摘要
为了有效解决风电场数据丢失时直接相加法无法进行区域风电功率预测的问题,提出了一种考虑时空分布特性的区域风电功率预测方法。为降低模型的复杂性,根据风电场及风能信息对子区域进行具体分析。在此基础上,利用相关系数法,选择风电场出力与子区域出力间相关系数绝对值大的风场为基准风电场。以所选基准风电场预测功率为输入,利用神经网络方法,直接预测各子区域功率,整个区域预测结果为各子区域预测值之和。算例结果表明:利用相关系数法选择基准风电场无需大量历史数据支撑,原理简单易于实现;模型与风电场所采用的预测系统无关,易于工程推广应用;模型无需考虑非基准风电场功率预测情况,成本更低、效益更高;采用该模型后子区域预测误差比直接相加的方法降低了5%,整个区域预测误差仅为20.8%。
A regional wind power prediction method considering temporal and spatial characteristics is proposed to effectively solve the unavailability of direct-adding-up method in the case of missing data.To reduce the model complexity,the region is divided into several subregions according to the information from this region.Furthermore,a correlation coefficient method is proposed to select reference wind farms.The correlation coefficient between wind-farm power output and sub-region power output is calculated separately and compared with each other.Farms with larger absolute correlation coefficient are chosen as the reference ones.A back propagation neural network is adopted to directly predict wind power output of each sub-region,and the predicted power of reference farms in this sub-region is considered as the input.The regional wind power prediction can be achieved after summing the prediction results of subregions.The application shows that the correlation coefficient method needs less history data and is easy to realize.The proposed model is compatible with various wind farm power prediction systems and independent on the power prediction of non-reference farms,thus the prediction is more efficient with lower cost.The root mean square error declines by 5 ％ compared with that in direct-adding-up method,and the regional wind power prediction error reaches only 20.8％.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2013年第10期68-74,共7页
Journal of Xi'an Jiaotong University
基金
国家高技术研究发展计划资助项目(2012AA050201)
关键词
子区域
基准风电场
神经网络
区域功率预测
sub-region
reference wind farm
neural network
regional power prediction
