基于长短期记忆网络的风电场发电功率超短期预测

Short-Term Wind Power Forecasting Based on LSTM

精确的风电场发电功率超短期预测对电力系统的经济调度和安全稳定运行意义重大。为充分利用多数据源中的有效信息来进一步提高风电场超短期发电功率的预测精度,提出了一种基于长短期记忆(long short-term memory,LSTM)网络的多变量风电场超短期发电功率预测方法。该方法首先利用距离分析法筛选出与风电功率相关程度高的变量,进而降低数据的规模和复杂程度。然后利用LSTM网络对多变量时间序列进行动态时间建模,最终实现对风电功率的预测。采用美国加州某风电场的实测数据进行验证,结果表明,文中方法能够有效利用多变量时间序列进行风电场发电功率的超短期预测,较人工神经网络和支持向量机拥有更高的预测精度。

Accurate ultra-short-term wind power forecasting is essential to economic dispatching and secure operation of power system with large-scale wind power integrated.To further enhance accuracy of ultra-short-term wind power forecasting,a multivariate method for ultra-short-term wind power forecasting based on long short-term memory（LSTM） was presented in this paper.It sifted multivariate time series highly relevant to wind power with distance analysis,and modeled the time series from the viewpoint of time with LSTM.The real-world data collected from a wind farm in California was applied to verify the conclusions.Results show that the proposed method can forecast wind power using multivariate time series,and outperform artificial neural network and support vector machine with respect to forecasting accuracy.