摘要
根据风电功率的波动特性,提出一种基于BP神经网络模型的预测方法.利用某风电场的实测数据,首先采用不同的分布拟合风电功率波动的阶跃变化和平稳系数的概率密度函数,发现采用t location-scale(tls)分布拟合效果最佳;其次利用tls分布的位置参数、尺度参数、形状参数和变异系数,对不同时间间隔的风电功率波动的阶跃变化和平稳系数进行分析,发现最佳预测时间间隔为10 min;最后通过神经网络预测检验,得出采用10 min间隔进行预测效果最好,此时的平均绝对误差为243 668,平均绝对百分误差为65.43%,标准误差为19 694.
According to the fluctuation characteristics of wind power, this paper proposes a forecasting method based on BP neural network model. Using the measured data of a certain wind farm, firstly the different distribution fitting step change of wind power fluctuations and stable coefficient of probability density function are adopted, and it is found that the location-scale （TLS） distribution fitting effect is the best; secondly using the location of the TLS distribution parameter and scale parameter, the shape pa- rameters and variation coefficient, the step change andthe smooth coefficient of wind power fluctuations on different time intervals are analyzed, and it finds that the best prediction intervals are for 10 min; Through the neural network prediction test, finally it draws a conclusion that 10 min interval prediction effect is the best, the mean absolute error is 243 668, the mean absolute percent- age error is 65.43%, the root mean square error is 19 694.
出处
《渭南师范学院学报》
2015年第2期41-48,共8页
Journal of Weinan Normal University
基金
四川省教育厅科研项目:基于volterra的数字放大电路预失真处理的研究(14ZB0436)
四川省教育厅科研项目:结合遗传算法和机会链模型的认知无线电频谱分配技术研究(15ZB0459)
四川师范大学成都学院科研基金项目:电力负荷的神经网络预测模型(14CSCD-QN-08)
