基于动态集成LSSVR的超短期风电功率预测

刘荣胜+彭敏放+张海燕+勋+沈美娥

摘 要：針对最小二乘支持向量回归（Least Square Support Vector Regression，LSSVR）建模风电功率时变特性的局限性，提出了一种基于动态集成LSSVR的超短期风电功率预测模型.首先利用风电场监测控制与数据采集（Supervisory Control And Data Acquisition，SCADA）与数值天气预报（Numerical Weather Prediction， NWP）系统的历史数据建立离线单体LSSVR模型库，然后根据预测时段与训练时段NWP序列的相似度从单体LSSVR模型库中动态选择候选集成成员，再后综合考虑正确性与多样性确定集成成员.最后由预测时段与训练时段NWP序列间的相似度分配集成LSSVR成员的权重.通过对湖南省某风电场输出功率进行预测，验证了动态集成LSSVR预测模型的有效性，与持续法、自回归求和移动平均法、单体LSSVR模型、常权重LSSVR组合模型及BPNN动态集成模型相比，动态集成LSSVR模型具有更高的精度，在天气非平稳变化阶段更加明显.

关键词：超短期风电功率预测； 最小二乘支持向量回归；动态集成；动态时间弯曲距离；数值天气预报

中图分类号：TM614 文献标志码：A

Ultra-short-term Wind Power Prediction Based on Dynamical Ensemble Least

Square Support Vector Regression LIU Rongsheng1，PENG Minfang1，ZHANG Haiyan1，WAN Xun2，SHEN Meie3

（1. College of Electrical and Information Engineering， Hunan University， Changsha 410082， China；

2. State Grid Hunan Electric Power Company Electric Power Research Institute， Changsha 410007， China；

3. Computer School Beijing Information Science & Technology University， Beijing 100101， China）

Abstract：For the limitation of least square support vector regression （LSSVR） in modeling the time varying feature of wind power， an ultra-short-term wind power prediction （USTWPP） model based on dynamical ensemble LSSVR was proposed. Firstly， the off-line LSSVR model library was created by making use of the historical data which were obtained from Numerical Weather Prediction （NWP） and supervisory control and data acquisition （SCADA） system of wind farm. Then， the candidate members of ensemble LSSVR were selected from off-line LSSVR model library dynamically according to the similarity between the NWP of forecasting period and the NWP of training period. The ensemble members were decided by considering the accuracy and diversity. Finally， the weights of ensemble LSSVR members were assigned according to the similarity between the NWP of training and NWP of prediction period. The validity of the dynamical ensemble LSSVR based predictor was verified by predicting the wind power of a wind farm in Hunan Province. Compared with persistence method （PM）， auto regressive integrated moving average （AGIMA）， LSSVR， constant weight ensemble LSSVR， and ensemble artificial neural networks （ANN）， the dynamical ensemble LSSVR is more accurate， especially when the weather changes severely.

Key words：ultra-short-term wind power prediction； least square support vector regression； dynamical Ensemble； dynamical time warp； numerical weather prediction

风力发电是目前技术最成熟、最具商业开发前景的可再生能源发电技术之一[1]，受风向、风速、气压、温度、海拔、地形地貌及风机特性等因素影响，其输出功率具有随机性和间歇性[1-2].风电大规模并入电力系统后，其随机性和间歇性会给电力系统的调度及稳定运行带来新的挑战[3-4].及时、准确的风电功率预测是大力发展风力发电的关键技术之一.国家能源局颁布的《风电场功率预测预报管理暂行办法》要求风电场必须及时……