,结果分析了上述各个算法的适用性和不可替代性。
[Abstract]:With the rapid development of wind power generation, the grid capacity of wind farm is increasing. In recent years, the construction of large-scale wind farms has been basically completed. In order to increase the installed capacity of new energy generation, China began to tap the resources of decentralized wind farms and vigorously develop decentralized wind farms. However, due to the volatility and uncertainty of wind speed, the wind power generated by wind power generation is also uncertain, and the disturbance to the power grid is becoming more and more serious. At present, many decentralized wind farms do not have short-term wind power forecasting equipment, which brings many difficulties to local scheduling and their own development. Therefore, short-term wind power prediction is imperative for decentralized wind farms. Accurate wind power prediction provides a reliable guarantee for wind farm bidding and has far-reaching significance for the realization of dispatching automation and field guidance. At present, wind power prediction methods mainly include time series method, BP neural network method and support vector machine. However, the time series method is difficult to determine the order, the BP neural network has the weight selection problem, and the support vector machine has the problem of parameter setting. In this paper, according to the actual situation and forecast demand of different wind farms, the improved artificial bee colony combined with three different methods is used to predict the wind power within 0-4 hours by the network method. The results show that, This method is suitable for cases with abundant historical data and high prediction accuracy. Finally, aiming at the problem that BP neural network needs large amount of data, easy to fall into local optimum and slow learning speed, the improved support vector machine method of artificial bee colony is applied to wind power prediction. The prediction accuracy is high and the complicated learning process of other learning algorithms is avoided. It has been proved that the algorithm is suitable for the situation where the data sample is small and the prediction accuracy is high. In order to meet the requirements of the State Grid for the development of decentralized wind power grid and wind farm, a decentralized wind power prediction software is also developed in this paper. The software is based on VB.net language, and it mainly depends on the dynamic link library file of MATLAB when dealing with complex computing process, and it can realize wind power prediction function in the future based on real-time numerical weather forecast and historical wind resource data. In addition, the software can also provide real-time monitoring and early warning function according to the data provided by scheduling. Finally, the software is used to verify and compare the algorithms mentioned in this paper. The results show that these algorithms are applicable and irreplaceable.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM614;TP18

【相似文献】